Latest / MBA Training Data / Load forecasting: why DER signals break utility models
Transcript
- 0:00Welcome to the Leaders Insights podcast.
- 0:07Today's episode, load forecasting for utility operators integrating weather, behavioral
- 0:13and DOR signals into demand prediction pipelines.
- 0:17Everyone in your field keeps saying load forecasting is a solved problem.
- 0:20You told me before we started, that's nonsense.
- 0:23Go.
- 0:24It was solved past tense.
- 0:26For 40 years the utility playbook was, temperature goes up, air conditioners turn on, demand
- 0:32climbs.
- 0:33You regress against weather and industrial schedules, and you're within 2%.
- 0:38That world is dead and a lot of forecasting teams haven't noticed the funeral.
- 0:43Dead how?
- 0:44Weather still drives demand.
- 0:45It does.
- 0:46But now half your grid fights back.
- 0:49Rooftop solar means a hot sunny afternoon can cut demand instead of raising it.
- 0:54The panels shave the peak.
- 0:55Then a cloud rolls over a neighborhood and 40 megawatts reappears in 90 seconds.
- 1:01Your old model sees weather and says, high demand.
- 1:04Reality says, actually no.
- 1:07The sign of the relationship flipped and nobody updated the equation.
- 1:11So the fix is just adding solar data to the pipeline.
- 1:14Problem solved again.
- 1:16That's exactly the mistake everyone makes.
- 1:18They bolt DR signals, distributed energy resources, meaning the solar panels, home
- 1:24batteries, and EV chargers sitting behind the meter, onto the same regression and call
- 1:28it modern.
- 1:30What they've built is a more complicated wrong answer.
- 1:33The behavior is the hard part, not the hardware.
- 1:35Explain the behavior part.
- 1:37Take EV charging.
- 1:39Electric vehicle charging.
- 1:40A utility in California assumes everyone plugs in at 6 p.m. when they get home.
- 1:45Fine.
- 1:47Then the utility offers a midnight rate discount.
- 1:50And suddenly 10,000 cars all start charging at 12 hours, zero minutes, and one second because
- 1:55they're on timers.
- 1:57You've engineered a synchronized spike that never existed in nature.
- 2:02Your forecast, trained on last year's behavior, has no idea the rate change even happened.
- 2:07That sounds like the utility shooting itself in the foot.
- 2:11Constantly.
- 2:12The price signal changes the thing you're trying to predict.
- 2:16Economists call that endogeneity.
- 2:18The variable you're measuring reacts to your own decisions.
- 2:22Most load models pretend the world holds still.
- 2:24It doesn't.
- 2:25Where does this actually cost money?
- 2:27Give me the pain.
- 2:29Two places.
- 2:30NERC reliability reports.
- 2:32That's the North American body that finds you when the lights flicker.
- 2:36And rate cases where regulators decide what you're allowed to charge.
- 2:40Under forecast the peak and you buy emergency power at 10 times the price.
- 2:45Or worse, you shed load and end up on the news.
- 2:49Over forecast and you've built a gas plant nobody needed.
- 2:52And the regulator makes shareholders eat it.
- 2:55Either direction, a 2% error, is now a nine-figure conversation.
- 2:59The vendors say their tooling fixes this.
- 3:02You buy that?
- 3:03Partly.
- 3:04DBB Labs, DWE, a data tooling vendor, publishes numbers on how much faster teams ship pipeline
- 3:11changes with proper version control.
- 3:14And it's genuinely useful for keeping these models auditable.
- 3:18But they sell the software, so cross-check it.
- 3:21The tooling helps you fix a model quickly.
- 3:24It does not tell you the model is wrong in the first place.
- 3:27That's a human problem wearing a software costume.
- 3:30So what does the independent research say people should do differently?
- 3:34MIT Technology Review had good reporting last year on probabilistic forecasting.
- 3:39Instead of predicting one demand number, you predict a range with confidence bands.
- 3:44As the honest answer on a partly cloudy day with high EV adoption isn't 4,200 megawatts.
- 3:51It's somewhere between 3,800 and 4,600.
- 3:55And I'm nervous.
- 3:58Operators hate that.
- 3:59They want a single number to plan against.
- 4:02But a range is harder to act on.
- 4:04Isn't the single number more useful?
- 4:06A confident wrong number is the most expensive thing in the building.
- 4:10The range tells you when to keep a reserve plant warm and when to relax.
- 4:15That's the whole game.
- 4:16Knowing your own uncertainty.
- 4:18The dry truth in one line?
- 4:20We spent 40 years teaching the grid to predict the weather.
- 4:23And then we filled every house with a robot that argues with it.
- 4:26Give me the one thing a forecasting team should do Monday morning.
- 4:30Go find every rate change and demand response program your company launched in the last
- 4:3418 months.
- 4:36Then check whether your model has any variable that reacts to them.
- 4:40If it doesn't, and it probably doesn't, your forecast isn't modeling the grid.
- 4:45It's modeling a grid that stopped existing years ago.
- 4:48Uncomfortable and useful.
- 4:50Thanks for coming in.
- 4:51Sources for today's episode.
- 4:53The new stack.
- 4:54MIT Technology Review.
- 4:56MIT Sloan Management Review.
- 4:58DBT Labs.
- 4:59Vendor.
- 5:00Data tooling.
- 5:01Done.
- 5:02Want to know where you actually stand?
- 5:04Take the CDO self-assessment at MBA-training.com.