With the continuous development of the power industry, the prediction of power load on the power us- er side has played an important part in meeting the balance between power supply and demand of users and power grid planning. Under the background of big data, in order to improve the accuracy of power load forecast- ing results, and in view of the influence of the time for historical data, considering the limitations in historical data of the same period and recent historical data respectively, based on the principle of time-dominant, the ex- ponentially weighted moving-average was introduced to distribute the weight of data at different time, and an improved power load forecasting model was proposed. Taking the power load forecasting in a certain area as an example, the predicted result was improved by 29.5% in root mean squared error and the mean absolute percent error increased by 25.7%. The analysis results show that the proposed model is feasible and has high accuracy, providing a reliable reference for the power load forecasting.