Accurate load forecasting is a powerful support for power system planning and design, and an important guarantee for the safe and economic operation of the power grid. In practical application, there are some problems, such as data acquisition equipment failure and system emergency, which lead to inaccurate data and affect the short-term load forecasting results. In this paper, WT-LSTM (wavelet transform long short term memory) is proposed, which uses the time-frequency characteristics of wavelet transform to refine the scaling transformation of load data and realize the quantification of high-frequency coefficients. The gradient calculation of long short term memory neural network is combined to improve the accuracy and reliability of load forecasting. Through the substation load data and regional office building experiments, the simulation results show that this method can effectively deal with the noise in the original load data, so as to improve the accuracy and robustness of load forecasting.
