Based on the analysis of the variation characteristics of Shanghai's daily electrical load, this paper comes to the conclusion that both the tendency function and the periodic functions with obvious physical meanings can be picked up from small samples of power load by means of the exponential weight moving average. The result is the formation of a stationary linear random series which can be approached by autoregressive model AR(1). Thus, the random model of the daily power load forecasting can be obtained. With regard to the future prospective sudden change, it can be modified by the forcasting value of daily electricity demand with the interactive mode. The model will thus automatically check the future forecasting value of hourly power load, so that a new level of forecasting technique will be reached. This essay also describes the actual result during operation and the average deviation rate is 1.79% from Oct.1983 to Nov. 1985. It proves that the forecasting accuracy of this model is satisfactory and its operation is very easy. It has been running smoothly and successfully in the Shanghai Power Supply Bureau up to now.