Short-term Load Forecasting using Convolutional Neural Network and Long Short-term Memory

Nowadays, electricity is one of the most basic needs of human societies, such that almost all industrial operations and a large part of social, economic, and agricultural activities rely on this energy; therefore, the quality and stability of electrical power are important. In this study, we seek to forecast the changes in short-term load consumption concerning the factors affecting electric load, mainly weather changes, as well as the daily and weekly consumption fluctuations that have complex nonlinear relationships. It should be noted that the prediction of short-term changes in load consumption is an essential and critical factor in power distribution systems. The proposed method is a hybrid neural network based on deep learning, which trains with energy consumption data and real weather changes. This neural network is developed by combining CNN and LSTM architectures and optimized for configuration parameters such as the number of network layers and neurons as well as filter size. CNN and LSTM architectures have been utilized to extract existing patterns in data and to predict time series, respectively. The proposed approach predicts the pattern of future consumption by forecasting weather in the next hours and the pattern of electric load consumption in the past hours. Tensorflow framework is employed to implement the proposed approach, and the results are compared with similar state-of-the-art methods. The evaluation results show that the prediction accuracy is improved compared to the best available methods.