A Method of Load Forecasting Based on Temporal Convolutional Network

With the wide application of RNN and the high accuracy of LSTM and GRU, it is more and more popular in the field of load forecasting. However, the stability of the training is always worrying, and the phenomenon of gradient vanishing often occurs. Aiming at the problems, this paper proposes an intelligent load forecasting method which is suitable for medium and long term and short term. The method is based on temporal convolutional network Based on the residual learning idea of RESNET network model, the extended causal convolution and self-attention mechanism are used to build the intelligent load forecasting model of low-voltage substation area, which reflects the timeliness, accuracy and intelligence.