Feature Engineering for Short-Term Forecast of Energy Consumption

This paper proposes a model for short-term load forecasting using conventional linear regression technique. The real electricity consumption data provided by local Estonian company Alexela AS is used. The choice of algorithm is motivated by specific design requirements imposed by Nord Pool and local transmission system operator data processing partner. Due to these requirements, most of attention was put toward feature engineering process and discovering important factors that are essential for the local energy consumption profiles. Proposed solution is validated on the real data, and shows 3.339% forecasting accuracy on a one year interval.