June 2019

Transparency in long-term electric demand forecasting: a

perspective on regional load forecasting

In this Insights, we illustrate the incremental effects of considering energy efficiency and distributed solar

on load forecasting accuracy. In addition, we examine the implications of future distributed energy

resources (i.e., distributed solar photovoltaic) integration and energy efficiency (EE) adoption on long-term

forecasts.1 Overall, we found:

Consideration of EE and distributed energy resources (DERs) in electricity use forecasting generates

different results from a scenario without EE and DERs.

Different methodologies affect the levels of accuracy performance, which can also vary by customer

segment, demonstrating the importance of accuracy and transparency in regional load forecasting

methods used by utilities.

Forecasted peak demand is different in these EE and DER integration scenarios from scenarios

without these considerations, and the differences widen over time.

The discrepancy between the peak demand forecasts increases with higher levels of solar integration

and the difference is not negligible, especially for forecasts far in the future.

Load forecasting in regional markets

The integration of distributed energy sources and energy efficiency in power markets has stimulated

greener and more sustainable energy, but also creates challenges for electric utilities, resource planners,

and policy-makers in forecasting demand. Currently, electric utilities independently choose methods to

forecast their long-term demand. The forecasts are used to help with planning for energy delivery to end-

use customers, revenue requirements, and rate design for each customer segment—residential,

commercial, and industrial. Customer-specific demand forecasts are typically estimated separately and

then aggregated to represent the total customer demand. Electric utilities provide these aggregated

forecasts to the regional transmission and system planners for regional load planning. However, regulators

1 The author would like to acknowledge and thank Bixuan Sun and Rao Konidena for their contributions to the research and ideas

contained in this paper originally published by the Institute of Electrical and Electronics Engineers,

https://smartgrid.ieee.org/newsletters/may-2019.

CRA Insights: Energy | 2

do not require electric utilities to use a uniform approach for their load forecasts. Figure 1 illustrates the use

of electric load forecasting, and the information flow among utilities and regional operators.

Since each utility has its own business models, service territories, and customer profiles there is often no

transparency in the granularity of forecasting approaches employed across utilities, raising the question: is

the aggregation of different forecast outcomes appropriate, given the different forecasting approaches

employed by utilities? In addition, there is no uniform approach on how to account for EE and the

integration of DERs in long-term load forecasting for different customer segments. Although it is not

common practice now, some utilities have made adjustments to account for demand-side energy

resources including demand response, distributed solar, and energy efficiency savings in their load

forecasts for different end-use customer segments. However, these adjustments are based on various

assumptions, which may introduce inaccuracies to the projected distributed solar generation and energy

efficiency savings.

Figure 1: Illustration of the use of load forecasting2

Method

We use two empirical methods to predict regional electricity demand for residential and commercial

customer segments and illustrate the results using the example of the Midcontinent Independent System

Operator (MISO) control area, including Minnesota, Iowa, Wisconsin, North Dakota, Illinois, Michigan,

Indiana, and Louisiana. State-level energy efficiency is measured using the American Council for an

Energy Efficient Economy (ACEEE) rating. The level of distributed energy resource integration is

measured by the “Freeing the Grid” rating on net metering standards. Our analysis focuses on forecasting

monthly electricity sales by customer segment. The following scenarios are analyzed and compared to

show the differences in electric forecasts:

Scenario 1 – Traditional model vs Updated model: forecasting residential and commercial electricity

demand using the same empirical method with and without EE and DER adjustments.

2 Our empirical specifications for residential and commercial demand estimation are widely used in academic and industry

literature (Hagen and Behr 1987; Kyriakides & Polycarpou, 2007; Hahn et al., 2009) and are widely used by the electric and gas

utilities to conduct short-term and long-term forecasting for their customer segments (Hahn et al., 2009). Data from 2007 to 2015

are used to estimate the model, and data from 2016 to 2017 are used for forecasting. Note that electricity demand forecast is

estimated controlling for weather and state-specific economic factors in addition to price of electricity and price of substitutes

(i.e., natural gas).

CRA Insights: Energy | 3

Scenario 2 – Different methods: forecasting residential and commercial electricity demand using different

empirical methods. Specifically, we employed ordinary least squares (OLS-Method 1) and time-series

(TS-Method 2) estimation methods. The time-series model takes into account the time trend and

correlations among past electricity consumptions, while the OLS model treats each month of consumption

as independent.

Scenario 3 – Different methods and regions: forecasting residential electricity demand in two different

states with different methods.

Illustration

Absolute forecast error, which is the absolute difference between actual and forecasted values, is often

used to evaluate the accuracy of forecasts and enables comparison of different forecasting methods.

Forecast error increases for several reasons including incorrect forecast approaches, model

misspecification, and data limitations.

Figure 2 shows how the forecasting models perform differently in predicting the average monthly electricity

sales per customer across the seven states in the MISO footprint , by plotting the absolute forecast error in

each month of 2016 and 2017. The OLS method with EE and DER adjustments does not seem to

significantly improve forecast accuracy in the residential sector (Figure 2, Panel A), but makes a significant

difference in reducing forecast errors in the commercial sector (Figure 2, Panel B). Overall, TS models

perform best in terms of reducing forecast errors in both sectors because they take into account the

seasonal and cyclical nature of electricity demand, as well as EE and DER adjustments.

Figure 2: Prediction errors of different methods

Panel A. Residential sector

CRA Insights: Energy | 4

Panel B. Commercial sector

Next, we forecast residential electricity sales per customer for two states within the MISO territory,

Minnesota (MN) and North Dakota (ND) for the months of 2016 and 2017. After estimating state-specific

forecasts, we aggregate electric load forecasts and compare forecasted loads using the same econometric

approach versus different approaches. Figure 3 shows that forecasted load using the same method has a

different prediction than forecasted load using different methods. We also calculate the forecast error for

forecasting using same v. different methods for MN and ND. There is a significant difference in accuracy

between the forecasts using the same method and the ones using different methods.

CRA Insights: Energy | 5

Figure 3: Predicted load for different states and methods (residential load in MN and ND)

Simulation – Projecting residential summer peak demand

To further illustrate the long-term implications of not including DER and EE savings on peak demand

forecasting in the longer term, we developed a simulation model to project residential summer peak

demand through 2030 in Minnesota considering different solar PV integration and EE scenarios. Our

model shows that the differences in forecasting outcomes widen over a longer time horizon and with

increasing uptake of solar PV.3 Figure 4 shows the percentage of residential peak demand reduction per

customer achieved by five solar PV penetration scenarios in 2020, 2025, and 2030, while holding the peak

demand reduction from EE at 10%.

Based on our peak demand simulations, we conclude the following:

• Considering residential distributed solar and energy efficiency in peak demand generates different

forecasting results from the scenario without these considerations.

• The forecasted peak demand is different in these DER and EE integration scenarios, and the

differences widen over time.

3 In the simulation, we first project residential electricity use (MWh) from 2018 to 2030 using the historical average annual growth

rate from 2007 to 2017. We then convert the electricity use to summer peak demand using the formula developed by State Utility

Forecasting Group at Purdue University: (forecasted annual energy (MWh)/total numbers of hours in a year)×1.568. Next, we

assume 5%, 10%, 15%, 20%, and 25% of residential summer peak demand is supported by distributed solar and average 10% of

residential summer peak demand is reduced by energy efficient appliances and programs (e.g., LED lights, ENERGY STAR®

appliances) in Minnesota by 2030. We then derive the paths of residential summer peak demand using the average annual

growth rates required to meet these solar PV and EE demand reduction goals.

CRA Insights: Energy | 6

• The discrepancy between the peak demand forecasts increases with higher levels of solar integration

scenarios and the difference is not negligible, especially for forecasts far in the future.

Figure 4: Peak demand reduction under different distributed PV integration scenarios

Discussion

Transitioning to a cleaner and more sustainable electric power system by adopting EE and DERs has

raised certain complications in using the existing load forecasting tools for policy-making and resource

planning. Energy efficiency measures, especially utility sector energy efficiency programs, can

substantially reduce peak energy demand and energy use and improve system reliability (Berg, 2016).

Distributed generation, especially distributed solar PV generation, brings opportunities and challenges to

electricity load management. The PV generation can bring electricity demand to extremely low levels at the

hottest hours of the day. In the evening when people return home to turn on their appliances, PV

generation also declines, which requires flexible generation capacity to come online quickly. The

intermittent nature of distributed sources, as well as the timing mismatch between generation and demand,

leads to an increase in cycling-related costs (Perez-Arriaga and Batlle, 2012).

One complication is that traditional forecasting tools do not necessarily consider the impacts of EE and

DERs on load forecasting. A survey conducted by Carvallo et al. (2016) shows that the sets of variables

used for load forecasting by load serving entities (LSEs) have not changed for a long time, and only a few

of them adopted new forecasting techniques. Utilities and agencies can make inefficient and suboptimal

procurement decisions based on these inaccurate load forecasts. This analysis shows that the inclusion of

EE and DER adjustments is important by showing improved forecasting results once these adjustments

are included into the electric load forecasting analysis.

In addition, long-term forecasts often have larger forecast errors due to greater uncertainty. Our analysis

demonstrated that a method with good forecast accuracy in one sector (e.g. residential) might not perform

as well in another (e.g. commercial). Therefore, aggregating load forecasts from multiple electric utilities,

without knowing their approaches, can introduce bias into load forecasting for regional operators. Regional

system operators, such as MISO, have limited authority to check long-term load forecasts. MISO can only

audit load forecasts one year into the future for its voluntary annual capacity auction. The remaining years

are not audited by MISO because individual utilities do not need to provide capacity more than one year

into the future. This introduces additional challenges when it comes to verifying long-term regional load

CRA Insights: Energy | 7

forecasting accuracy. Transparency across utilities in terms of their load forecasting methods is important

to inform regional forecasts.

Another implication of load forecasting is resource-planning related. Transmission planning load forecasts

are “point” forecasts, which represent forecast load at a snapshot in time, while the wholesale market load

forecast is predicted in real time. The former is submitted to meet the North American Electric Reliability

Corporation (NERC) obligations under the functional model, while the latter is submitted to adhere to the

Federal Energy Regulatory Commission (FERC) tariff. Therefore, inaccurate load forecasts aggregated by

regional regulators can potentially impact both transmission and wholesale operational planning. Although

some utilities make adjustments to their load forecasts to account for solar and energy efficiency savings,

these savings are usually netted out from the forecasts instead of being included in the forecasting models.

This study serves as an applied demonstration of these potential issues using basic econometric methods

and further research in this area is essential for the US power industry.

References

Aung, Z., Toukhy, M., Williams, J., Sanchez, A., & Herrero, S. (2012, February). “Towards accurate

electricity load forecasting in smart grids.” In The Fourth International Conference on Advances in

Databases, Knowledge, and Data Applications, DBKDA, available at

http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.707.5956.

Berg, W., Nowak, S., Kelly, M., Vaidyanathan, S., Shoemaker, M., Chittum, A., ... & Kallakuri, C. (2016,

September). “The 2016 State Energy Efficiency Scorecard.” American Council for an Energy-Efficient

Economy.

Carvallo, J., Larsen, P., Sanstad, A., and. Goldman, C. (2016). “Load Forecasting in Electric Utility

Integrated Resource Planning.” Energy Analysis and Environmental Impacts Division, Lawrence

Berkeley National Laboratory. LBNL- 1006395. Available at: http://eta-

publications.lbl.gov/sites/default/files/lbnl-1006395.pdf.

Cook, A., “MISO Looks to Align Load Forecasting, Tx Planning,” RTO Insider, January 23, 2018,

available at: https://www.rtoinsider.com/miso-load-forecasting-transmission-planning-84614/.

Eryilmaz, D., & Sergici, S. (2016). “Integration of residential PV and its implications for current and

future residential electricity demand in the United States.” The Electricity Journal, 29(1), 41–52.

Eryilmaz, D., Smith, T. M., & Homans, F. R. (2017). “Price Responsiveness in Electricity Markets:

Implications for Demand Response in the Midwest.” Energy Journal, 38(1), 23–49

Freeing the Grid, “Freeing the Grid” (2017).

Hahn, H., Meyer-Nieberg, S., & Pickl, S. (2009). “Electric load forecasting methods: Tools for decision

making.” European Journal of Operational Research, 199(3), 902–907.

Hippert, H. S., Bunn, D. W., & Souza, R. C. (2005). “Large neural networks for electricity load

forecasting: Are they overfitted?” International Journal of Forecasting, 21(3), 425–434.

Hippert, H. S., Pedreira, C. E., & Souza, R. C. (2001). “Neural networks for short-term load forecasting:

A review and evaluation.” IEEE Transactions on power systems, 16(1), 44–55.

Hobbs, B. F. (1995). “Optimization methods for electric utility resource planning.” European Journal of

Operational Research, 83(1), 1–20.

Hyndman, R. J., & Koehler, A. B. (2006). “Another look at measures of forecast accuracy.”

International Journal of Forecasting, 22(4), 679–688.

Kyriakides E., Polycarpou M. (2007) Short Term Electric Load Forecasting: A Tutorial. In: Chen K.,

Wang L. (eds) Trends in Neural Computation. Studies in Computational Intelligence, vol 35. Springer,

Berlin, Heidelberg.

MISO, 2017. MTEP17– MISO Transmission Expansion Plan.

CRA Insights: Energy | 8

Panapakidis, I. P., & Dagoumas, A. S. (2016). “Day-ahead electricity price forecasting via the

application of artificial neural network based models.” Applied Energy, 172, 132–151.

Perez-Arriaga, I. J., and Batlle, C. (2012). “Impacts of intermittent renewables on electricity generation

system operation.” Economics of Energy & Environmental Policy 1, no. 2: 3–18.

Soares, L. J., & Medeiros, M. C. (2008). “Modeling and forecasting short-term electricity load: A

comparison of methods with an application to Brazilian data.” International Journal of Forecasting,

24(4), 630–644.

State Utility Forecasting Group. (2017) MISO Independent Load Forecast Update. Available at:

https://www.purdue.edu/discoverypark/sufg/docs/publications/MISO/MISO%20ILF%20Update%20Rep

ort.pdf, accessed on 5/7/2018.

Wooldridge, J. M. (2001). Econometric analysis of cross section and panel data. The MIT Press.

Western Area Power Administration (WAPA). 2016. MISO – Market Overview. Available at:

https://www.wapa.gov/About/the-source/Documents/2016/discussion-markets-todd-ramey-MISO.pdf,

accessed on 3/5/2018.

About CRA’s Energy Practice

Charles River Associates is a leading global consulting firm that offers strategic, economic, and financial

expertise to major corporations and other businesses around the world. CRA’s Energy Practice provides

services to a wide range of industry clients, including utilities, ISOs, RTOs, large customers, and investors.

The Energy Practice has offices in Boston, New York City, Washington, DC, Toronto, and London. Learn

more at www.crai.com/energy.

Contact

Derya Eryilmaz, PhD

Associate Principal, Energy

Boston

+1-617-425-3080

deryilmaz@crai.com

The conclusions set forth herein are based on independent research and publicly available material. The views expressed herein do not purport to reflect or represent the views of Charles River Associates or any of the organizations with which the author is affiliated. The author and Charles River Associates accept no duty of care or liability of any kind whatsoever to any party, and no responsibility for damages, if any, suffered by any party as a result of decisions made, or not made, or actions taken, or not taken, based on this paper. If you have questions or require further information regarding this issue of CRA Insights: Energy, please contact the contributor or editor at Charles River Associates. This material may be considered advertising. Detailed information about Charles River Associates, a registered trade name of CRA International, Inc., is available at www.crai.com. Copyright 2019 Charles River Associates