Automation of probabilistic time series forecasting: a new multimodel framework with optimal selection of explanatory variables applied to electricity consumption series.

The R&D project carried out with Energisa aims to create a new multimodel automation framework for the simulation (probabilistic forecasting) of time series with optimal selection of explanatory variables. The framework allows for automated and systematized comparison of various models through the execution of backtests and other classic criteria. So far, we have considered in the tool 1) boosting, 2) random forest, 3) SARIMA, 4) state space, 5) ETS, 6) GAS. New models will be included throughout the project.

Below are some screenshots of the tool: