General approach

Energy forecasting is about creating mathematical models to predict future energy supply and demand.

How it works

The idea behind forecasting in general is making use of what we know today to predict what might happen tomorrow. This knowledge might be based a theoretical understanding (such as calculating how much solar radiation is incident on a tilted plane) or more empirical (such as finding trends and seasonality in data). In reality, forecasting is usually done with a blend of these approaches.

For energy forecasting specifically, one of the main uncertain factors is in many cases the weather, which is why weather forecasting becomes of high importance. In other cases, such as for behaviour-driven or industrial electricity demand, the best predictor is the time series itself that is forecasted. It contains information regarding how behaviour is correlated specific times of the day, days of the week or days of the year.

Input data sources

The input data sources used in the energy forecasting models are:

  • Target time series itself
  • Weather forecasts
  • Other exogenous information

Other exogenous information typically consist of day types (as mentioned previously), but can also be production plans, planned maintenance and even historical weather measurements. Find more information about our data APIs here.

Approaches and frameworks

Any Python framework and code can be used in the Rebase Platform to develop an energy forecast models. Several of the more common frameowrks already have pre-implemented energy forecasting models. Below is a list of frameworks that have an implemented model:

ModelDeveloperModel typeCode and docs
sklearn.LinearRegressionScikit-learnLinear regressionCode, Docs
statsmodels.ARIMAStatsmodelsARIMACode, Docs
sklearn.RandomForestRegressorScikit-learnRandom forest regressionCode, Docs
sklearn.RandomForestRegressorScikit-learnRandom forest regressionCode, Docs
LightGBMMicrosoftGradient boosting decision treesCode, Docs
XGBoostDMLCGradient boosting decision treesCode, Docs
tensorflowGoogleNeural networksCode, Docs