A Novel Short-term Load Forecasting Model by TCN-LSTM Structure with Attention Mechanism

The main challenges of short-term load forecasting are that loads are highly uncertain and time-dependent at shortterm scales, and short-term loads are affected by many factors, including temperature, humidity, and electricity prices, and these characteristics hinder the accuracy of load forecasting. To further improve load forecasting accuracy, this paper proposes a TCN-LSTM model with attention mechanism for short-term load forecasting. The temporal convolutional network (TCN) constructs very long effective histories by dilated causal convolution to establish the long-term dependencies of load data. The proposed attention mechanism is coupled with the long short-term memory network (LSTM) to collect crucial input sequence details and reduce uncertainty-related distraction. A case study of load forecasting based on a public dataset is investigated to illustrate the feasibility and benefits of the model. According to simulation findings, the suggested TCN-LSTM model with attention mechanism increases the load forecasting's accuracy.