With the arrival of the era of power big data, higher requirements are put forward for the accuracy of power load forecasting. Accurate power load forecasting is of great significance to the safe and stable operation of the power system and to reduce the cost. According to the characteristics of different types of short-term user loads showing different load operation laws in different periods of time, different forecasting methods are adopted in different periods of timewithin the daily range. According to the applicable characteristics of the model, different weights are given in different periods, so that it can reflect the forecasting effect of each single model in different periods of time, and use this weight for combined forecasting. The simulation results show that the accuracy of load forecasting can be effectively improved by considering the characteristics of load behavior in different time periods.