In recent years. The Beijing-Tianjin-Tangshan region is building the electricity market and promoting the restructuring of the capital's functions to achieve industrial adjustment. With the rapid changes in the socio-economic structure, higher requirements are placed on power system services, and accurate power forecasting results are urgently needed. However, previous research on electricity consumption forecasting has focused on the changing patterns of historical electricity consumption data, with insufficient research on the correlation between historical electricity consumption data and other socio-economic data, resulting in a lack of analysis of the impact of socio-economic data on future electricity consumption. This paper first uses distance correlation coefficient to evaluate the correlation between socio-economic data and electricity consumption data. Then, filter the socio-economic data with high correlation for the prediction model. Afterwards, based on the XGBoost algorithm, multiple XGboost models are trained using different features to complete preliminary electricity consumption forecasting. Take multiple prediction results as input and input them into LSTM. Get the final prediction result by LSTM. Finally, based on historical electricity consumption data and socio-economic data, compare multiple models with the model proposed in this paper to verify the effectiveness.