A CNN Based Model for Short-Term Load Forecasting: A Real Case Study on the Romanian Power System
Andrei M. Tudose, Dorian O. Sidea, Irina I. Picioroagã, Valentin A. Boicea, Constatin Bulac
Abstract
Short-term load forecasting (STLF) is of crucial importance in power systems operation, for it represents the basis of fundamental processes, such as power dispatch, storage scheduling or equipment maintenance planning. Considering the high degree of uncertainties that describes the load time series, advanced techniques are needed to be implemented for an accurate electrical load forecasting. In this paper, a convolutional neural network (CNN) based model is proposed, in order to solve the day-ahead load forecasting problem. The evaluation and validation of the developed model are carried out based on real public dataset from the Romanian power system. By comparing our model output with the Romanian TSO's forecasting results, using the Mean Absolute Percentage Error (MAPE) as evaluation index, the proposed methodology shows great potential in accurate forecasting and high generalization capability.