June 17 to 22, 2024, brought high electricity demand, driven by high temperatures, to the Pennsylvania-New Jersey-Maryland Interconnection (PJM). The Regional Transmission Organization (RTO) issued a Hot Weather Alert for June 25 and 26, warning of large loads.
Let’s dive into how accurate the RTO’s load forecasts turned out to be.
While RTOs continue to forecast system load up until the hour when power is delivered, many market participants need to make decisions further ahead. For instance, the version of the load forecast published just before the day-ahead market closes can be important for financial traders as well as physical operators deciding whether to buy and sell power in the day-ahead (DA) market or the real-time (RT) market.
Other physical participants, such as generators planning fuel purchases or large industrial consumers deciding when to reduce their demand, may need accuracy in even earlier versions of load forecasts – as much as three days before the delivery date. After all, you can't just flip a switch to turn a generator on or a warehouse cooling system off.
Regardless of which forecast version you want, Yes Energy vintages every RTO load, wind, and solar forecast as well as our proprietary TESLA load forecasts. That allows us to compare the load forecasts from the RTO and TESLA forecasts to the actual load that materialized.
Looking at the latest forecast as of the DA market close time (11 a.m. EPT), TESLA forecast’s accuracy outperformed the RTO's by 22% in terms of mean absolute percent error (MAPE) at both one-day-ahead and three-day-ahead time horizons.
Source: Yes Energy’s TESLA load forecasting
Looking at historical load records, PJM’s three-day-ahead forecast for June 21 of 152.635 GW would have made it the eighth-highest load peak in PJM history.
Source: Yes Energy’s Time Series Analysis Module in PowerSignals®
Even one day out, it was still projected to be the 18th-highest load peak.
However, the actual load came in at 147.927, which was the 28th-highest load peak.
Source: Yes Energy’s DataSignals Cloud®, our cloud database for power markets
Aside from peak load days, holidays are another source of market uncertainty due to their infrequent nature, which poses challenges for forecasting models.
One way that we account for those concerns in our forecasts is by training the TESLA model on 11 years of weather and load history which guarantees at least one instance of every holiday falling on a given day of the week.
This year, Juneteenth presented a unique challenge with only three years of history and continued changes in how many companies observe it. (It’s only been a federally observed holiday since 2021.)
TESLA load forecasts came through though. The day-ahead peak load forecast (11 a.m. on June 18) was within 415 MW (0.3%) of the actual load, better than the ISO forecast peak which was off by 2,700 MW (1.8%).
Source: Yes Energy’s Time Series Analysis Module in PowerSignals®
TESLA was also 74% better than the ISO in terms of MAPE over the whole day.
Source: Yes Energy’s TESLA load forecast
It matters which version of a load forecast you look at, and in Yes Energy, you can see any version of the load forecast (ISO, RTO, or TESLA forecasts).
Historical context also matters, and in Yes Energy you can see historical market data back to the ISO or RTO’s inception.
As we saw in June, RTOs often find it challenging to predict electricity demand during extreme weather temperatures. This is where Yes Energy’s TESLA forecasts shine.
Want to learn more about predicting unpredictable days? Talk to our team or request a demo.