A Deep Learning Approach for Load Demand Forecasting of Power Systems

Power load forecasting has been an important problem in which machine learning and neural networks have been largely studied, providing effective solutions. Nevertheless, there are still discrepancies among generated load forecasts and actual demands in real life environments, where big amounts of data are continuously created. In this paper we propose an approach for evaluating existing short term load forecasting and predicting such discrepancies. Our approach is based on generating and using deep convolutional - recurrent neural networks that are able to process the data, either as time series, or as 2-D information. An experimental study is provided that illustrates the ability of this approach to improve the accuracy of forecasting in a real life scenario.