To obtain higher short-term load forecasting accuracy, it is necessary to fully consider the relationship between load variation trend and regional overall power consumption behavior pattern, so in this paper a short-term power load forecasting method driven by feature space closed operation is proposed. With this method we first divide the historical load data into multiple components to describe the characteristics of the electricity consumption behavior of the delineated area; then, according to the regional load change trend, the characteristics of the power consumption behavior are selected, the redundant input features are eliminated, and the prediction model is optimized; finally, the remaining input features are brought into the prediction model to further estimate the distribution of the load in each period. The results show that the prediction accuracy of the proposed method is higher.