Short-Term Load Forecasting By Learning Load Characteristics Using Deep Convolutional and Recurrent Networks

Forecasting the consumed electrical load is critical for reliable operation of power systems as well as for management of planning and load storage. In this paper, short-term load forecasting is performed by extracting the characteristics of the load history using deep neural networks. Recurrent neural networks, especially improved recurrent neural networks such as LSTM and GRU, are able to retain short-term and long-term memory to extract the relationships between load values from the time series. On the other hand, convolutional neural networks are able to automatically learn features and can directly generate a vector for prediction. The proposed method is to extract the load characteristics using convolutional neural networks and then extract the load sequence information using the GRU network. The results of experiments on three data sets, Toronto, ISO-NE, and North American Utility, show a decrease in the forecasting error of the proposed method compared to other competitor methods.