Nowadays, power companies are trying to monitor energy consumption to provide demand response. Energy management and scheduling are possible through short-term load forecasting. Energy supply stability and efficiency depend on accurate forecasting, which balances demand and supply. In this paper, a novel hourly energy prediction method is introduced. A new parallel deep learning network is presented based on CNN and GRU networks. Firstly, some features are extracted from the dataset during pre-processing. Then, the CNN models extract more information from these features in two parallel paths. Afterward, Bi-GRU networks are used to observe the extracted features from previous layers in both two directions to learn long dependency patterns. The real-world data collected by Mashhad energy distribution company is used to evaluate the proposed method. The results demonstrate that the proposed method reached the lowest values containing 49.04, 34.37, and 3.81 for RMSE, MAE, and MAPE metrics in comparison with existing methods.

Keywords