Temporal convolutional neural network with self-attention for shortterm load forecasting

Abstract:
The operation of the power system becomes more flexible and uncertain as the electrical market reform continues to develop, requiring smart grid data processing efficiency, intelligence analysis, and processing technology. Temporal Convolutional Networks (TCN) are a specific design for forecasting problems that offers advantages over recurrent networks. In this paper, a hybrid method is proposed temporal convolutional neural network with self-attention (TCN-SA) for short-term load forecasting that can make more accurate predictions by extracting multi-scale characteristics than other existing methods. We apply it to an open dataset and comparisons with existing models show that the proposed model provides accurate load forecasting results.