Load Forecasting in Distribution Grids with High Renewable Energy Penetration for Predictive Energy Management Systems

In this paper we present a new approach for load forecasting in distribution grids with high renewable energy penetration. The method is based on multiple neural networks and the application focuses on predictive energy management systems which use a model predictive control (MPC) approach. These control algorithms need predictions of demand profiles from 15 minutes up to several days. The short-term forecast values are more important than the long-term prediction values beyond six or 24 hours. Thus, the new method takes instantaneous measurements into account in order to provide a high accuracy for the first prediction values. In addition, weather forecast data is included as input variables of the neural networks for the purpose of mapping the influence of renewable energy generation on the load profiles. With this approach, the method improves the Root-Mean-Squared Error up to 80 % compared to a reference model based on a weekly persistence.