Medium-to-long Term Electricity Consumption Forecasting Using Deep Hybrid Neural Networks

Abstract

This paper presents a deep hybrid neural network method to predict the regional electricity consumption in the future. The electricity consumption forecasting problem is usually modelled as the time-series problem using the recurrent neural network. Considering the advantage of convolutional neural network on the feature extraction, it is potential to enhance the performance of long short-term memory neural network on improving the training accuracy. It is the regional electricity consumption that is correlated to its economic growth. To address such implicated issue, the proposed approach combines the convolutional and recurrent neural networks to investigate the impact of gross domestic product on the electricity consumption. Furthermore, it is based on the relationship between the electricity consumption and gross domestic product to establish the forecasting model. The available data mainly includes the annual energy consumption data and gross domestic product from the different social industries. The numerical results imply that the deep hybrid neural network model generally correspond to lower relative error when compared with the autoregressive integrated moving average model, and solves the problem of overabundance training database in deep neural work.