Forecasting vertical grid load using machine learning algorithms

The vertical grid load is one of the main input variables to determine the load flow on the transmission grid in system operation centers. As it depends on a variety of input variables, a method is needed that allows a good forecast despite incomplete and uncertain information about underlying grids, loads and feed-ins. With each underlying distribution grid having its own characteristics, the authors propose to model the vertical grid load individually for each interconnection point between the transmission and distribution grid. The following machine learning algorithms are tested and compared to develop a reliable forecast model: Linear regression, Lasso and Ridge regression, generalized additive regression, Random Forests and XGBoost. The case study shows the results from a German transmission system operator for a forecast horizon of one day ahead.