Summer Short-Term Electric Load Forecasting Based on Hierarchical Model

The operation safety of the power grid during the summer peak period is affected by the accuracy of short-term summer load forecasting directly. Firstly, we point out that temperature is essential to determine the summer load variation through the analysis. Then, we propose a hierarchical model based on temporal convolutional network (TCN) and long short-term memory network (LSTM) to solve the problem of forecast accuracy decline caused by the non-linearity between temperature and load. The model emphasizes the influence of temperature on load variation structurally and reflects the nonlinear correspondence between temperature and load accurately. By actual data analysis, the accumulated days of temperature and the input sequence length of the model are determined. The example shows that the established hierarchical model has higher forecast accuracy than the conventional single-layer neural network model.