A data center is not only a large electrical load. It is a facility made of IT equipment, cooling systems, UPS systems, batteries, backup generation, power converters, protection schemes and control logic. Each part affects how the facility imports power from the grid and how it behaves during disturbances.

This post is the first step in a slower technical sequence. The objective is to define the basic concept clearly, so that planners, reviewers and study engineers can understand the modelling structure before moving into detailed grid studies.

1. Understand the Facility, Not Only the MW

From the grid perspective, a data center should first be broken into major electrical blocks:

• IT load: servers, CPUs, GPUs, storage and networking equipment

• cooling load: chillers, fans, pumps, CRAH units and liquid-cooling systems

• UPS and battery systems

• backup generation

• auxiliary systems such as lighting, controls, security and fire systems

• internal MV/LV distribution and transformers.

This figure is a good starting point because it shows that a data center is not just a server hall. It is a complete electrical and mechanical facility.

For grid studies, this distinction is very important. The utility does not supply only the IT equipment. The utility supplies the complete facility demand at the point of interconnection.

2. From IT Load to Grid Import Demand

A data center is often announced in terms of its IT capacity, but this value does not always represent the actual demand seen by the grid.

The IT load mainly includes servers, CPUs, GPUs, storage systems and networking equipment. However, the utility supply must serve the complete facility, including cooling systems, UPS losses, batteries, auxiliary loads, transformers and internal distribution losses.

This means a data center announced as 100 MW IT load may require approximately 125 MW grid import demand at full facility loading. However, this is still only a preliminary estimate. The final demand at the point of interconnection may also depend on commissioning stages, actual IT utilization, cooling technology, battery charging, redundancy philosophy, contractual import limits, power factor and reactive power requirements.

Therefore, the first modelling question should be:

Is the announced MW value referring to IT load, total facility load, contracted grid import demand, or future ultimate capacity?

3. Why Data Center Type Matters

Not all data centers behave the same way.

Traditional enterprise and colocation facilities are often designed around continuity of service. Their load profile may be relatively stable, but they still require careful assessment because UPS and backup systems affect disturbance behaviour.

Hyperscale data centers introduce a planning challenge because their final capacity may be very large and developed in phases. The requested capacity, energized capacity and actual utilized capacity may be different at each stage.

AI data centers add another layer. GPU-based workloads can have higher rack density and more variable power behaviour compared with conventional CPU-based loads. For the grid, this means the facility type must be understood before selecting the modelling approach.

The same 100 MW figure can mean very different things depending on whether the facility is enterprise, colocation, hyperscale or AI-focused.

4. UPS, Batteries and Backup Generation

UPS systems are central to data-center reliability. They are also central to grid disturbance behaviour.

A double-conversion UPS normally includes a rectifier, DC link, battery and inverter. During normal operation, power is supplied from the grid through the UPS system to the critical IT load. During a voltage disturbance, the battery/DC link can support the IT load for a short duration. If the disturbance continues, the facility may transfer to backup generation.

From the grid perspective, the important question is not only whether the data center has a UPS.

The important questions are:

1. Will the facility remain connected during a voltage dip?

2. At what voltage and duration will the UPS transfer?

3. Will the grid see sudden load reduction?

4. Will the facility transfer to backup generation?

5. How long will it remain on backup generation?

6. How will it reconnect to the grid?

7. Will batteries recharge immediately after the event?

This is where data centers differ from ordinary static loads. Their grid response depends on UPS control, rectifier current limits, battery status, bypass logic, protection settings and backup generation philosophy. A data center may be designed to protect the IT load internally. However, the utility may require the facility to behave in a way that also supports grid reliability. These two objectives must be coordinated.

5. Bridge Toward Modelling

• Once the basic facility blocks are understood, the next step is to decide how the data center should be represented in power-system studies.

• At the early planning stage, a simple aggregated MW/MVAr model may be enough for initial load-flow studies. However, for disturbance behaviour, ride-through, harmonics or fast power-electronic interactions, more detailed models may be required.

• The PNNL report on EMT modelling provides a useful next step because it shows how different data-center power architectures can be represented for grid-level studies.

This is not required in the first concept article, but it is useful as a bridge to the next stage of the series.

6.References

Pacific Northwest National Laboratory, PNNL-38817, Electromagnetic Transient Modelling of Large Data Centers for Grid-Level Studies, Alpha Release, December 2025.

In my forecasting series, I have been writing about how future demand appears in planning studies: base load, new customer load, bottom-up additions, scenario uncertainty and the practical conversion of announcements into numbers.

Data centers now deserve a parallel discussion. They are among the most important new loads entering power-system planning, but they are not ordinary industrial loads. Their grid behavior is shaped by UPS systems, rectifiers, inverters, batteries, variable-frequency drives, cooling systems, protection logic and, increasingly, highly synchronized AI workloads.

This opening post uses PNNL-38817, Electromagnetic Transient Modelling of Large Data Centers for Grid-Level Studies, as the main technical reference. The report is important because it does two things at the same time: it explains how data-center electrical architectures are evolving, and it shows why grid-level EMT models are becoming necessary for some interconnection studies.

The key message is simple: a large data center should no longer be treated only as a fixed MW and MVAr value at the point of connection. That representation is still useful for load flow, but it is not enough to answer the more difficult questions about voltage ride-through, load recovery, oscillations, harmonics, power quality and interaction with nearby generation or power-electronic devices.

2. Why Data Centers Are Moving From Facility Design Into Grid Studies

Traditionally, the electrical engineer looking at a data center focused on reliability inside the fence: utility incomers, MV switchgear, transformers, LV distribution, UPS, generators, static transfer, earthing, protection, cooling support and redundancy. The grid planner usually saw the same facility in a much simpler form: a requested demand capacity and a power factor assumption.

That separation is becoming weaker. Large data centers can now be comparable in size to major industrial plants. Some campuses are planned in hundreds of megawatts, and AI-oriented facilities can push rack densities and electrical architectures beyond the assumptions used for conventional enterprise or colocation sites.

The PNNL report starts from this same concern: the magnitude and complexity of data-center electricity usage are now large enough to affect grid operation and dynamics. For utilities, developers and consultants, this means the study question is no longer only, ‘Can the grid supply this load?’ It is also, ‘How will this load behave during and after a disturbance?’

3. First Principle: The Data Center Is a Power-Electronics-Rich Load

The grid does not directly feed the computing chips. Between the utility connection and the processor, power may pass through HV and MV transformers, switchgear, UPS rectifiers, DC links, battery converters, output inverters, PDUs, rack-level AC/DC power supplies, DC/DC converters and low-voltage distribution paths.

This matters because power electronics control how the facility exchanges current, active power and reactive power with the grid. During voltage sags, the response may be governed by converter current limits, DC-link support, battery control and protection settings, not by the physical computing load alone.

In that sense, the data center starts to look familiar to engineers who work on inverter-based resources. It is still a load, but its interface is increasingly converter-dominated. The technical language of grid-following controls, grid-forming controls, current limits, DC-link voltage, ride-through, damping and EMT validation is becoming relevant on the demand side as well.

4. Six Electrical Architecture Families

PNNL organizes data-center power systems into six aggregated architecture families. The first three are already used today; the later three are near-future or longer-term designs driven mainly by AI hyperscale requirements and the need for much higher power density.

This table is one of the most useful starting points for a grid-connection engineer. It shows that two facilities with the same requested MW capacity may have very different grid-facing behavior. A centralized UPS facility, a distributed UPS facility and a future LVDC AI facility can all appear as ‘data centers’ in a connection queue, but their response to voltage sags, harmonics, short-time load variation and ride-through requirements may be very different.

The trend is also important: future high-density AI designs are moving toward three-phase active front ends and DC distribution. That makes data-center modelling closer to power-electronic plant modelling, even though the commercial objective is computing reliability rather than electricity generation.

5. From IT Capacity to Grid Demand: A Simple Planning Calculation

Many discussions begin with IT capacity, but the grid sees total facility demand. IT load must be converted into facility demand by adding cooling, electrical losses and site support loads. A common planning conversion uses power usage effectiveness, or PUE:

PUE = total facility energy / IT equipment energy

For early planning, the same idea is often used approximately on a power basis:

P_facility = P_IT x PUE

Example: assume a proposed AI data center has 100 MW of IT demand and a design PUE of 1.25.

P_facility = 100 MW x 1.25 = 125 MW

If the point-of-connection power factor is 0.98 lagging, the apparent power is approximately:

S = P / PF = 125 / 0.98 = 127.6 MVA

This means that a project described commercially as a 100 MW IT data center may represent around 125 MW and 127.6 MVA at the grid interface before considering staged commissioning, contractual import limits, redundancy philosophy, contingency operation, harmonic limits or dynamic behavior.

PNNL’s model parameters reinforce the same engineering idea. A grid-level model may need total peak load, IT fraction, cooling fraction, site-support fraction, transformer rating, UPS capacity ratio, LV nominal voltage, individual UPS size, fan load, pump load and PFC converter size. These are not only facility-design details; they become modelling inputs.

For example, if 75 percent of the 125 MW facility demand is IT load, the IT portion is 93.75 MW. If the UPS capacity ratio is between 1.2 and 2.0 times peak IT demand, the installed UPS capacity range may be roughly 112.5 to 187.5 MVA. That range does not mean the normal grid demand is doubled. It means the equipment rating, redundancy philosophy and disturbance behavior must be understood before the grid study assumptions are finalized.

6. Why EMT Modelling Enters the Discussion

The PNNL report is focused on electromagnetic transient modelling because some data-center questions occur at a level of detail that positive-sequence studies cannot always capture. However, one of the most practical messages from the report is that EMT should be used with judgment. Not every data-center study needs EMT.

PNNL groups EMT-relevant interconnection studies into three categories: oscillatory interactions, voltage and frequency regulation, and power quality. This is a useful framework for consultants because it prevents the discussion from becoming too broad.

• Oscillatory interactions: Could rapid data-center load changes excite turbine shaft modes, local modes, interarea modes, series-compensated network resonance or control interactions with HVDC, FACTS or IBR plants?

• Voltage and frequency regulation: Can the facility ride through a voltage sag, recover load smoothly, support voltage or frequency, and be represented reliably in positive-sequence models?

• Power quality: What harmonic currents, voltage fluctuations, flicker risk or modulation effects may appear at the point of common coupling?

A key distinction is frequency range. For lower-frequency electromechanical behavior, positive-sequence transient-stability tools may be sufficient when the model is validated. EMT becomes more useful when fast converter behavior, switching effects, sub synchronous interactions, voltage-sag response, protection transitions, PFC behavior and harmonic issues are material to the study.

7. UPS Ride-Through Is a Grid-Integration Topic

For many large facilities, the double-conversion UPS is one of the most important grid-facing components. PNNL notes that modern large data centers commonly use three-phase UPS blocks, often around the 1 MVA class, and that the rectifier side is especially important from the grid perspective.

During a voltage sag, the UPS rectifier may try to maintain the DC-link voltage by drawing more current. If the rectifier reaches current limit or trips, the battery converter may support the DC link so that the downstream IT load continues to see a regulated supply. Whether the facility stays connected, transfers, disconnects or recovers smoothly depends on the actual UPS control, current limit, battery converter, bypass logic and protection settings.

This is why a simple static load model can be misleading for ride-through questions. The question is not only how much MW is connected. The question is what the facility does during the first few cycles and seconds after a disturbance.

This point also matters commercially. Historically, many data centers were designed primarily to protect IT continuity, and conservative trip settings could be acceptable from a facility perspective. At grid scale, a sudden loss or transfer of hundreds of megawatts can become a system event. Therefore, ride-through requirements, UPS settings and grid-code expectations must be coordinated earlier in the project lifecycle.

8. AI Workloads Can Change the Load Profile

For many years, data centers were treated as smooth and dependable loads because individual compute, storage and communication activities were not strongly correlated. Aggregation made the site-level demand stable.

AI training changes that assumption. Large parallel computing processes can synchronize many processors, creating site-level active-power variations over seconds or tens of seconds. PNNL describes AI training load profiles with distinct stages such as synchronization, forward propagation, loss-function computation and backward propagation. The exact waveform depends on the workload and hardware, but the principle is clear: synchronized computing can turn processor-level variation into facility-level power variation.

For planners, this matters in three ways. First, the connection demand may not be perfectly constant. Second, rapid active-power variation can interact with weak grids, nearby machines or other power-electronic devices. Third, mitigation may be possible through better workload management, UPS control, rack-level storage, LVDC storage, utility-scale BESS or E-STATCOM-type solutions.

9. Cooling and Balance of Plant Should Not Be Forgotten

Most of the energy consumed by IT equipment becomes heat, so climate control is a major part of the facility load. PNNL notes that cooling can represent roughly 10 to 40 percent of total data-center electricity consumption, depending on climate, design, cooling technology and operating philosophy.

From a grid-modelling perspective, cooling load is often represented more simply than the IT load, but it cannot be ignored. Pumps, fans and chillers are commonly driven through VFDs, so their behavior during voltage disturbances may depend on rectifiers, DC links, motor-drive control and protection. Even if an individual drive can technically ride through, the cooling process may still trip if sensors, control circuits or associated devices reset.

This is a good reminder that facility process behavior and electrical component behavior are not always the same thing. A grid study needs a representation that is simple enough to be usable but still realistic enough to capture the disturbance response that matters at the point of connection.

10. Energy Storage Is Becoming More Than Backup

Traditional data-center storage was mainly about continuity: support the critical load long enough to transfer to backup generation. That role remains important, but the grid-facing role of storage is expanding.

PNNL describes several possible storage locations: UPS batteries, rack-level battery backup units, rack-level supercapacitors, LVDC-connected batteries, utility-scale BESS and E-STATCOM-type systems. The closer the storage is to the DC portion of the data-center architecture, the easier it may be to smooth rapid computing load changes before they appear on the AC grid.

This creates an interesting design question for future projects: should mitigation be installed at the grid interface, inside the facility MV/LV architecture, at the LVDC bus, or directly at the rack? The answer depends on cost, cycling duty, response speed, reliability philosophy, grid-code requirements and the level of cooperation between the developer, OEM and utility.

11. The Modelling Journey: From MW Block to EMT Library

The PNNL Data Center Model Library is not intended to be a ready-made representation of any specific project. It is a generic PSCAD library that can accelerate the development of site-specific EMT models and help the industry converge on modelling approaches.

This distinction is very important. A generic model can support learning, screening and early technical discussion. A real interconnection study still needs site-specific data, OEM input, validated controls, realistic protection settings, measured or supplied harmonic information, disturbance records where available and careful engineering judgment.

PNNL’s alpha release includes components for UPS representation, PFC converters, grid-following and grid-forming inverter functions, DC/DC converters, simplified rectifier loads, protection blocks, low-voltage cable representation, AI training load profiles and playback-based electronic loads. For a consultant, this is valuable because it gives a practical vocabulary for asking the right questions during model-data collection.

Instead of asking only for a peak demand value, the grid-integration team can ask: What is the UPS topology? What are the ride-through settings? Is the load behind centralized UPS or distributed storage? Is cooling behind UPS? What fraction of the load is IT, cooling and support? Is there rack-level storage? Are AI training profiles expected? What harmonic data are available? Which controls are OEM-protected, generic, or validated?

12. What This Series Will Cover

This series will use the PNNL report as the main technical foundation for the first set of posts, then connect it with standards, utility requirements, data-center design practices and practical planning calculations. The goal is to make data-center grid integration understandable for both facility designers and power-system engineers.

1. Data-center types and why enterprise, colocation, hyperscale and AI facilities are different from the grid perspective.

2. IT load, rack density, PUE, cooling demand and how announced MW becomes grid import demand.

3. Electrical distribution architectures: centralized UPS, distributed UPS, LVDC, MV SST and MVDC.

4. UPS, PFC converters, VFDs, batteries and their role in voltage ride-through.

5. Load-flow, short-circuit, protection, harmonics, flicker and voltage-stability studies.

6. RMS and EMT modelling methods for data-center interconnection studies.

7. Model validation, commissioning tests and practical data requests from developers and OEMs.

13. Closing Thought

The data center and the grid can no longer be treated as two independent systems separated by a meter at the point of connection.

The facility design determines how the load behaves during grid events. The grid design determines whether the facility can be supplied reliably and within acceptable voltage, frequency and power-quality limits. Between these two sides sits the real engineering challenge: converting a large, power-electronics-rich, mission-critical load into a validated model and a reliable connection.

For power-system engineers, data centers are no longer just demand growth. They are becoming a new class of dynamic load that must be planned, designed, connected, modelled and operated with the same seriousness that the industry has learned to apply to inverter-based generation.

Next post: Types of Data Centers - Enterprise, Colocation, Hyperscale and AI, and Why the Difference Matters to the Grid.

Reference

Ross, Brett A., and Jim Follum. Electromagnetic Transient Modeling of Large Data Centers for Grid-Level Studies: Alpha Release. PNNL-38817. Pacific Northwest National Laboratory, Richland, Washington, December 2025.

In the previous article, I discussed long-term base load forecasting and why historical demand alone is no longer enough for modern demand planning.

The base load forecast represents the normal underlying demand trend of the system. It is usually built from historical demand, weather normalization, customer growth, population, economic activity, and customer-class behaviour.

However, many modern power systems are now facing demand growth that does not follow historical trends.

• data centers

• industrial plants

• hydrogen projects

• mining loads

• EV charging depots

• new airports, ports, or rail electrification

• large commercial developments

• new cities or urban zones

These loads can create step changes in the forecast.

A system may have an organic demand growth of only 1% to 2% per year, but one new data center or industrial facility can add hundreds of MW at a single location. This is why known new customer loads must be studied as a separate forecast layer.

2. Base Load vs Known New Customer Load

A base load forecast answers:

What will demand look like if the system continues to grow under normal trends?

A known new customer load forecast answers:

What additional demand should be added because of specific new customers, projects, or connection requests?

The difference is important.

For example, if a city grows gradually, that demand may be captured in the base forecast. But if a 300 MW data center applies for connection at a specific transmission node, it should not be hidden inside a general growth rate. It must be assessed separately.

A simple forecasting structure is:

Final Planning Forecast = Base Load Forecast + Known New Customer Loads + Electrification Adjustments - DER / Efficiency Adjustments ± Uncertainty

3. Why Requested Capacity Is Not Always Planning Demand

A common mistake is to assume that the requested connection capacity is equal to the planning demand.

For example, a customer requests 300 MW. Does this mean the planner should directly add 300 MW to the forecast? Not always.

The requested capacity may be the maximum contractual requirement, but the actual planning demand may depend on:

• expected utilization

• project realization probability

• coincidence with system peak

• phased commissioning

• auxiliary load

• standby or redundant capacity

• probability of delay or cancellation

• whether the request is duplicated across different locations

Therefore, planners need a method to convert a connection request into a realistic planning value.

4. Main Planning Formula

Expected Planning Load = Requested Capacity × Utilization Factor × Realization Factor × Coincidence Factor + Auxiliary Load

5. Numerical Example: 300 MW Data Center

Assume a data center submits a connection request for 300 MW. The planning team reviews the project and applies the following assumptions:

The expected planning load is:

Expected Planning Load = 300 × 80% × 70% × 95% + 20

Step by step:

300 × 80% = 240 MW

240 × 70% = 168 MW

168 × 95% = 159.6 MW

159.6 + 20 = 179.6 MW

So the central planning value is approximately 180 MW.

This means the planner may not use the full 300 MW request in the central forecast. Instead, the project may be represented as approximately 180 MW, depending on the confidence level and planning methodology.

6. Why Realization Factor Matters

The realization factor is one of the most important assumptions for large loads.

Not every announced project is built. Not every project is built on time. Not every project reaches its full requested demand.

These values are illustrative. Each utility, system operator, or planner may use its own criteria.

The key point is that a 300 MW application at early stage should not always be treated the same as a 300 MW project already under construction.

7. Ramp-Up Schedule

Large loads do not always appear at full demand immediately.

A data center, industrial plant, or hydrogen facility may be energized in phases. This means the forecast should include a ramp-up schedule.

Using the previous example:

Full Expected Planning Load = 180 MW

Assume the project ramps up over three years:

This is more realistic than adding the full load in the first year.

Forecast Load in Year n = Expected Planning Load × Ramp-Up Percentage

For Year 1: 180 × 40% = 72 MW

For Year 2: 180 × 70% = 126 MW

For Year 3: 180 × 100% = 180 MW

This approach helps planners decide when reinforcement is actually required.

8. Network Capacity Gap Calculation

After estimating the expected planning load, the next step is to compare it with the available local network capacity.

Total forecast local load:

Total Forecast Load = Existing Local Peak + Expected New Customer Load

Total Forecast Load = 420 + 180 = 600 MW

Capacity gap:

Capacity Gap = Total Forecast Load - Available Network Capacity

Capacity Gap = 600 - 550 = 50 MW

This indicates an approximate 50 MW capacity gap.

The planner may then consider:

• transformer upgrade

• new feeder or circuit

• new substation

• transmission reinforcement

• staged connection

• operational limitation

• demand flexibility

• local BESS

• commercial security before investment

9. Low, Central and High Scenarios

Because large customer loads are uncertain, planners should normally develop scenarios.

Using the same 300 MW data center example:

The same connection request can therefore produce very different planning outcomes.

This is why a single deterministic number may not be enough. Scenario planning helps the planner understand the range of possible impacts.

10. Avoiding Double Counting

One of the most important checks is double counting.

A known new load should not be added twice.

For example, suppose a region’s base load forecast already includes strong industrial growth based on economic development assumptions. If the same industrial project is then added again as a separate known new load, the final forecast may become overstated.

A practical check is:

Final Forecast = Base Load Forecast + Specific New Loads Not Already Included in Base Forecast

Before adding a known new load, planners should ask:

1. Is this load already included in the economic growth assumption?

2. Is it part of a general industrial growth forecast?

3. Is it included in another developer application?

4. Is the same project requesting connection at multiple locations?

5. Is the requested capacity firm, speculative, or phased?

6. Is the project included in both transmission and distribution forecasts?

7. Is the timing consistent with the planning horizon?

This step is especially important for data centers, industrial parks, hydrogen projects, and large electrification programs.

11. Planning Impact of Known New Loads

Known new customer loads can affect the power system in several ways.

Generation level

• capacity requirement

• energy demand

• reserve margin requirement

• BESS and flexibility needs

Transmission level

• corridor loading

• transformer loading

• voltage performance

• reactive power requirement

• transfer capability

• contingency performance

• system strength

Distribution level

• feeder loading

• substation capacity

• voltage drop

• protection coordination

• reliability margins

• hosting capacity for DERs

Commercial and regulatory level

• connection queue management

• reinforcement cost allocation

• investment timing

• security requirements

• staged connection agreements

This is why known new loads are not only a forecasting issue. They are also a planning, technical, commercial, and regulatory issue.

12. A Simple Methodology for Known New Load Forecasting

A practical workflow can be:

1. Collect project information: requested capacity, location, voltage level, customer type, connection date, and operating profile.

2. Classify project maturity: conceptual, application submitted, feasibility completed, agreement signed, under construction, or committed.

3. Estimate utilization factor: expected operating demand compared with requested capacity.

4. Apply realization factor: reflect the probability or confidence that the project will materialize.

5. Apply coincidence factor: assess whether the load occurs during system peak, local peak, or off-peak conditions.

6. Add auxiliary load: include cooling, pumps, supporting facilities, or other associated loads.

7. Build ramp-up profile: represent staged development year by year.

8. Check network capacity gap: compare total forecast load with available network capacity.

9. Develop scenarios: low, central, and high planning cases.

10. Check for double counting: ensure the same demand is not already included in the base forecast.

13. Key Takeaway

Known new customer loads convert forecasting from historical trend-based planning into project-aware planning.

A connection request is not always equal to planning demand. The planner must consider utilization, realization, coincidence, auxiliary demand, ramp-up timing, and double-counting risk.

The key message is simple:

Base load forecasting tells us how demand may grow under normal trends. Known new customer load forecasting tells us how specific projects can change the future demand trajectory.

In the next article, the discussion will cover:

DER and Electrification Adjustments: How Rooftop PV, BESS, EVs and Heat Pumps Change Net Load Forecasting.

Reference

[1] Energy Systems Integration Group, Long-Term Load and DER Forecasting, A Report by the Energy Systems Integration Group’s Long-Term Load and DER Forecasting Task Force, Aug. 2025.

In the previous articles, I discussed forecast granularity, top-down and bottom-up forecasting, and how forecasting approaches change depending on the planning use case.

Now we move to the first major technical building block of long-term load forecasting:

”Base load forecasting”

Base load forecasting estimates the underlying future demand before adding separately modelled adjustments such as data centers, industrial projects, EV charging, electrification, DERs, energy efficiency, and demand response.

In simple terms, base load forecasting answers the question:

If the system continues to grow under normal economic, demographic, weather, and customer-consumption trends, what would the future demand look like?

Mathematically, the final planning forecast can be represented as:

Final Forecast = Base Load Forecast + New Loads + Electrification − DER / Efficiency Adjustments ± Scenario Adjustments

or:

Lᶠ(t) = Lᵇ(t) + Lⁿ(t) + Lᵉ(t) − Lᴰᴱᴿ(t) − Lᵉᶠᶠ(t) ± U(t)

Where:

• Lᶠ(t) = final planning load forecast at time t

• Lᵇ(t) = base load forecast

• Lⁿ(t) = known new customer loads

• Lᵉ(t) = electrification load

• Lᴰᴱᴿ(t) = DER impact on net load

• Lᵉᶠᶠ(t) = energy-efficiency reduction

• U(t) = uncertainty or scenario adjustment

This article focuses only on Lᵇ(t), the base load forecast.

2. What Is the Base Load Forecast?

The base load forecast represents the expected demand trend before adding major external adjustments.

It captures normal growth due to:

• population growth

• customer growth

• economic activity

• historical consumption behaviour

• weather-normalized demand

• customer-class trends

• normal urban and commercial development

A simple mathematical expression for base load growth is:

Lᵇ(t) = L₀ × (1 + g)ᵗ

Where:

• Lᵇ(t) = base load forecast in year t

• L₀ = current base load

• g = annual organic growth rate

• t = number of years into the forecast horizon

For example, if the current base peak load is 5,000 MW, annual growth is 1.8%, and the forecast horizon is 10 years:

Lᵇ(10) = 5,000 × (1 + 0.018)¹⁰

Lᵇ(10) = 5,975 MW

This means the base peak demand increases from 5,000 MW to approximately 5,975 MW over 10 years before adding other forecast layers.

3. Why Base Load Forecasting Matters

The base load forecast is the foundation of the entire planning forecast.

If the base forecast is too high, the final forecast may overstate the need for:

• generation capacity

• transmission reinforcement

• distribution upgrades

• BESS and flexibility resources

• network investment

  If the base forecast is too low, the system may face:

• delayed reinforcements

• insufficient capacity

• reliability risk

• connection constraints

• underestimated investment needs

Mathematically, any error in the base load forecast propagates into the final forecast.

If:

Lᶠ(t) = Lᵇ(t) + A(t)

where A(t) represents all later adjustments, then an error in the base forecast creates:

ΔLᶠ(t) = ΔLᵇ(t) + ΔA(t)

This means a poor base forecast can distort the entire planning outlook.

4. Historical Demand as the Starting Point

The first input to the base load forecast is historical demand.

Historical demand may be represented as a time series:

Lₕ = {L₁, L₂, L₃, …, Lₙ}

Where each Lᵢ represents historical demand at a given time interval.

Depending on the planning requirement, the interval may be:

• annual

• monthly

• daily

• hourly

• sub-hourly

For long-term planning, planners often extract key indicators from the historical time series:

Annual energy:

Eᵧ = Σ Lₕ × Δt

Annual peak demand:

Pᵧ = max(Lₕ)

Load factor:

LF = Eᵧ / (Pᵧ × 8,760)

Where:

• Eᵧ = annual energy consumption

• Pᵧ = annual peak demand

• LF = load factor

• 8,760 = number of hours in a normal year

These indicators help planners understand whether demand is growing in energy, peak, or both.

5. Weather Normalization

Electricity demand is strongly affected by weather.

In hot climates, peak demand may be driven by cooling load. In cold climates, heating demand may become dominant. Therefore, historical demand must often be adjusted to remove abnormal weather effects.

A simplified weather-normalized demand equation can be written as:

Lᴺ = Lᴬ − β × (Wᴬ − Wᴺ)

Where:

• Lᴺ = weather-normalized load

• Lᴬ = actual recorded load

• β = weather sensitivity coefficient

• Wᴬ = actual weather variable

• Wᴺ = normal weather variable

For example:

• Actual recorded peak = 10,500 MW

• Normal weather peak estimate = 10,000 MW

• Weather impact = 500 MW

Then:

Weather-normalized peak = 10,500 − 500 = 10,000 MW

This normalized value is a better starting point for long-term forecasting because it removes the effect of an abnormal weather event.

6. Customer-Class Forecasting

A more rigorous base load forecast should separate demand by customer class.

Instead of forecasting the whole system with one growth rate, the forecast can be written as:

Lᵇ(t) = Σ Lᵢ,₀ × (1 + gᵢ)ᵗ

Where:

• Lᵇ(t) = total base load forecast

• Lᵢ,₀ = current load of customer class i

• gᵢ = annual growth rate of customer class i

• t = forecast year

• i = residential, commercial, industrial, agricultural, public sector, etc.

The total base forecast is:

Lᵇ(10) = 3,657 + 2,321 + 1,657 + 541

Lᵇ(10) = 8,176 MW

This is more informative than applying one average growth rate to the entire system.

7. Peak Demand and Energy Forecasting

Base load forecasting should distinguish between peak demand and annual energy.

Peak demand is used for capacity and network planning.

Annual energy is used for energy supply, fuel planning, market modelling, and revenue forecasting.

Peak demand forecast:

Pᵇ(t) = P₀ × (1 + gᴾ)ᵗ

Energy forecast:

Eᵇ(t) = E₀ × (1 + gᴱ)ᵗ

Where:

• Pᵇ(t) = base peak demand forecast

• Eᵇ(t) = base annual energy forecast

• gᴾ = peak demand growth rate

• gᴱ = energy growth rate

These two growth rates do not always have to be the same.

For example:

• Energy demand may grow due to more consumption.

• Peak demand may remain controlled due to demand response.

• Peak demand may grow faster than energy due to air-conditioning, heat pumps, or EV charging.

This is why both energy and peak forecasting are required.

8. Hourly Base Load Forecasting

Modern planning increasingly requires hourly forecasts, not only annual peak and energy.

An hourly base load forecast can be represented as:

Lᵇₕ(t) = Sₕ × Pᵇ(t)

Where:

• Lᵇₕ(t) = base load forecast for hour h in year t

• Sₕ = normalized hourly load shape factor

• Pᵇ(t) = forecast annual peak demand

The shape factor can be defined as:

Sₕ = Lₕ / P

Where:

• Lₕ = historical load in hour h

• P = historical peak demand

This allows the planner to create an 8,760-hour base load forecast.

Hourly forecasting is important because it can show:

• peak demand

• minimum load

• evening ramp

• seasonal variation

• load coincidence

• flexibility needs

• storage charging opportunities

• DER impact on net load

A forecast that only provides annual peak demand may miss important operational and planning conditions.

9. Spatial Allocation of the Base Forecast

Base load forecasting must also consider location.

A system-level forecast may be mathematically correct but still insufficient for transmission or distribution planning.

A simple spatial allocation method is:

Lᵇⱼ(t) = αⱼ × Lᵇ(t)

Where:

• Lᵇⱼ(t) = base load forecast for zone, node, substation, or feeder j

• αⱼ = allocation factor for location j

• Lᵇ(t) = total base load forecast

The allocation factors should satisfy:

Σ αⱼ = 1

For example, if the total base forecast is 8,176 MW and one load zone has an allocation factor of 20%:

Lᵇⱼ = 0.20 × 8,176 = 1,635 MW

However, using fixed historical allocation factors can be risky. Future load growth may not follow historical spatial patterns. Some areas may grow faster due to new housing, commercial activity, industrial zones, EV charging hubs, or urban expansion.

Therefore, spatial allocation should be reviewed against local information.

10. Base Forecast Validation

After developing the base forecast, planners should validate it.

Validation can include:

1. Back-casting

Apply the forecasting method to an earlier historical year and compare the result with actual observed demand.

Forecast error can be measured as:

Forecast Error = Forecast Value − Actual Value

Percentage error:

Error (%) = [(Forecast − Actual) / Actual] × 100

2. Comparison with historical growth

Check whether the forecast growth rate is reasonable compared with historical demand growth.

3. Comparison with customer and economic indicators

Check whether the demand forecast is consistent with customer growth, GDP, population, and industrial activity.

4. Comparison across customer classes

Check whether residential, commercial, industrial, and other class forecasts are individually reasonable.

5. Scenario sensitivity

Test low, central, and high growth cases.

For example:

This helps planners understand the range of possible outcomes.

11. Why Historical Trends Alone Are No Longer Enough

A purely historical model may be written as:

Lᵇ(t) = f(Lₜ₋₁, Lₜ₋₂, …, Lₜ₋ₙ)

This means future demand is estimated mainly from past demand.

This approach may work when the power system is stable and demand drivers are changing slowly.

However, modern power systems are affected by many forward-looking drivers:

• EV adoption

• heat pumps

• data centers

• industrial electrification

• rooftop PV

• behind-the-meter BESS

• demand response

• energy efficiency

• flexible tariffs

• climate change

• changing customer behaviour

• policy targets

Therefore, a stronger model is:

Lᵇ(t) = f(Historical Load, Weather, Economy, Population, Customer Growth, Technology Trends, Behaviour)

This means historical demand remains important, but it must be combined with forward-looking variables.

12. Simple Step-by-Step Base Load Forecasting Methodology

A practical mathematical workflow can be summarized as follows:

Step 1: Collect historical load data

Lₕ = {L₁, L₂, …, Lₙ}

Step 2: Normalize for weather

Lᴺ = Lᴬ − β × (Wᴬ − Wᴺ)

Step 3: Estimate growth drivers

Estimate growth rate g using historical demand, population, GDP, customer numbers, or class-specific indicators.

Step 4: Forecast by customer class

Lᵇ(t) = Σ Lᵢ,₀ × (1 + gᵢ)ᵗ

Step 5: Develop peak and energy forecast

Pᵇ(t) = P₀ × (1 + gᴾ)ᵗ

Eᵇ(t) = E₀ × (1 + gᴱ)ᵗ

Step 6: Convert to hourly profile where required

Lᵇₕ(t) = Sₕ × Pᵇ(t)

Step 7: Allocate spatially

Lᵇⱼ(t) = αⱼ × Lᵇ(t)

Step 8: Validate forecast

Error (%) = [(Forecast − Actual) / Actual] × 100

Step 9: Develop scenarios

Low Case, Central Case, High Case

Step 10: Use the result as the base forecast layer

The base forecast is then used as the starting point for known new loads, DERs, electrification, flexibility, and uncertainty adjustments.

13. Numerical Example: Class-Based Base Load Forecast

Assume a utility has the following current peak demand:

The 10-year forecast for each class is:

Residential = 3,000 × (1 + 2.0%)¹⁰ = 3,657 MW

Commercial = 2,000 × (1 + 1.5%)¹⁰ = 2,321 MW

Industrial = 1,500 × (1 + 1.0%)¹⁰ = 1,657 MW

Public / Other = 500 × (1 + 0.8%)¹⁰ = 541 MW

Therefore:

Total current demand = 7,000 MW

Total 10-year base forecast = 8,176 MW

The total base load growth is:

Growth = 8,176 − 7,000 = 1,176 MW

Percentage increase:

Growth (%) = (1,176 / 7,000) × 100 = 16.8%

This result represents only the base load forecast. It does not yet include major new loads, EV charging, data centers, DER adjustments, or high electrification scenarios.

14. How the Base Forecast Fits into the Final Forecast

Once the base forecast is prepared, additional forecast layers can be added:

Lᶠ(t) = Lᵇ(t) + Lⁿ(t) + Lᵉ(t) − Lᴰᴱᴿ(t) − Lᵉᶠᶠ(t) ± U(t)

For example:

• Base forecast = 8,176 MW

• Known new loads = 600 MW

• Electrification = 400 MW

• DER net-load reduction = 300 MW

• Energy efficiency reduction = 150 MW

• Scenario adjustment = ±500 MW

Central final forecast:

Lᶠ = 8,176 + 600 + 400 − 300 − 150

Lᶠ = 8,726 MW

High case:

Lᶠ,high = 8,726 + 500 = 9,226 MW

Low case:

Lᶠ,low = 8,726 − 500 = 8,226 MW

This shows why the base forecast is not the final forecast. It is the first layer.

15. Key Takeaway

Base load forecasting is the mathematical foundation of long-term demand forecasting.

It begins with historical load, weather normalization, customer-class analysis, growth drivers, peak and energy forecasting, hourly profiles, spatial allocation, validation, and scenario testing.

However, historical demand alone is no longer sufficient.

Modern power systems are changing due to electrification, DERs, EVs, data centers, customer behaviour, flexible demand, and climate impacts.

The key message is:

The base load forecast explains the underlying demand trend. The final planning forecast adds the future changes that historical demand alone cannot capture.

In the next article, the discussion can move into:

Known New Customer Loads: How Data Centers, Industrial Growth and Large Connections Change the Forecast.

Reference

[1] Energy Systems Integration Group, Long-Term Load and DER Forecasting, A Report by the Energy Systems Integration Group’s Long-Term Load and DER Forecasting Task Force, Aug. 2025.

In the previous article, I discussed the difference between top-down and bottom-up load forecasting. However, in real planning work, the forecasting approach is not selected in isolation. It depends on the planning application.

A forecast used for resource adequacy is different from a forecast used for a distribution feeder. A forecast used for generation expansion is different from a forecast used for a data-center connection. Similarly, an integrated system plan cannot rely only on one method. It must combine system-level consistency with local network and customer-level detail.

This article presents numerical examples for different power system planning applications. The purpose is not to provide market-specific values, but to explain how the forecasting approach changes depending on the planning question.

The examples are illustrative and are intended to support methodology understanding.

2. Indicators for the planning application

The Excel workbook includes three indicators for each planning application:

a) Top-Down Reliance

This indicates how much the planning case depends on system-level or regional assumptions.

Examples of top-down inputs include:

• system peak demand

• annual energy demand

• population and economic growth

• reserve margin

• regional demand growth

• electrification assumptions

• system-level DER and flexibility assumptions

A high top-down reliance means the study mainly begins from a system-wide or regional forecast.

b) Bottom-Up Reliance

This indicates how much the planning case depends on local, customer-specific, feeder-level, or project-specific inputs.

Examples of bottom-up inputs include:

• feeder loading

• substation loading

• customer connection requests

• data-center or industrial project demand

• EV charging hub location

• DER capacity on a feeder

• local development plans

• project realization factors

A high bottom-up reliance means the study mainly depends on local or project-level details.

c) Uncertainty Score

The uncertainty score is a relative assumption-risk indicator. It is not a formal probability value.

It reflects how uncertain the forecast assumptions are, considering:

• project realization risk

• customer behaviour

• DER adoption uncertainty

• EV charging behaviour

• technology adoption

• local network variability

• data quality

• timing uncertainty

• policy and market uncertainty

For example, data-center connection and EV corridor planning have higher uncertainty because project timing, utilization, simultaneity, and customer behaviour can vary significantly.

3. Planning Applications and Preferred Forecasting Approaches

The table below summarizes the preferred forecasting approach for each planning application.

Figure 1: Summary of planning applications, preferred forecasting approaches, key numerical logic, planning outputs, method-dependence indicators, and relative uncertainty scores.

This summary shows that system-wide studies normally rely more on top-down forecasting, while local network and connection studies rely more on bottom-up forecasting. Integrated planning requires both approaches to be reconciled.

4. Resource Adequacy: Primarily Top-Down

Planning Question

Will the total power system have enough supply, storage, imports, and demand-side resources to serve future peak demand reliably?

Numerical Example

Input and Assumptions

• Current system peak demand = 10,000 MW

• Annual demand growth = 2.3%

• Forecast horizon = 10 years

• DER and flexibility reduction = 600 MW

• Planning reserve margin = 15%

First, calculate the gross future peak:

Gross future peak = 10,000 × (1 + 2.3%)¹⁰

Gross future peak = 12,553 MW

Then subtract the expected DER and flexibility reduction:

Net forecast peak = 12,553 − 600

Net forecast peak = 11,953 MW

The indicative firm capacity requirement is:

Firm capacity requirement = 11,953 × (1 + 15%)

Firm capacity requirement = 13,746 MW

Interpretation

The adequacy study should test whether firm generation, storage contribution, imports, and demand-side resources can meet a future net peak demand of approximately 11,953 MW, with an indicative firm capacity requirement of approximately 13,746 MW.

Why Top-Down Is Preferred

Resource adequacy is mainly a system-wide supply-demand balance question. Therefore, it normally starts from a top-down system demand forecast.

5. Generation Expansion : Primarily Top-Down

Planning Question

How much new firm generation, storage, import capability, or demand-side capacity is required to meet future system demand and reliability targets?

Numerical Example

Input and Assumptions

• Forecast peak demand = 12,300 MW

• Planning reserve margin = 15%

• Existing firm capacity = 11,200 MW

• Planned retirements = 500 MW

• Accredited storage / demand response addition = 400 MW

Required firm capacity:

Required firm capacity = 12,300 × (1 + 15%)

Required firm capacity = 14,145 MW

Effective existing capacity:

Effective existing capacity = 11,200 − 500 + 400

Effective existing capacity = 11,100 MW

New firm capacity need:

New capacity need = 14,145 − 11,100

New capacity need = 3,045 MW

Interpretation

The generation expansion study should identify how to meet approximately 3,045 MW of additional firm capacity need. This may come from new generation, BESS, imports, demand response, or a combination of resources.

Why Top-Down Is Preferred

Generation expansion planning starts from system-wide demand, reserve requirements, retirement assumptions, and long-term reliability needs. This makes top-down forecasting the natural starting point.

6. Market Design : Primarily Top-Down

Planning Question

Does the future market design provide enough ramping, flexibility, and investment signals to operate the system reliably?

Numerical Example

Input and Assumptions

• Evening peak demand = 9,800 MW

• Midday minimum net load = 4,200 MW

• Existing ramp capability = 4,500 MW

• Demand response ramp contribution = 300 MW

• BESS ramp contribution = 250 MW

Required ramp:

Required ramp = 9,800 − 4,200

Required ramp = 5,600 MW

Available ramp capability:

Available ramp = 4,500 + 300 + 250

Available ramp = 5,050 MW

Ramping gap:

Ramping gap = 5,600 − 5,050

Ramping gap = 550 MW

Interpretation

The market design study may identify an indicative 550 MW ramping or flexibility gap. This could support the need for flexibility products, ancillary services, storage procurement, demand response, or revised market signals.

Why Top-Down Is Preferred

Market design is mainly concerned with system-level operation, flexibility, pricing, and investment signals. Therefore, it generally begins with a top-down view of future net-load behaviour.

7. Regional Transmission Outlook : Top-Down with Local Adjustments

Planning Question

Will a regional transmission corridor remain adequate after organic demand growth and known local developments are considered?

Numerical Example

Input and Assumptions

• Current corridor load = 2,800 MW

• Annual regional growth = 2%

• Forecast horizon = 8 years

• Data center requested load = 500 MW

• Data center realization factor = 70%

• Industrial load addition = 250 MW

• Industrial realization factor = 80%

Base regional forecast:

Base regional forecast = 2,800 × (1 + 2%)⁸

Base regional forecast = 3,281 MW

Expected local additions:

Expected local additions = 500 × 70% + 250 × 80%

Expected local additions = 350 + 200 = 550 MW

Forecast corridor load:

Forecast corridor load = 3,281 + 550

Forecast corridor load = 3,831 MW

Interpretation

The regional transmission corridor should be tested against approximately 3,831 MW of forecast load. This includes both top-down regional growth and bottom-up local project adjustments.

Why Local Adjustments Are Required

A regional forecast may allocate growth too evenly across the network. Local developments such as data centers, industrial zones, EV corridors, and urban expansion can create corridor-specific constraints.

8. Transmission-Node Forecasting : Reconciled Top-Down and Bottom-Up

Planning Question

What demand should be assigned to a specific transmission bus for power-flow, contingency, and planning studies?

Numerical Example

Input and Assumptions

• Load-zone forecast = 5,000 MW

• Historical node share = 14%

• Bottom-up local forecast = 950 MW

• Top-down reconciliation weight = 40%

• Bottom-up reconciliation weight = 60%

Top-down node forecast:

Top-down node forecast = 5,000 × 14%

Top-down node forecast = 700 MW

Reconciled node forecast:

Reconciled forecast = 700 × 40% + 950 × 60%

Reconciled forecast = 850 MW

Difference compared with top-down allocation:

Difference = 850 − 700

Difference = 150 MW

Interpretation

A purely top-down allocation would assign 700 MW to the node. However, bottom-up local evidence suggests higher demand. A reconciled forecast of 850 MW provides a more defensible planning value.

Why Reconciliation Is Required

Transmission-node forecasting sits between system planning and local planning. Historical allocation factors may miss future local load pockets, while bottom-up forecasts may overstate demand if project realization is not properly considered.

9. Distribution Planning : Primarily Bottom-Up

Planning Question

Can a distribution substation or feeder support expected local demand growth, electrification, and DER impacts over the planning horizon?

Numerical Example

Input and Assumptions

• Substation capacity = 60 MW

• Current peak load = 48 MW

• Annual organic growth = 1.5%

• Forecast horizon = 5 years

• EV load addition = 8 MW

• Rooftop PV net-load reduction = 2 MW

• Planning margin trigger = 90%

Organic forecast load:

Organic forecast load = 48 × (1 + 1.5%)⁵

Organic forecast load = 51.7 MW

Final substation forecast:

Final forecast = 51.7 + 8 − 2

Final forecast = 57.7 MW

Loading ratio:

Loading ratio = 57.7 / 60

Loading ratio = 96.2%

Planning trigger:

Planning trigger = 60 × 90%

Planning trigger = 54 MW

Interpretation

The final forecast load is approximately 57.7 MW, which is below the 60 MW capacity but above the 90% planning trigger of 54 MW. This indicates that the utility should review upgrade timing, contingency loading, load transfer capability, and non-wires alternatives.

Why Bottom-Up Is Preferred

Distribution constraints are local. They depend on substation loading, feeder loading, EV adoption, rooftop PV, customer growth, and local development activity.

10. Data-Center or Industrial Connection : Bottom-Up with System-Level Validation

Planning Question

Can a large new customer be connected at the requested location, and what reinforcement or system validation is required?

Numerical Example

Input and Assumptions

• Requested connection capacity = 300 MW

• Expected utilization factor = 80%

• Project realization factor = 70%

• Auxiliary load = 20 MW

• Available local capacity = 120 MW

• System peak demand = 10,000 MW

Expected project demand:

Expected demand = 300 × 80% × 70% + 20

Expected demand = 188 MW

Local upgrade need:

Upgrade need = 188 − 120

Upgrade need = 68 MW

System peak impact:

System impact = 188 / 10,000

System impact = 1.88%

Interpretation

The expected project demand is approximately 188 MW. If only 120 MW of local spare capacity is available, an indicative 68 MW reinforcement requirement may arise. Because the project is large, the planner should also validate wider system impacts.

Why Bottom-Up Is Preferred

Large-load connections are project-specific. The forecast must consider location, requested capacity, utilization, ramp-up schedule, auxiliary demand, realization factor, and connection timing.

11. EV Corridor Planning : Bottom-Up

Planning Question

Can a feeder or substation support a planned EV fast-charging corridor or charging hub?

Numerical Example

Input and Assumptions

• Number of chargers = 20

• Charger rating = 0.35 MW

• Simultaneity factor = 55%

• Auxiliary load = 0.5 MW

• Available feeder spare capacity = 2.5 MW

• Managed charging reduction = 20%

Unmanaged charging peak:

Unmanaged peak = 20 × 0.35 × 55% + 0.5

Unmanaged peak = 4.35 MW

Managed charging peak:

Managed peak = 4.35 × (1 − 20%)

Managed peak = 3.48 MW

Remaining feeder capacity gap:

Capacity gap = 3.48 − 2.5

Capacity gap = 0.98 MW

Interpretation

Managed charging reduces the peak from 4.35 MW to 3.48 MW, but the feeder still has an indicative remaining gap of approximately 0.98 MW. The planner may need to consider feeder reinforcement, local BESS, staged charger deployment, or stronger charging controls.

Why Bottom-Up Is Preferred

EV charging is location-specific and time-specific. A small number of high-power chargers can create a local network constraint even if the annual energy demand appears modest.

12. DER Hosting-Capacity Assessment: Bottom-Up

Planning Question

How much additional DER can be connected on a feeder before thermal, voltage, reverse-flow, or protection limits are reached?

Numerical Example

Assume:

• Feeder minimum load = 3 MW

• Reverse power export limit = 5 MW

• Existing PV capacity = 4 MW

• Thermal spare capacity = 4.5 MW

• Voltage-based limit = 6 MW

• Protection-based limit = 5.5 MW

Reverse-flow headroom:

Reverse-flow headroom = minimum load + export limit − existing PV

Reverse-flow headroom = 3 + 5 − 4

Reverse-flow headroom = 4 MW

Additional DER hosting capacity is the minimum of:

• reverse-flow headroom = 4 MW

• thermal spare capacity = 4.5 MW

• voltage-based limit = 6 MW

• protection-based limit = 5.5 MW

Therefore:

Additional DER hosting capacity = 4 MW

Existing plus additional PV:

Total PV after using headroom = 4 + 4

Total PV = 8 MW

Interpretation

The binding constraint is the reverse-flow headroom. Even though the thermal, voltage, and protection limits are higher, the feeder can host only approximately 4 MW of additional DER before reaching the reverse-flow limit.

Why Bottom-Up Is Preferred

DER hosting capacity depends on local feeder topology, voltage profile, transformer loading, protection settings, reverse power flow, and the exact location of DER connections.

13. Integrated System Planning : Hybrid and Reconciled

Planning Question

How should system-level demand, local projects, DERs, electrification, generation, transmission, and distribution plans be combined into one internally consistent outlook?

Numerical Example

Input and Assumptions

• Top-down system forecast = 15,000 MW

• Bottom-up aggregate forecast = 15,800 MW

• Top-down reconciliation weight = 60%

• Bottom-up reconciliation weight = 40%

• Uncertainty margin = 8%

Reconciled planning forecast:

Reconciled forecast = 15,000 × 60% + 15,800 × 40%

Reconciled forecast = 15,320 MW

Forecast gap:

Forecast gap = 15,800 − 15,000

Forecast gap = 800 MW

High planning case:

High case = 15,320 × (1 + 8%)

High case = 16,546 MW

Interpretation

The reconciled central planning forecast is approximately 15,320 MW. The difference between the top-down and bottom-up views is 800 MW, which should be investigated and documented. The high case of approximately 16,546 MW can be used for sensitivity analysis.

Why Hybrid and Reconciled Forecasting Is Preferred

Integrated planning must preserve system consistency while capturing local constraints, large loads, DER growth, electrification, and distribution-level impacts.

14. Graphical Comparison of Use Cases

The graphical comparison in the workbook provides a useful way to understand the overall pattern.

The comparison highlights four main observations:

1. System-wide studies rely more on top-down forecasting.
   Resource adequacy, generation expansion, and market design are mainly driven by total system demand, reserve requirements, and system-level flexibility needs.

2. Local network studies rely more on bottom-up forecasting.
   Distribution planning, EV corridor planning, DER hosting capacity, and large-load connections require local data and project-specific assumptions.

3. Transmission planning often sits between both approaches.
   Regional transmission outlooks may begin top-down, but transmission-node forecasting requires reconciliation with bottom-up local information.

4. Integrated system planning requires a reconciled view.
   Generation, transmission, distribution, DER, electrification, and large-load assumptions must be internally consistent.

15. Comprehensive Analysis

The examples show that each planning application requires a different forecasting lens.

1) For resource adequacy, the focus is total system demand and firm capacity. The numerical example converts current peak demand into a future net peak and then applies a reserve margin. This is why the approach is primarily top-down.

2) For generation expansion, the study converts future demand into new capacity requirements after considering reserve margin, retirements, and accredited resources. The approach remains primarily top-down because the key question is system-wide capacity need.

3) For market design, the example focuses on ramping and flexibility. The top-down forecast identifies a system-level ramping gap, which may justify new market products, ancillary services, or demand response mechanisms.

4) For regional transmission outlook, the example begins with a regional top-down forecast but adds local project adjustments. This is important because regional demand growth is rarely distributed evenly across all corridors.

5) For transmission-node forecasting, reconciliation becomes essential. A historical top-down allocation gives one value, while bottom-up local evidence gives another. The planning value should be justified through reconciliation.

6) For distribution planning, the example shows that a substation may remain below its absolute capacity but still exceed a planning trigger. This demonstrates why local planning margins and contingency considerations are important.

7) For data-center or industrial connection, the example shows that requested capacity should not automatically be treated as firm demand. Utilization and realization factors help convert a project request into a planning assumption. However, system-level validation is still needed for large loads.

8) For EV corridor planning, the example shows that managed charging can reduce the local peak but may not fully remove the feeder gap. This highlights the need for local studies, not only system-level EV energy forecasts.

9) For DER hosting capacity, the example shows that the binding limit may be reverse power flow, not thermal loading or voltage. This is why feeder-level assessment is essential.

10) For integrated system planning, the example shows that top-down and bottom-up forecasts can differ materially. The difference should not be ignored. It should be reconciled, explained, and reflected through central and high planning cases.

16. Excel Workbook

Forecasting Approache By Planning Use Case Numerical Examples

97.8KB ∙ XLSX file

Download

Download

The workbook includes one numerical example for each planning application, detailed case sheets, formula-based calculations, planning interpretation, method-dependence indicators, uncertainty scores, and graphical comparisons.

Detailed numerical examples for resource adequacy, generation expansion, market design, transmission planning, distribution planning, data-center connections, EV corridors, DER hosting capacity, and integrated system planning.

17. Key Takeaway

There is no single forecasting approach that fits every planning application.

A. Top-down forecasting is most suitable for system-wide planning decisions such as resource adequacy, generation expansion, and market design.

B. Bottom-up forecasting is most suitable for local network and connection studies such as distribution planning, EV corridor planning, DER hosting capacity, and large-load connections.

C. Hybrid and reconciled forecasting is essential when system consistency and local accuracy are both required.

The key message is simple:

The right forecasting approach depends on the planning question, the required level of detail, and the decision that the forecast is intended to support.

In the next article, the discussion will move into the first technical component of the forecasting process:

Long-Term Base Load Forecasting: Why Historical Demand Alone Is No Longer Enough.

Reference

[1] Energy Systems Integration Group, Long-Term Load and DER Forecasting, A Report by the Energy Systems Integration Group’s Long-Term Load and DER Forecasting Task Force, Aug. 2025.

In the previous article, I discussed why the temporal and spatial granularity of a load forecast must match the planning question. Once the required level of detail has been defined, the next methodological question is:

Should the forecast be developed from the system level downward, or from customers and local networks upward?

These two approaches are commonly known as top-down forecasting and bottom-up forecasting.

1. A top-down forecast begins with the wider system, market, utility service territory, or regional economy. Future demand is estimated at an aggregated level and then allocated to load zones, transmission nodes, substations, or customer classes.

2. A bottom-up forecast begins with individual customers, buildings, technologies, feeders, substations, or proposed projects. These local forecasts are then aggregated to form a wider system outlook.

Neither approach is universally better. Each serves different planning purposes, requires different datasets, and introduces different forms of uncertainty. In modern power system planning, the strongest methodology often combines both approaches and reconciles their results before investment decisions are made [1].

2. What Is Top-Down Load Forecasting?

Top-down forecasting starts with aggregated system demand.

The forecast is normally developed using broader variables such as:

• historical system demand

• population growth

• gross domestic product and economic activity

• employment and industrial trends

• weather and climate assumptions

• electricity prices

• energy-efficiency trends

• electrification targets

• DER and technology-adoption assumptions

• government policies and regulatory targets

The result may initially be a forecast for an entire power system, utility service territory, or load zone. It can then be disaggregated to more detailed network locations.

A typical top-down process may follow this sequence:

Top-down methods are commonly used for:

• resource adequacy

• generation expansion planning

• market design

• long-term transmission planning

• utility financial planning

• regional energy outlooks

• policy and electrification scenarios

The main advantage of this approach is consistency. It provides one system-wide view of future demand and ensures that the sum of the allocated regional forecasts remains aligned with the overall planning outlook.

Its main limitation is that it may fail to capture concentrated local growth. A system forecast may appear reasonable at the aggregate level while missing a data-center cluster, industrial development, EV charging corridor, or rapidly growing distribution area.

3. What Is Bottom-Up Load Forecasting?

Bottom-up forecasting starts with more detailed local information.

The forecast may be developed using:

• customer billing or AMI data

• feeder and substation SCADA measurements

• customer classes and end-use consumption

• known customer connection requests

• industrial and commercial developments

• data-center applications

• EV charging infrastructure

• building electrification

• rooftop PV and behind-the-meter BESS

• local population and land-use development

• customer technology-adoption models

The local forecasts are then aggregated upward through the network hierarchy.

A typical bottom-up process may follow this sequence:

Bottom-up methods are particularly useful for:

• distribution planning

• feeder and substation reinforcement

• known new customer loads

• large-load connection assessment

• data centers and industrial clusters

• EV charging infrastructure

• DER adoption and hosting capacity

• geographically concentrated load growth

The main advantage is geographic detail. Bottom-up forecasting can identify local load pockets that may not be visible in a system-wide forecast.

Its limitation is that it can overstate future demand if project realization, customer diversity, load coincidence, and uncertainty are not properly considered. The sum of all requested or proposed loads does not necessarily represent the amount of demand that will actually materialize.

4. Comparing the Two Approaches

Top-down and bottom-up forecasting differ mainly in their starting point, data requirements, level of geographic detail, and typical planning application.

The comparison shows that top-down forecasting provides a consistent system-wide outlook, while bottom-up forecasting provides greater visibility of local demand growth.

The choice should therefore depend on the planning decision:

• System-wide adequacy and strategic outlooks generally require a top-down framework.

• Local network and connection decisions usually require bottom-up evidence.

• Integrated planning requires the two forecasts to be reconciled.

5. Use Cases for Top-Down Approach

a. Approach Regional Resource Forecast

Consider a system operator preparing a 20-year demand forecast for resource adequacy and generation planning.

The starting data may include:

• historical annual energy and peak demand

• population and economic growth

• weather-normalized demand

• electrification assumptions

• energy efficiency

• rooftop PV

• behind-the-meter storage

• industrial development

• policy and market scenarios

The system operator may develop low, central, and high demand-growth scenarios. Each scenario may include annual peaks, energy forecasts, and 8,760 hourly demand profiles.

The results are then used to assess:

• future generation capacity

• storage requirements

• reserve margins

• resource adequacy

• market procurement

• transmission transfer needs

This approach provides a consistent system-wide control total.

However, it may not show exactly where the demand will appear. A 3% increase in regional demand does not mean every transmission node or substation will grow by 3%.

b. Allocation to Transmission Nodes

A transmission planner may receive a load-zone forecast and need to allocate it to individual buses for power-flow and stability studies.

The forecast may be allocated using:

• historical peak-demand contribution

• annual energy contribution

• customer numbers

• previous growth trends

• planned development zones

• information from transmission and distribution owners

This process is called disaggregation.

A key risk is the use of fixed historical allocation factors. Future load growth may not follow historical patterns. Data centers, industrial plants, EV hubs, and DER adoption may be concentrated in specific areas.

For this reason, transmission-node allocations should be reviewed against bottom-up local information.

6. Use Cases for Bottom-Up Approach

a. Distribution Forecast

Consider a utility preparing a five- to ten-year forecast for a distribution substation.

The utility may begin with:

• historical substation and feeder loading

• organic customer growth

• new service applications

• planned housing and commercial developments

• EV charging

• heat-pump adoption

• rooftop PV

• behind-the-meter storage

• energy efficiency

• demand flexibility

The utility may forecast each feeder separately and then aggregate the feeder forecasts to the substation.

The resulting forecast can support decisions on:

• transformer replacement

• substation expansion

• feeder reinforcement

• voltage improvement

• protection upgrades

• BESS or non-wires alternatives

• connection capacity

This method reveals local constraints, but planners must consider diversity. Individual feeders and customers may not reach their maximum demand at the same time.

b. Large-Load Pipeline

Large-load forecasting is one of the clearest applications of a bottom-up approach.

Consider three proposed data centers, each requesting a 300 MW connection. The total requested capacity is 900 MW, but this does not necessarily mean 900 MW will materialize.

Each project should be assessed using information such as:

• development stage

• land and planning approvals

• financial commitment

• connection agreement status

• expected energization date

• construction progress

• ramp-up profile

• expected utilization

• duplication across multiple connection applications

• probability of realization

The projects may be classified using confidence levels such as:

• certain

• probable

• possible

• uncertain

• forecast or conceptual

A realization factor can then be applied to each project.

For example:

• Project A: 300 MW × 90% realization

• Project B: 300 MW × 60% realization

• Project C: 300 MW × 30% realization

The resulting central forecast would be lower than the total requested connection capacity.

Low, central, and high scenarios should also be developed because project timing and utilization remain uncertain.

7. Numerical Example: Why the Results May Differ

The Excel workbook includes an illustrative numerical example comparing a top-down system forecast with a bottom-up local forecast.

The top-down example starts from the current system peak and applies:

• organic growth

• electrification uplift

• energy-efficiency reduction

• DER and net-load reduction

The resulting system forecast is then allocated to five areas using adjusted growth and allocation factors.

The bottom-up example forecasts the same areas using:

• current local demand

• organic growth

• known new customer loads

• realization factors

• EV and electrification demand

• DER reductions

The total bottom-up forecast is higher because it includes a pipeline of known local projects that is not fully captured by the macro-level system outlook.

The numerical gap does not automatically mean one forecast is wrong. It indicates that planners should investigate:

• whether the system outlook is missing known projects

• whether local projects have been overestimated

• whether realization factors are reasonable

• whether demand has been double-counted

• whether local peaks are coincident

• whether DER assumptions are consistent

• whether the same electrification growth has been included in both forecasts

8. Why Reconciliation Is Essential

Top-down and bottom-up forecasts often produce different results.

A top-down forecast may understate local growth because it is based on macroeconomic trends and historical allocation factors.

A bottom-up forecast may overstate system demand because it includes:

• speculative projects

• duplicate connection requests

• high utilization assumptions

• non-coincident local peaks

• limited geographic diversity

• projects that may be delayed or cancelled

Forecast reconciliation is the process of understanding and managing these differences.

A practical reconciliation process may include the following steps:

1. Establish the top-down system forecast as the control total.

2. Aggregate the bottom-up forecasts to the same geographic level.

3. Compare the assumptions, components, and planning horizon.

4. Identify known large loads that are absent from the top-down forecast.

5. Review realization factors and commissioning dates.

6. Check DER, electrification, and energy-efficiency assumptions.

7. Assess coincident and non-coincident peaks.

8. Remove duplication and double-counting.

9. Develop low, central, and high reconciled scenarios.

10. Document the remaining difference and its planning implications.

Reconciliation should not force both forecasts to become identical. Its purpose is to ensure that the difference is understood, justified, and appropriately reflected in planning studies.

9. A Hybrid Forecasting Approach

Modern power system planning increasingly benefits from a hybrid methodology.

A practical hybrid approach is:

1. Top-down system forecast
   Provides the system-wide outlook and overall control total.

2. Bottom-up local forecast
   Identifies known projects, local demand pockets, DER concentrations, and infrastructure-specific needs.

3. Reconciliation and scenario analysis
   Aligns the two perspectives and manages uncertainty.

This can be expressed simply as:

System consistency + local intelligence + uncertainty analysis = planning-ready forecast

The top-down forecast gives direction. The bottom-up forecast gives location and detail. Reconciliation ensures that planning decisions do not rely blindly on either one.

10. Choosing the Appropriate Method

The forecasting method should be selected according to the use case.

The objective is not to select one method and reject the other. The objective is to use each method where it provides the greatest value.

11. Excel Workbook - Numerical Examples

The supplementary workbook includes a detailed comparison of top-down and bottom-up forecasting, practical example cases, numerical calculations, forecast reconciliation, and a comparison chart. The values are illustrative and are intended to demonstrate the methodology rather than represent a specific market or utility forecast.

Top Down Vs Bottom Up Load Forecasting Numerical Examples

27.6KB ∙ XLSX file

Download

Download

12. Key Takeaway

Top-down and bottom-up forecasting answer different parts of the same planning question.

Top-down forecasting provides a consistent view of total future demand. Bottom-up forecasting explains where that demand may appear and which customers, technologies, or developments are driving it.

Used alone, either method may produce an incomplete picture.

The key message is:

Top-down forecasting provides system-wide direction, bottom-up forecasting provides local detail, and forecast reconciliation connects the two into a planning-ready outlook.

In the next article, the discussion can move deeper into the first technical component of the forecasting process.

13. Reference

[1] Energy Systems Integration Group, Long-Term Load and DER Forecasting, A Report by the Energy Systems Integration Group’s Long-Term Load and DER Forecasting Task Force, Aug. 2025.

In the previous articles, I discussed how future demand shapes grid investment and how different stakeholders use load forecasts for planning decisions. Before moving deeper into base load forecasting, it is important to first discuss one key methodological question:

At what level of detail should a load forecast be prepared?

This question is usually described as forecast model granularity.

In simple terms, granularity means the level of detail used in the forecast. A forecast can be developed at system level or feeder level. It can be annual, monthly, daily, hourly, or even sub-hourly. It can represent one large region, one load zone, one transmission node, one distribution substation, or one customer class.

This matters because a forecast is only useful if its level of detail matches the planning question. A forecast that is good enough for long-term resource planning may not be detailed enough for distribution planning. Similarly, a feeder-level forecast may be too detailed for market-wide adequacy studies unless it is properly aggregated.

The ESIG report highlights that modern load and DER forecasting increasingly requires both temporal granularity and geographic granularity to capture changing demand patterns, DER adoption, electrification, and localized load growth [1].

2. What Is Forecast Model Granularity?

Forecast model granularity can be understood in two dimensions:

1. Temporal granularity

This refers to how demand is represented over time, e.g.

• annual peak demand

• monthly energy demand

• daily load profile

• hourly load profile

• 8,760 hourly time-series forecast

2. Spatial granularity

This refers to where demand is represented in the network, e.g.

• system level

• load zone level

• transmission node level

• distribution substation level

• feeder level

• customer or meter level

Both dimensions are important.

A planner may know the total demand growth of a region, but if the forecast does not show where the demand appears, it may miss local constraints. Similarly, a planner may know the annual peak demand, but if the forecast does not show when the demand appears, it may miss ramping needs, midday minimum load, evening peaks, or seasonal operating stress.

3. Why Temporal Granularity Matters

Traditional long-term load forecasts often focused on annual energy and annual peak demand. This was useful when electricity demand was relatively predictable and the system was dominated by centralized generation.

However, modern demand is becoming more dynamic.

Electric vehicles may charge outside traditional peak periods. Rooftop PV may reduce midday net load. Batteries may shift demand between hours. Heat pumps may create new winter peaks. Data centers may introduce large and relatively flat demand profiles. Extreme weather may create short-duration but high-impact demand events.

This is why hourly forecasting is becoming more important.

An 8,760 hourly forecast represents every hour of the year. It allows planners to see not only the annual peak, but also:

• hourly load shape

• seasonal variation

• minimum load conditions

• evening ramps

• non-coincident local peaks

• DER impact on net load

• EV charging behaviour

• weather-driven demand patterns

This does not mean every study always needs an 8,760 forecast. It means the time resolution should be selected based on the planning purpose as shown in below table.

The key point is simple: as the grid becomes more dynamic, time resolution becomes more important.

4. Why Spatial Granularity Matters

Spatial granularity is equally important.

A system-level forecast may show that the overall grid has enough capacity, but this does not mean every local area is secure. A specific substation, feeder, industrial zone, data center cluster, or EV charging corridor may become constrained even if the wider system appears adequate.

This is especially important for:

• EV fast-charging hubs

• data center clusters

• industrial electrification

• rooftop PV concentration

• behind-the-meter BESS

• heat pump adoption

• new urban developments

• distribution-level load pockets

For example, a 500 MW increase in demand spread across a large region may be manageable. But the same 500 MW concentrated near one transmission node or one group of substations may require major reinforcement.

This is why modern planning increasingly needs forecasts that can move between different geographic levels:

• system forecast

• load zone forecast

• transmission node forecast

• substation forecast

• feeder forecast

• customer class forecast

The forecast must be detailed enough to reveal the planning issue.

5. The Trade-Off: Detail Versus Complexity

Higher granularity improves visibility, but it also increases complexity.

A highly detailed forecast needs better data, more assumptions, stronger modelling capability, and more validation. For example, an hourly forecast at feeder level over a 20-year horizon creates a large amount of data. A customer-level DER adoption forecast may be very detailed, but it may also carry significant uncertainty.

Therefore, planners must balance:

• accuracy

• data availability

• modelling complexity

• computational effort

• planning horizon

• stakeholder needs

• decision impact

The objective is not to make every forecast as detailed as possible. The objective is to make the forecast fit for purpose.

A financial planning forecast may be acceptable at customer-class or service-territory level. A transmission planning forecast may need allocation to transmission nodes. A distribution planning forecast may need substation or feeder-level detail.

6. Aggregation and Disaggregation

Another important methodological concept is the movement of forecasts between different levels of the grid.

• Aggregation means combining lower-level data into a higher-level forecast.

For example:

• Disaggregation means allocating a higher-level forecast down to more detailed grid locations.

For example:

Both approaches are used in planning.

However, both can create errors if not handled carefully. A top-down system forecast may miss local load pockets. A bottom-up distribution forecast may overestimate total system demand if diversity and coincidence are not properly considered.

This is why reconciliation is important. Planners need to check whether system-level, transmission-level, and distribution-level forecasts are consistent enough for the intended planning use.

7. Matching Granularity with the Planning Use Case

A good forecasting methodology starts by asking:

What decision will this forecast support?

Different planning studies need different forecast resolution.

This shows why a single forecast cannot answer every planning question. The same demand outlook may need to be represented differently depending on whether the question is about adequacy, congestion, voltage, system strength, distribution reinforcement, tariff design, or grid connection.

8. Why This Comes Before Base Load Forecasting

Before developing a base load forecast, planners should first understand the required forecast resolution.

This is important because the base load forecast may be built differently depending on the intended use. A system-level base forecast may rely on economic, demographic, and weather-normalized trends. A feeder-level forecast may require SCADA, AMI, customer connection data, and local development information.

Therefore, the correct sequence is:

1. Define the planning use case

2. Select the temporal and spatial granularity

3. Collect and validate historical data

4. Develop the base load forecast

5. Add known new customer loads

6. Add DERs, electrification, and flexible demand

7. Apply scenarios and uncertainty analysis

8. Produce the final planning forecast

This sequence helps avoid one common problem: developing a forecast first and only later discovering that it does not have the right resolution for the planning study.

9. Key Takeaway

Forecast model granularity is not just a modelling detail. It is a planning decision.

If the forecast is too broad, it may hide local constraints. If it is too coarse in time, it may miss ramping needs, minimum load conditions, or non-coincident peaks. If it is too detailed without good data, it may create false confidence.

The key message is simple:

A load forecast must match the planning question. The right level of time and location detail is essential for reliable planning, efficient investment, and effective grid connection decisions.

In the next article, the discussion can move into the components of the long-term load and DER forecasting process, including top-down and bottom-up approaches, base load forecasting, known new customer loads, DERs, electrification, uncertainty, and scenario planning.

Reference

[1] Energy Systems Integration Group, Long-Term Load and DER Forecasting, A Report by the Energy Systems Integration Group’s Long-Term Load and DER Forecasting Task Force, Aug. 2025.

A load forecast is not prepared for one department, one study, or one decision. In modern power systems, it is a shared planning input used by utilities, system operators, regulators, developers, consultants, and other stakeholders to decide how the grid should evolve.

In the previous article, I discussed how future demand shapes grid investment and connection opportunities.

The next logical question is:

who actually uses these forecasts, and why do different stakeholders need them?

The answer is important because load forecasting is no longer limited to estimating annual energy demand and peak load. It now supports resource planning, transmission planning, distribution planning, financial planning, customer programs, market design, DER integration, and grid connection strategy.

As demand patterns become more complex due to electrification, data centers, EV charging, distributed solar, battery storage, flexible demand, and extreme weather, the value of a forecast depends not only on its accuracy, but also on how well it is understood, shared, and coordinated across planning stakeholders [1].

2. Load Forecasting as a Shared Planning Function

In a traditional power system, load forecasting was often treated as a relatively narrow planning function. Planners estimated future peak demand and annual energy, and those values were used to guide generation and network investment. That approach is no longer sufficient.

Today, the same forecast may be used by different teams for different purposes. A system operator may use it for resource adequacy. A transmission planner may use it to assess network reinforcement. A distribution utility may use it to identify overloaded feeders or substations. A financial planning team may use it to estimate revenue requirements. A customer program team may use it to design demand response, energy efficiency, EV charging, or DER programs.

This means load forecasting is not just a technical calculation. It is a coordination point between different parts of the electricity sector.

A useful way to think about it is:

If the forecast is inconsistent, incomplete, or poorly coordinated, the planning decisions that depend on it may also become inconsistent.

3. Key Stakeholders in Load Forecasting

Long-term load forecasting is used by several stakeholders, each with a different role in the planning process. Utilities use forecasts for distribution planning, infrastructure investment, customer programs, financial planning, and regulatory submissions. System operators, ISOs, and RTOs use forecasts for resource adequacy, market design, transmission planning, and system reliability studies.

Regulators and government agencies influence the forecast through policy targets, electrification programs, DER incentives, and planning requirements. Developers and large customers also shape the forecast when new generation, BESS, data centers, industrial loads, or hydrogen projects enter the planning pipeline. Consultants, vendors, and research organizations support the process through modelling tools, datasets, scenario studies, and technical analysis.

In simple terms, load forecasting is not owned by one stakeholder. It is a shared planning input used across the electricity sector to guide technical, commercial, regulatory, and investment decisions.

4. Main Use Cases of Load Forecasting

Different stakeholders use the same forecast in different ways. This is why the forecast must be clear, traceable, and suitable for the intended planning purpose.

4.1. Resource Planning and Market Design

Resource planners use load forecasts to determine the future need for generation, storage, capacity, reserves, and flexibility. The forecast helps answer questions such as:

• How much capacity will be needed?

• What type of generation mix is required?

• How much storage is needed?

• What level of reserve margin is required?

• How will demand shape affect dispatch and market design?

As load patterns become more variable, resource planning increasingly needs hourly or time-series forecasts, not only annual energy and peak demand values.

4.2.Transmission Planning

Transmission planners use load forecasts to identify future power flow patterns, network constraints, transfer needs, and reinforcement requirements. Forecasts help determine whether the system needs:

• new transmission lines

• transformer upgrades

• reactive power support

• revised transfer limits

• system strength solutions

• new substations or switching stations

• reinforcement near large load centers

Transmission planning usually requires a longer-term view because major network projects take many years to plan, approve, permit, and build.

4.3. Distribution Planning

Distribution planners use forecasts at a more granular level. They are concerned with local substations, feeders, transformers, cables, voltage levels, protection limits, and local hosting capacity.

This is where geographic detail becomes very important. A system may look adequate at the transmission level, while a local feeder or substation may be overloaded because of EV charging, rooftop solar, heat pumps, or a new commercial development.

Distribution planning therefore needs local forecasts that can identify load pockets and non-coincident peaks.

4.4. Financial Planning and Tariff

Forecasts also support utility financial planning and tariff design.

If energy sales, peak demand, and customer behavior change, the revenue requirement and cost recovery structure may also change. This becomes especially important when customers install behind-the-meter PV, batteries, EV chargers, or participate in demand response and time-varying rates. In this context, load forecasting supports:

• revenue projections

• tariff design

• cost-of-service studies

• investment planning

• customer class analysis

4.5. Customer Programs and DER Planning

Customer-facing teams use forecasts to design programs for:

• energy efficiency

• demand response

• EV charging

• rooftop solar

• behind-the-meter BESS

• building electrification

• flexible demand

These programs can reduce, shift, or reshape demand. Therefore, they are not only outputs of planning; they also become inputs into future forecasts.

5. Why Coordination Is Becoming More Important

The major challenge is that different stakeholders often produce different forecasts for the same or overlapping regions. For example:

• a system operator may prepare a system-level forecast

• a utility may prepare a service-territory forecast

• a distribution team may prepare feeder-level forecasts

• a financial team may use customer-class forecasts

• a regulator may use policy-driven adoption scenarios

• a consultant may develop project-specific assumptions

All of these forecasts may be valid for their own purpose, but they can create problems if they are not coordinated.

The ESIG report highlights that multiple entities often produce forecasts across overlapping areas and planning horizons, which can create discrepancies in assumptions, methods, and data. This is why coordination, data-sharing, and consistent assumptions are becoming essential in modern grid planning [1]. The following Figure describe the Data Sharing and Coordination in a Region or State with Market Operators

6. What Happens When Forecasts Are Not Aligned

When forecasts are not aligned, the consequences can be serious:

A transmission planner may underestimate local growth if distribution-level data is not reflected. A distribution planner may overbuild if a local forecast is too conservative. A resource planner may miss future adequacy needs if large loads are not properly included. A system operator may underestimate future ramping needs if DERs and electrification are not represented correctly.

Misalignment can lead to:

• underinvestment

• overinvestment

• delayed grid connections

• poor timing of network upgrades

• missed local load pockets

• inconsistent planning assumptions

• duplicated or double-counted demand

• resource adequacy gaps

• transmission bottlenecks

• higher costs for customers

This is why a forecast should not be treated as a single number. It should be treated as a shared planning dataset with clear assumptions, ownership, resolution, and use case.

7. Moving Toward Integrated Planning

Modern power system planning requires better integration between forecasting entities.

This means improving:

• data sharing between utilities, system operators, and regulators

• consistency between transmission and distribution forecasts

• coordination between customer account teams and planning teams

• transparency of assumptions for DERs and large loads

• alignment between policy targets and technical models

• reconciliation between top-down and bottom-up forecasts

• scenario planning for uncertainty

Integrated planning does not mean that every stakeholder uses the same model. It means that the assumptions, data flows, and planning outputs are coordinated enough to avoid major misalignment. This is especially important as future demand becomes more uncertain and more location-specific.

8. Key Takeaway

The value of a load forecast depends not only on its technical accuracy, but also on how well it is coordinated across the stakeholders who use it.

A forecast that is accurate at system level but disconnected from local distribution realities may miss load pockets. A forecast that is detailed at feeder level but not reconciled with system planning may create confusion. A forecast that ignores large loads, DERs, electrification, or customer behavior may lead to poor investment decisions.

Load forecasting is not only a planning input. It is a coordination tool that connects stakeholders, aligns assumptions, and shapes the future development of the power system.

9. Reference

[1] Energy Systems Integration Group, Long-Term Load and DER Forecasting, A Report by the Energy Systems Integration Group’s Long-Term Load and DER Forecasting Task Force, Aug. 2025.

Every power system planning process begins with one fundamental question: how much electricity will be needed, where will it be needed, and when will it be needed?

This question may look simple, but it sits at the centre of almost every major planning decision in the electricity sector. Future demand influences how much generation capacity is required, where transmission reinforcements may be needed, which substations and feeders may become constrained, and where new connection opportunities may emerge.

Traditionally, long-term load forecasting was mainly focused on two key indicators:

1. total annual energy demand and annual peak demand. This approach was generally sufficient when the grid was dominated by large centralized generation and relatively predictable customer demand patterns.

2. However, the modern power system is changing quickly. Electrification, data s, manufacturing growth, customer-sited DERs, EV charging, building electrification, behind-the-meter solar, battery storage, and extreme weather are changing not only the amount of demand, but also its timing, location, and operating characteristics [1].

This is why load forecasting is no longer just a statistical exercise. It is now one of the most important inputs to generation planning, transmission planning, distribution planning, market design, infrastructure investment, and grid connection strategy.

2. Future Demand as the First Planning Signal

Future demand is often the first signal that the power system may need to change. If demand is expected to grow in a particular region, planners must ask whether the existing network can serve that growth reliably. If demand is expected to shift because of EV charging, heat pumps, or data centres, planners must assess whether existing peak assumptions are still valid. If behind-the-meter PV or battery storage changes the shape of net load, planners must understand whether the system will need more ramping capability, storage, flexible demand, or operational tools.

The references reports highlights that long-term load and DER forecasting supports many planning functions, including generation capacity, energy markets, grid infrastructure, DER planning, customer rates, and programs. In simple terms, the forecast becomes the foundation on which future system needs are assessed [1] - [4].

A useful way to understand this is:

When the load forecast changes, almost every downstream planning decision can change with it. A higher demand forecast may create a need for new generation or storage. A geographically concentrated forecast may create a need for local network reinforcement. A changing hourly demand shape may create a need for flexibility, demand response, or BESS. A new large load may trigger a transmission connection, substation expansion, or system impact study.

3. Why Traditional Forecasting Is No Longer Enough

Traditional load forecasting often worked well when demand followed relatively stable historical patterns. In that environment, planners could rely heavily on economic growth, population growth, historical load trends, and weather-normalized peak demand.

That approach is now becoming less sufficient. Modern demand is being reshaped by several new drivers:

These drivers do not simply increase or decrease total electricity demand. They change the shape of demand. They affect when demand appears during the day, which season produces the highest stress, whether the system peak occurs in summer or winter, and whether a local feeder or substation peaks at the same time as the wider system.

For this reason, modern demand forecasting emphasizes the growing need for high-resolution, time-based forecasting, including 8,760 hourly forecasts. Annual energy and peak demand values may still be useful, but they may miss important operational details such as midday minimum load, evening ramping, local non-coincident peaks, weekend or seasonal peaks, and the interaction between weather, demand, and renewable generation [1] - [4].

In modern planning, the question is not only: How much demand will there be?

It is also: Where will the demand appear? When will it peak? What will drive it? How flexible is it? How certain is it? And how will it interact with DERs and the network?

4. How Demand Drives Network, Generation and Investment Decisions

Once future demand is forecast, planners begin to assess whether the existing system can accommodate it.

1. For generation planning, demand forecasts help determine how much capacity, energy, storage, reserve, and flexibility may be required. If demand grows faster than expected, additional generation or storage may be needed to maintain adequacy. If demand becomes more variable, the system may require more flexible resources. If electrification creates winter peaks, the generation portfolio may need to be assessed under different seasonal conditions.

2. For transmission planning, demand forecasts influence power flow patterns, transfer requirements, congestion risk, voltage performance, and reinforcement needs. A new industrial zone, data cluster, or electrified transport corridor may shift loading patterns across the transmission network. This can create a need for new lines, transformer upgrades, reactive support, system strength solutions, or revised operating limits.

3. For distribution planning, demand forecasts are even more location-specific. A system may appear adequate at the bulk level while individual feeders, substations, or transformers become constrained. This is particularly important where EV charging, heat pumps, rooftop PV, behind-the-meter BESS, or new commercial loads are concentrated in specific areas. ESIG notes that greater geographic granularity can help planners identify localized demand variations and load pockets that may otherwise be missed [1].

A forecast is not just a prediction. It becomes a signal for:

• new generation capacity

• BESS and flexibility services

• transmission reinforcement

• substation expansion

• feeder upgrades

• voltage support

• system strength solutions

• demand response programs

• connection capacity

• grid modernization

In other words, demand drives the need, and the need drives the investment.

5. From Load Forecast to Connection Opportunity

Grid connection opportunities often appear where the forecast reveals a future system need.

• If a region is expected to face load growth, developers may see opportunities for generation, BESS, or hybrid projects. If a corridor is expected to become congested, transmission reinforcement or local generation may become valuable. If a distribution area is expected to experience EV or data growth, utilities may need new substations, feeders, transformers, or flexibility solutions. If the system requires more ramping or peak-shaving capability, BESS and demand-side resources may become attractive.

• Large loads are becoming especially important in this discussion. The references report identifies data s as a major component of growth in utility load forecasts and highlights the challenges of forecasting large load growth, including uncertainty in realization rates, regional variation, and load clustering effects
  [1] - [4].

• A data may not only create a new connection request; it may also change the planning outlook for generation, transmission, resource adequacy, and local grid infrastructure.

This is why the connection process should not be viewed only as an application submitted by a project developer. The opportunity often begins earlier, when load forecasting identifies where the system is likely to need capacity, flexibility, reinforcement, or operational support.

A simple example is a data cluster. If several large loads are forecast in one region, the system may need new transmission capacity, additional local transformation, reactive power support, backup supply arrangements, or dedicated connection studies. This may create opportunities not only for the data customer, but also for network owners, generation developers, BESS providers, OEMs, EPC contractors, consultants, and system operators. The same logic applies to electrified transport corridors, industrial zones, hydrogen production hubs, renewable energy zones, and urban distribution networks with high DER adoption.

6. Key Takeaway

Future demand is not just an input to a planning model. It is one of the earliest indicators of where the power system may need to evolve. A well-developed load forecast helps planners understand future stress points, identify network and generation requirements, assess uncertainty, and prepare investment decisions. A weak or outdated forecast can lead to underinvestment, overinvestment, connection delays, poor allocation of resources, or missed opportunities.

The key message is simple:

Load forecasting is the starting point of power system planning, and power system planning is the starting point of many grid investment and connection opportunities.

In the next article, the discussion can move deeper into the main components of load forecasting: base load, new customer loads, DERs, electrification, large loads, uncertainty, and scenario planning.

7. Reference

[1] ESIG, Long-Term Load and DER Forecasting, Aug. 2025.
[2] AEMO, 2025 Electricity Statement of Opportunities, Aug. 2025.
[3] AESO, 2025 Long-Term Transmission Plan, Jan. 2025.
[4] PJM, 2026 Long-Term Load Forecast Report, Jan. 2026.

The grid connection projects do not begin with a substation drawing, an OEM data sheet, or a grid study. They usually begin much earlier inside a planning cycle run by a transmission system operator, distribution network operator, independent system operator, market operator, or regulator. That planning cycle may be called an integrated system plan, annual planning outlook, long-term outlook, network development plan, network investment plan or reliability assessment, depending on the region. The terminology changes from market to market, but the purpose is broadly the same: identify future system needs early enough that generation, storage, transmission, demand-side resources, and large new loads can be connected in a reliable and economic way [1]–[6].

In this article, the term annual planning is used in a broad sense to describe the recurring process through which a power system looks ahead, tests future scenarios, identifies emerging constraints, and signals where investment or interconnection opportunities may arise. In practice, these planning outputs are not always issued on a strictly annual basis. Like Ontario’s IESO, for example, publishes an Annual Planning Outlook, whereas AEMO’s Integrated System Plan is updated on a multi-year cycle, and AESO prepares its Long-Term Transmission Plan every two years using inputs from its broader Long-Term Outlook. In Europe, ENTSO-E publishes the Ten-Year Network Development Plan, while in North America wider reliability assessments are also undertaken through institutions such as NERC, alongside regulated transmission planning processes under FERC [1]–[6].

How the Power System Planning Cycle Works;

Before examining how projects emerge, it is useful to consider at a high level, how electricity markets function. Each market has its own legal, regulatory, and commercial design, yet the physical realities of electricity remain universal. Electricity must be produced, transported, balanced, and consumed in real time. Because demand evolves, generation portfolios change, networks age, public policy develops, and new technologies continue to enter the system, the power sector cannot rely solely on the existing asset base. It requires a forward-looking planning process capable of translating uncertain futures into practical actions, including transmission reinforcements, additional interconnection capacity, flexibility procurements, new operational tools, and, in some cases, entirely new technical requirements for connection [1]–[6].

This is why planning documents are so important. They are not merely administrative reports or records of institutional thinking. Rather, they serve as the bridge between long-term system needs and real project opportunities. They reveal how system planners interpret the future and where they expect technical, operational, or economic pressures to arise. In this sense, planning documents often provide the earliest meaningful signals to the market [1]–[6].

So how is this planning actually done?

At its core, every planning cycle begins with a fundamental question: what kind of power system will be required in the future, and what factors are most likely to place that system under stress? To answer this, planners construct future scenarios using a combination of demand forecasts, anticipated generator retirements, project development pipelines, fuel assumptions, policy targets, technology cost trajectories, weather and climate considerations, electrification trends, industrial growth, transmission constraints, and reliability criteria [1]–[6].

The principal drivers behind these plans are remarkably similar across all boundaries, even if the language used to describe them differs. Common drivers include load growth, generator retirement, renewable energy integration, storage deployment, industrial electrification, data centre expansion, decarbonization targets, transmission congestion, resilience concerns, and reliability margins. Although each market frames these issues in its own way, the planning challenge is essentially the same: to determine how the future system can remain secure, operable, and economically sustainable under changing conditions [1]–[6].

How Grid Connection Opportunities Emerge;

Once these future scenarios have been developed, planners do not simply publish them as forecasts. They test them. Power flow studies, contingency assessments, transfer analyses, stability studies, adequacy assessments, and, increasingly, combinations of operational and market modelling are carried out to determine where the system may face limitations or new requirements. It is at this stage that where opportunities begin to emerge [1]–[6].

Most grid connection opportunities do not originate as incidental or self-contained ideas. Rather, they arise when the planning process reveals a technical, economic, or reliability need within the system. In many cases, planning identifies one or more deficiencies or requirements: a region may need greater transfer capability, additional voltage support, more dispatchable capacity, faster frequency response, improved system strength, new local supply, or a more balanced relationship between generation and demand. In some markets, such needs are translated into transmission plans or procurement pathways. In others, they appear through congestion signals, hosting-capacity limitations, network expansion, policy-supported development zones, or specific connection windows. The form may differ, but the logic remains consistent: the system first identifies the need, and projects subsequently emerge or compete in response to it. [1]–[6].

That is why developers, OEMs, EPCs, consultants, and major load customers should read planning documents carefully. These reports often tell the market where the system expects future stress, which corridors may need reinforcement, where new capacity is likely to be valuable, what kinds of projects may face stronger scrutiny, and which technical issues are becoming more important. Even when a report does not list a project explicitly, it may signal the type of opportunity that will later appear as an interconnection wave: utility-scale renewables, BESS, synchronous support, HVDC links, flexible demand, or large load connections such as data centers. This is an inference from how the planning frameworks are structured, but it is strongly supported by the fact that these documents are designed to identify needs, evaluate candidate solutions, and inform subsequent investment and regulatory actions [1]–[6].

The most important takeaway is this: grid connection projects do not start at the connection application stage. They start much earlier, when the system begins asking what it will need in five, ten, or twenty years. That is where demand growth, policy, reliability, economics, and engineering first come together. By the time a specific PV plant, BESS project, HVDC scheme, or data centre appears in the queue, it is usually entering a system context that has already been shaped by a planning cycle. Understanding that planning cycle is the first step to understanding the full project lifecycle [1]–[6].

In the next article, the follow-on is to map the stakeholders in that lifecycle: who does what, when the developer enters, where the OEM becomes critical, how consultants shape the technical path, and why the system operator’s early planning assumptions can materially affect the final connection route.

References

[1] Australian Energy Market Operator, 2024 Integrated System Plan (ISP), Melbourne, VIC, Australia, Jun. 2024.

[2] Australian Energy Market Operator, ISP Methodology, Jun. 2025.

[3] Independent Electricity System Operator, Annual Planning Outlook, Ontario, Canada, Apr. 2025.

[4] Alberta Electric System Operator, 2025 Long-Term Transmission Plan, Alberta, Canada, Jan. 2025.

[5] ENTSO-E and ENTSOG, TYNDP 2024 Scenarios Report, Jan. 2025.

[6] North American Electric Reliability Corporation, 2025 Long-Term Reliability Assessment, Atlanta, GA, USA, Jan. 2026.