An ensemble framework for short-term load forecasting based on parallel CNN and GRU with improved ResNet

Accurate and efficient load forecasting is of great significance for stable operation and scheduling of modern power systems. However, load data are usually nonlinear and non-stationary that make accurate forecasting difficult. Although some serial hybrid models effectively extracted the spatiotemporal features of load data, the extraction of features in order are not efficient due to the loss of some important features. To address these issues, this paper proposes a novel ensemble framework for short-term load forecasting based on parallel convolutional neural network (CNN) and gated recurrent unit (GRU) with improved ResNet (iResNet). Firstly, the original data is preprocessed to reconstruct the electrical features. Secondly, the spatial and temporal features are extracted by the CNN and GRU, respectively. Then, both of the extracted features are dynamically combined with attention mechanism. Finally, the iResNet is utilized to efficiently forecast the power load. Compared with the GRU and serial CNN-GRU-iResNet models, the mean average percentage error (MAPE) of the proposed model decreases by 40% and 30%. The proposed model even outperforms the parallel CNN-LSTM-iResNet model by 12%, in terms of the MAPE.