top of page

FAQ

What data is used as inputs?

We have over 1,300 data series from which to select as inputs from data published by the ISO including scheduled generation, transmission outages, regional load, and reserve capacity. We also include commodities futures data from ICE, as well as more robust weather data from several vendors. Some of our most accurate forecasts use load forecasts, weather forecasts (temperature, wind, humidity, solar irradiance), scheduled and forecasted generation, previous days imports and exports, previous days congestion, planned/forced outages on plants and lines, time variables, price of natural gas, heat rates, ancillary services procurement, tie line flows and liquidity and pricing data from ICE

How accurate are the models?

It typically depends on the target timeseries, but a good rule of thumb is the following: • Load forecasts: 2% MAPE • Next-day DALMP forecasts: $7 MAE • Next-day RTLMP forecasts: $12 MAE • Other generation forecasts: 5% MAPE Some forecasts can beat the above benchmarks by as much as 50% (our next-day NYISO load forecast MAPE is

Will the models improve over time?

Yes – we expect every target to be more accurate over time. We believe this for three reasons: 1) Models are automatically retrained weekly using more data with the same parameters 2) Our cloud-based architecture allows us to constantly train new models in the cloud using a slightly different dataset (for example, by adding in outages for a key backbone), and comparing those results against the current top performer 3) Forecasting power markets is our raison d’etre, and we’re constantly tweaking our engine and rolling out improvements to all forecasts

I want to use the forecasts for automatic battery dispatch – do you have an API?

Yes we have an easy-to-use API that can provide our most recent forecasts, all of which can be run at a frequency specified by the client. See our docs under "Resources" for more information.

Why would we use Enertel instead of doing it ourselves?

There are some clients with a high technical aptitude for price forecasting, and we certainly understand that companies wish to perform some modelling in house. We don’t believe they’re mutually exclusive, given the importance of forecasting for managing assets. With that said, there are a couple advantages to also procuring forecasts from a vendor: - Yes Energy Access: Your traders are already in Yes Energy on a regular basis, so there’s no change to their day-to-day process - Economies of Scale: Forecasting vendors consistently invest their time into improving forecasts because small incremental improvements have a compound effect over many models, vs. in-house models that may be left for weeks without retraining or improvements - Attrition: In-house models can sometimes rely on a few people within the company who may leave the company. A forecasting vendor provides reliability and long-term accuracy gains

How do you evaluate accuracy?

We rigorously track the accuracy of every forecast generated by the Enertel platform to make decisions on training new models and promoting model architectures. Typically, clients care about the accuracy of our ‘best’ forecasts for a given timeseries, so we provide an easy-to-use dashboard that lets them evaluate the forecast accuracy for every time a given model was run and the forecasts it generated.

How does a trial work?

We invite companies to trial our price forecasts for one or two price nodes for a duration of 4 weeks, after which we'll send you an accuracy report that will measure the MAE and RMSE of the forecasts you recieved from Enertel vs. a benchmark of your choosing, which is usually the seasonal naive model (using the previous day's clear).

bottom of page