Latest Trends in Estimated Load Impacts of COVID-19 Mitigation Policies

As shelter-in-place policies are enacted to help reduce the spread of coronavirus (COVID-19), the Itron Forecasting group is leveraging publicly available hourly load data for most North American Independent System Operators (ISOs) to build a picture of the load impacts by region.

To assess the load impact of COVID-19 mitigation strategies, actual loads since March 22, 2020, when many of the shelter-in-place policies began, were compared to baseline loads without COVID-19 policy impacts. The Forecasting team recently conducted a special webinar on this topic - we encourage you to view the recording if you missed it.

Although shelter-in-place policies vary across the U.S., in general, the policies have led to school closures and reduced operations or closures for many non-essential businesses. Many other businesses have a large portion of their employees working remotely. The net effect is a shift of weekday loads from the non-residential sector to the residential sector. In some control regions, that is leading to an evolution of the system load shape toward a residential load pattern. The net impact on system loads depends on the original mix of residential and nonresidential loads, prevailing weather conditions and the specific shelter-in-place policy.

Baseline Loads: Estimates of the load impacts resulting from the shelter-in-place policies are developed by comparing Actual loads to Baseline loads. For this initial set of impact estimates, baseline loads are computed as the average hourly load by day-of-the-week over the time period March to April 2017, 2018 and 2019. It is important to note that the baseline loads are not adjusted for prevailing temperature and solar conditions. As a result, the hot spell that rolled through ERCOT the week of March 22, 2020 drove actual loads above the baseline, reducing the estimated shelter-in-place policy impacts.

Estimated Load Impacts: The estimated net impact of the shelter-in-place policies for the time period March 22, 2020 to April 6, 2020 by ISO and for the aggregate ISO load is presented below. The values in this table represent the percent difference between Actual loads and Baseline loads. The percentage difference is computed by day (d) and hour (h) as:

Again, these estimated impacts do not control for differences between the weather that prevailed over the comparison period and the period over which the baseline load is computed.

Estimated Load Impact by ISO Control Region and Time-of-Use Period: March 22, 2020 to April 6, 2020:


Total ISO Impacts: Total ISO load is computed as the sum of the hourly loads across the eight (8) ISOs for which hourly load data are publicly available. The estimated maximum impact in one hour is a reduction of -10.7%. On average, hourly loads at the aggregate ISO level are down by about -4.7% across all hours. The morning hour loads (i.e. 6 a.m. to Noon) have the biggest average load reduction of -6.1%. A comparison of Actual versus Baseline loads for the week of March 29, 2020 are shown in the following figure.

New York ISO (NYISO): The NYISO control region spans New York state, including New York City which is the epicenter of the COVID-19 impact in the United States. Like California, New York began shelter-in-place policies around mid-March. On average, the policies have reduced NYISO hourly loads by about -8.6%. The mid-day hours, 6 a.m. to 6 p.m., show the biggest load reductions with average loads down by about -11%.

California ISO (CAISO): The CAISO control region spans California. On average, the shelter-in-place policies that were enacted in mid-March have led to an average hourly load reduction of about -8.5%. The afternoon hours show the biggest load reduction of about -12%. The impact in the morning hours is also significant, at roughly -8.4%, but this region has had significant growth in solar PV installations year over year which muddies a straight comparison of baseline to actual loads.

PJM Interconnection (PJM): The PJM control region spans a large geographical footprint from the eastern seaboard across to Chicago. Certain areas within the PJM control region have been hard hit by the COVID-19 virus including parts of New Jersey, Detroit and Chicago. There has also been a mix of shelter-in-place policies for each area. The net load impact of the various shelter-in-place policies is estimated to be an average load reduction of -6.8%.

Midcontinent Independent System Operator (MISO): The MISO control region also spans a large geographical footprint from Canada down to Louisiana. The impact of COVID-19 and shelter-in-place policies is mixed across this control region. Cities like New Orleans, which is a growing epicenter, have enacted severe shelter-in-place policies while other areas have little to no policies in place. The average non-weather adjusted net load impact is estimated to be a reduction in loads of about -6.0%.

ISO New England (ISO NE): The ISO NE control region spans the New England states. Boston and other government jurisdictions have enacted a mix of shelter-in-place policies. The average non-weather adjusted net load impact is estimated to be a reduction in loads of about -4.7%.

Independent Electricity System Operator (IESO): The IESO control region spans the province of Ontario, with Toronto as the largest city. Like many areas with large international cities, the province of Ontario has put in place shelter-in-place policies to help mitigate the spread of the COVID-19 virus. The average non-weather adjusted net load impact is estimated to be a reduction in loads of about -2.4%.

The Electric Reliability Council of Texas (ERCOT): ERCOT operates the electric grid and manages the deregulated market for 75% of Texas. During the week of March 22, 2020, there was a significant heat wave that rolled through Texas. With weather-driven actual loads significantly above baseline loads, the average estimated impact of COVID-19 mitigation polices is an increase in average loads of 2.8%. Removing the week of March 22 from the analysis leads to a non-weather adjusted estimated reduction in average hourly loads of -1.2%.

Alberta Electric System Operator (AESO): The AESO is responsible for the operation of the Alberta Interconnected Electric System. In this case, the non-weather adjusted baseline leads to a positive increase in loads over this period. Additional work on the baseline model is required to estimate the impact the mitigation strategies are having on AESO’s loads.

Model Recommendation: If you are responsible for developing operational load forecasts in a control region impacted by COVID-19 mitigation strategies, the recommendation is to add an Endshift variable to the set of explanatory variables. Specifically, the Endshift variable should be designed based on the forecast modeler’s best estimate of when the mitigation strategies started and when the strategies are expected to be removed. The Endshift variable can contain a phased in period. For example, if mitigation policies were set to begin on March 15, the value of the Endshift variable could be something like: 0.0 on March 14, 0.2 on March 15, 0.4 on March 16, 0.6 on March 17, 0.8 on March 18 and 1.0 on March 19. From March 20, the value of the Endshift variable would be 1.0.

The recommendation is to add the Endshift variable to each forecast equation. The effect would be to shift the intercept down(up) as the impact of COVID-19 mitigation strategies hit. As additional weeks of data are gathered, it is recommended that interactions between the Endshift variable, Temperature variables and Day-of-the-Week binary variables be added to allow the load impacts of the mitigation policies to vary with weather conditions and day-of-the-week. For areas with significant solar PV generation, it is recommended that interactions between the Endshift variable and the behind-the-meter solar PV generation variables be evaluated since the net load impact of solar PV could be dampened with people working at home.


Upcoming Annual Energy Forecasting Webinars

Itron continues to monitor and navigate through the coronavirus pandemic, keeping the safety and wellbeing of our employees, customers and partners first and foremost in our decisions. As a result of the near-term dynamic environment, we have decided to cancel this spring’s 18th Annual Energy Forecasting Meeting and Workshop. We look forward to seeing you next year in New Orleans (rescheduled to April 20-23, 2021).

While social distancing and shelter-in-place mandates mean that we can’t meet in person, Itron will still be hosting several webinars throughout the originally planned dates – April 22 through April 24 – to keep you informed. We hope that you will join us!

Participation for these webinars is free, like our Brown Bags, but prior registration is required. Each seminar will last approximately one hour, allowing 45 minutes for the presentation and 15 minutes for questions. If you can't attend or if you miss one, each session will be recorded and a link to the recording will be sent to you automatically after the event.

Wednesday, April 22

  • 8 a.m. - EIA Residential and Commercial Updates - Erin Boedecker, Energy Information Administration
  • Noon - EIA Electric Vehicle Forecast - Michael Dwyer, Energy Information Administration

Thursday, April 23

  • 8 a.m. - Economic Outlook - Ryan Sweet, Moody’s Analytics
  • Noon - Addressing Co2 Emissions Standards and Greenhouse Gas Reduction Targets - Arthur Maniaci, New York Independent System Operator

Friday, April 24

  • 8 a.m. - Forecasting Challenges with COVID-19 - Itron’s Forecasting Team

For more information on these webinars and other forecasting events, please go to our forecasting workshop page at http://www.itron.com/forecastingworkshops.


NYISO Climate Change Study: Phase 1

Serving nearly 20 million people and having a peak demand of roughly 32,000 megawatts, the New York Independent System Operator (NYISO) manages the wholesale power market and daily operations for New York State’s electrical transmission grid. For more than a decade, NYISO has successfully utilized Itron’s software and services for long-term energy and demand forecasting in the area of transmission planning, as well as short-term operational forecasting to support NYISO’s real-time and day-ahead market operations.

As part of an ongoing project with NYISO, Itron completed the New York ISO Climate Change Impact Study – Phase 1 report in December 2019. The report identifies historical weather trends across more than 20 weather stations in New York State. The ultimate goal of the project was to develop long-term energy, peak and hourly load forecasts that reflect the potential continuation of such weather trends during the next 30 years, as well as the effects of such trends for electricity consumption and the requirements for the transmission grid. Complicating factors include continued growth in behind-the-meter solar generation, increasing proliferation of electric vehicles and state policy to address climate change through electrification.

Contact Itron at forecasting@itron.com for inquiries about the report.


How to Account for COVID-19 in Your Load Forecast

There are unprecedented load forecasting effects due to the COVID-19 pandemic. On March 26, Itron’s forecasting team will discuss how to best model the sudden load shifts due to the various COVID-19 mandates.

Register for this special webinar today.


Itron’s New Forecasting YouTube Library

Itron’s Forecasting group has conducted webinars on a variety of forecasting and load research-based topics for many years and continues to host new webinars every quarter. Past webinars have been recorded and are available in a YouTube library found here.

To watch current recordings, you must register on the forecasting workshop page for one of the upcoming free webinars at http://www.itron.com/forecastingworkshops. After you register, a link will be sent to you.

Our first free brown bag of 2020 is just around the corner on Tuesday, March 31, and is entitled “A Method for Combining Load Forecasts.” This brown bag will present a high-level review of the econometric literature on combining multiple forecasts focusing on the alternative combining frameworks. The session will conclude with alternative recommendations for how the combining frameworks can be extended to meet the operational load forecasting problem.

Participation is free, but prior registration is required. Each seminar lasts approximately one hour, allowing 45 minutes for the presentation and 15 minutes for questions. Seminars start at noon Pacific-time.

If you can’t attend a seminar or missed one, don’t worry! Your registration ensures that a link to the recording will be sent to you automatically approximately one week after the seminar date.


Annual Energy Forecasting Survey

Itron’s forecasting group just launched their Annual Energy Survey. As we have done in previous years, we collect information to assess industry growth expectations and forecast accuracy for electricity and natural gas. We continue to focus on forecasts of customer and sales growth, weather-normalized growth and forecast accuracy. All survey participants will receive a summary report. Preliminary results will be presented at the annual Energy Forecasting Meeting in April and final results will be shared during our brown bag seminar on Sept. 15. Utility specific data will not be disclosed.

To sign up for the meeting, brown bags or other workshops, go to www.itron.com/forecastingworkshops.

Surveys like this benefit the entire energy community by supplying valuable knowledge. If you would like to participate, contact Paige Schaefer at paige.schaefer@itron.com.


Timing is Everything

When evaluating peak loads, load forecasters commonly focus on the severity of peak producing weather, considering meteorological factors such as temperature, humidity and wind speed. The peak loads are then weather adjusted to represent what the peak load would have been had the peak producing weather been normal.

However, the recent proliferation of AMI data facilitates deeper analysis. In particular, this data supports the decomposition of system peak loads into class-level and new technology (e.g. solar PV, EV and battery storage). Decomposing the peak load unveils the fact that while the severity of the peak producing weather is impactful, the time at which the peak producing weather occurs is also important, particularly in areas where the base load shape has a profound seasonal pattern.

In this blog, we will discuss two examples of the impact of timing on the winter and summer peaks, respectively.

Case 1: The Extreme Winter Peak
Case 1 considers a utility located in the heart of the Canadian prairies. Here, the winter low temperatures are frigid, approaching -40 degrees F. As most reasonable people would guess, this is a winter peaking utility. However, because the majority of customers in this area have natural gas – while the electric space heating load (driven by furnace fans) contributes to the winter peak – it is actually not the primary driving factor. Rather, the combination of business class, residential lighting and residential furnace fan loads contribute to drive the winter peak, which typically occurs just after 5 p.m.

Located far north from the equator, this area experiences significant shifts in hours of light.

  • On Dec. 21, the sun set at 4:56 p.m. local time
  • On Jan. 21, the sun set at 5:32 p.m. local time
  • On Feb. 21, the sun will set at 6:25 p.m. local time

The sunset time oscillation creates a narrow window in mid-December during which the base load is particularly high at 5 p.m. In addition, holiday lights also provide additional lift during this period. Therefore, a -40-degree F day which hits in this window drives a much stronger peak than it would if it occurs in February or March.

Case 2: The Extreme Summer Peak
Case 2 considers a utility located in the Northwest of the United States. Here, the summer temperatures approach 105 degrees F, and as one might expect, this is a summer peaking utility. However, while the air conditioning load is high, the combination of strong air conditioning and irrigation loads drives the peak.

Irrigation loads have a profound seasonal pattern, reaching peak levels in late June and early July. This produces a narrow window during which a hot day can produce an extremely strong peak. While the extreme, hot weather tends to hit in late July and August, it can occur earlier, coincident with the irrigation season peak.

Therefore, a 105-degree F day which hits in this window drives a much stronger peak than it would if it occurs in late July or August.

Quantifying Sensitivity
Both of these cases lend themselves toward bottom-up hourly approaches to peak forecasting. AMI data supports the disaggregation of system load data into the relevant components.

Weather simulations support the quantification of peak load sensitivities to both peak producing weather severity and timing. As the above cases demonstrate, both of these factors prove influential in determining both the 50:50 and 90:10 peak forecasts.


Forecasting is Hard – Especially for the Future

Predicting behavior in the future is no easy task. Yet, we energy forecasters do this every day. The complexity of the problem should not be understated. Behind all of the software and the complicated models, we are attempting to predict how humans will act today, tomorrow and even 20 years from now.

In the context of short-term operational forecasting, the key unknowns are weather and solar conditions. Here, we depend upon third-party weather vendors to predict accurately various weather concepts, including dry bulb temperature, dewpoint temperature, cloud-cover and solar irradiance.

When we build statistical models (i.e., regressions, neural networks, etc.), we are estimating the relationship between load and a variety of other variables. In so doing, we are implicitly assuming that the relationship captured by the model (via the coefficients on the variables) in the past will persist into the future. The variables include many factors that we know with certainty: day-of-week, month and holidays. We know without question whether tomorrow is Thursday or Friday, or if tomorrow is Martin Luther King Jr. Day or New Year’s Day. We know with much less certainty whether tomorrow will be hot or cold, or if the sun will be obscured by clouds. Further, the accuracy with which the weather vendors can predict these concepts degrades as we go further into the forecast horizon. It is easier to predict tomorrow’s temperature than it is to predict the temperature a week into the future. None of this is especially surprising or revelatory, but these ideas are often assumed, rather than explicated.

In the context of medium- and long-term models, which extend 1 to 50 years into the future, we must depend even further upon others, including economic vendors (to provide household and GDP forecasts), the U.S. EIA (Energy Information Administration – to provide saturation and efficiency trends of major end uses) and utility staff (to provide estimates of demand-side management – DSM). That is most certainly a lot of moving parts, which are outside of our control, unless of course, we feel empowered and motivated to change those values ourselves.

Let’s think briefly about economic forecasts. The U.S. Bureau of Economic Analysis (BEA) produces the historical estimates of Gross Domestic Product (GDP). It is well-known that U.S. GDP values are frequently and substantially re-stated. Q3 GDP values could potentially be revised from 1.9% annual growth to 2.1%. In this example, that is a 10.5% change calculated as (2.1/1.9) – 1! Without putting too fine a point on it: a 10.5% adjustment is a lot. There are three important points here:

  1. This is merely the historical value. We are not even looking at the forecast yet.
  2. We are focusing on U.S. GDP. If this national value is updated so substantially, how confident can we be in the Gross State Product (GSP) or Gross Metro Product (GMP), which are at dramatically lower levels of aggregation?
  3. The official numbers are at a quarterly frequency. In many cases, we utilize monthly values, which have been interpolated (via some mathematical approach) from the quarterly values.

To be fair, these numbers are based on surveys, statistical methods and various data collected by the government. The GDP numbers do not fall out of the sky and magically appear on the desks of government employees. Indeed, humans (with the help of computers) report and calculate these statistics. There is much space here for error. It is not as if there is some kind of metering device that collects all the data on every transaction in our economy and transmits it to the federal government for quick and easy reporting – that is not how any of this works.

Now, let’s think about forecasting this constantly moving target. If we have little confidence in the historical GDP value from last quarter or even the prior quarter, how confident can we be in the forecast for the coming year? By way of analogy, let’s imagine we are generating load forecasts every five minutes, based on updated weather conditions and the most recently reported historical five-minute load data. If the deltas of the most recent observations were bouncing around by a factor of 10%, it would come as exactly no surprise to anybody if our near-term load forecast would be (how shall I put this?) bad.

We can make similar arguments about the saturation and efficiency drivers. The long-term weather forecast has its own set of issues: it is typically based on some measure of ‘normal’ weather, which can be calculated variously. Maybe we use a 10-year normal or a 20-year normal or a trended-normal. Again, there is much space for interpretation and for error.

The point of this is not to denigrate the economic vendors or the EIA, but rather to bring a few issues into the daylight for evaluation. The fact that there are GDP numbers and saturation/efficiency drivers at all is a big accomplishment. We also do not let a lack of data or a lack of confidence in the data stop us from generating forecasts. We must do the best we can with the tools we have available to us!


Energy Forecasting 101 Workshop

New to energy forecasting? Itron can provide you with an introduction to the use of regression for forecasting applications, which is designed for analysts and managers who are new to the forecasting area and do not have a background in statistical analysis.

Using hands-on examples, workshop participants will build sales/load forecasting models using linear regression and exponential smoothing models. At the end of this three-day session, attendees will have a strong understanding of basic regression theory, how to apply it to energy forecasting applications, and how to use the model statistics to develop accurate forecasting models. No background in statistics is required, although it will be useful, as mathematical concepts will be taught.

Itron is always at the forefront of the energy forecasting field. We help our customers understand the complexities of forecasting. Take advantage of Itron’s expertise and experience to help you improve your forecasts. For over 20 years, Itron has trained hundreds of energy forecasters. See why our blend of lecture on real-world examples, demonstration of modeling techniques and hands-on exercises executed by attendees is a verified recipe for successful learning.

There are still a few spaces left for this workshop on Feb. 24 through Feb. 26 in Washington, DC. Register today at http://www.cvent.com/d/shqpj6.

If this workshop doesn’t fit your needs, you may be interested in one of the others or one of the upcoming forecasting meetings. See a list of them all at www.itron.com/forecastingworkshops.


Developing a Combined Operational Load Forecast

With the deep penetration of non-grid connected renewable generation and storage, electric vehicle charging, smart load control and time-of-use rates now come with greater load volatility. This in turn leads to eroding operational load forecast performance. To improve system operator’s confidence with the load forecasting process, there has been a movement toward developing and presenting an ensemble of load forecasts, which could include forecasts designed to handle the impact of rooftop solar PV and electric vehicle charging, forecasts that incorporate the impact of TOU pricing and smart load control, and load forecasts produced under alternative weather forecasts.

If the alternative load forecasts are clustered closely around each other then system operations may have greater confidence in the system conditions predicted by the ensemble. On the other hand, a forecast ensemble with a wide range could raise doubts about the forecasted system conditions, leading to system operators taking actions to hedge against the worst-case scenario. In effect, the forecast ensemble quantifies the plausible range of loads that are given uncertainty around future meteorological conditions such as temperatures, wind and solar conditions, and uncertainty around price sensitive loads and load control actions.

Within this new world of ensemble forecasting, there remains the reality that most downstream applications (e.g. transmission and distribution energy management systems and market models) require as an input only one load forecast. This means the load forecasting process needs a way of combing the alternative forecasts into a single “optimal” forecast that is then used for downstream processing.

Much of the academic literature on combining forecasts starts with the seminal paper “The Combination of Forecasts”, by J. M. Bates and C. W. J. Granger (source Operational Research Society, Vol. 20, No. 4 (Dec. 1969), pp.451-469). Bates and Granger start their analysis by observing that a combined forecast formed by taking a simple average of two alternative forecasts outperforms the two alternative forecasts. They then asked, if a 50/50 weighting of two independently generated forecasts can improve the overall forecast performance, is it possible to define a method for computing an “optimal” weighting scheme that leads to a combined forecast with the smallest mean squared forecast error?

I have drafted a white paper that provides a high-level review of some of the academic literature on combining forecasts and put forth a recommendation for how to develop an optimal forecast given an ensemble of alternative load forecasts. I am looking for comments on the white paper draft. If you are interested in providing your comments, please send an email to frank.monforte@itron.com.

The final white paper should be available later in 2020.


The Story is in the Residuals

In any field that involves heaps of data and information, details are everything. Load forecasting is no exception. If you’ve spent any amount of time building load forecast models, trying to improve existing ones, conducting out-of-sample tests, etc., then you know these kinds of processes and assessments require paying close attention to the details. “Does that coefficient make sense? What happened in October that made the residual so big? Why does my forecast for Tuesday look a little funky? What is my model missing?” These are some of the questions I find myself asking (or being asked) a lot, and invariably, it forces you to get your hands dirty.

One could argue that the art of building good models is in the details, or as I like to sometimes call them, the residuals. Sure, in-sample fit statistics are useful, but they don’t tell the whole story. Oftentimes overlooked, the residuals can tell a powerful story to those modelers willing to listen. They can reveal outliers and patterns or trends in the data that otherwise might not be easily identified by just looking at the data. And they can give the modeler a sense for what is and isn’t working in their model.

To view a residual chart in MetrixND, you simply go to the Err tab of the model window (press the eyeglasses button) and press the button that looks like a little residual chart on the toolbar. Whether you’re building a model from scratch or polishing off an existing one, this step is a must. What I like to do is take it a step further and plot the residuals against a key variable in my model and see if I’m capturing the right pattern or relationship in my data.

For example, whenever I get asked the question, “Why can’t we just drop average temperature into our linear regression model?” I always show a scatterplot like the one below that illustrates the nonlinear relationship between loads and temperatures.

A simple regression line isn’t going to capture this nonlinearity, but I think this point is really driven home when we put average temperature on the right-hand side of our linear regression equation and then plot the residuals against temperatures.

The pattern seen in this scatterplot tells us that our model is mis-specified. The horizontal line at zero is essentially our regression line, which is showing that we’re underpredicting loads at low and high temperatures and overpredicting at mid-range temperatures. If we specify our model correctly, then we would hope to see our residuals reduced to white noise with no discernible pattern (i.e., most of the data hovering around the zero line). Fortunately, when it comes to temperature, we can leverage a polynomial functional form or heating and cooling degree variables to accomplish this.

As the saying goes, “the devil is in the details.” Or as I like to say, “the story is in the residuals.” Perhaps it’s not as catchy, but I think it rings true when it comes to building good load forecast models.
Happy Holidays everyone! See you in the new year!


2020 Forecasting Meetings – Registration Now Open!

The 18th annual Energy Forecasting/EFG Meeting will be in New Orleans, Louisiana on April 22 – April 24 with optional one-day training workshops on April 21. Forecasters continue to tell Itron that there is no greater value than meeting with their peers. Itron understands the importance of these interactions. This meeting is an excellent forum to exchange ideas and hear more about modeling concepts, data development, efficiency trends and related issues from industry speakers. Your participation is invaluable. The agenda is already filing up, but we are still accepting abstracts. Submit something today! Review the current agenda, see who has already registered and sign up at http://www.cvent.com/d/mhqxg4.

In addition, the 14th ISO/RTO/TSO Forecasting Summit will gather in Atlanta, Georgia on May 5 – 7. The summit provides a forum that brings energy forecasters together to address the unique forecasting challenges faced by independent system operators around the world. Over the past 13 years of meetings, summit discussions have ranged from sub-hourly forecasting models and techniques that support generation scheduling and dispatching to long-term forecasting supporting capacity planning. Participation and attendance at this event are limited to ISO/RTO/TSO representatives. Register at http://www.cvent.com/d/khqdfn.

Don’t miss out on these networking opportunities to discuss real world issues and practical solutions. We look forward to seeing you.


DON'T MISS A POST!
I agree to have my personal information transfered to AWeber ( more information )
Opt in to receive notifications when a blog post is published. Don't miss the thought leadership, insight and news from Itron.
We hate spam. Your email address will not be sold or shared with anyone else.