Abstract: In view of the poor prediction accuracy of existing short-term load forecasting methods for power systems, a short-term load forecasting method based on the combination of long short term memory (LSTM) and CatBoost is proposed. Firstly, in view of the time-series and nonlinear characteristics of the power load data, and the fact that the long-term and short-term memory networks can not directly deal with the categorical features, the LSTM load forecasting model and the CatBoost load forecasting model are established for the processed power load data. Secondly, the weighted coefficients are determined by the inverse variance method, and the predicted values of the LSTM and CatBoost models are obtained. Finally, the validity of the algorithm is verified by the actual load data. The prediction results show that the combined model of LSTM and CatBoost can significantly improve the accuracy of load forecasting.