Abstract: It is very important to provide accurate and fast load forecasting for users in power Internet of things system. Because of the variability of user activities in station area,the load of power consumption fluctuates greatly,and it is difficult to predict accurately with traditional methods. In order to meet the intelligent and multi-functional monitoring of the power Internet of things,the power load forecasting scheme based on gradient lifting decision tree is proposed. Firstly,the historical power consumption data in station area is preprocessed and the time window features are constructed. Then,the prediction model is constructed by using XGBoost and LightGBM based on gradient lifting decision tree,and the short-term power load results in the next period are predicted by the model. Based on the above predicted value of power consumption,the power consumption analysis of the station area is realized. Compared with the existing schemes,the proposed scheme can provide accurate load forecasting results,and can give out early warning in time when overload power or large-scale power outage is about to occur in current station area.