Ensemble based forecasts of electricity demand

In this notebook, we use an ensemble comprised of a linear model, a prophet model, a catboost and a theta model to forecast electricity demand. The theta model is a purely univariate method i.e. does not take any inputs besides the historic series itself. The linear model on the other hand uses 30 lags and calendar features such as time of day and day of year to make the forecast. Finally, both the catboost and prophet models make use of both historic data and also exogenous variables such as weather conditions and forecasts. Finally, the optimal parameters for the relevant models were found by a detailed grid search.