本文对韩志刚提出的“多层递阶预报方法”作了改进,其中涉及松驰因子的选取,初值确定,非平稳序列的处理,参数建模等方面,还给出了预报的置信区间。改进后的方法经过验证和在上海地区用电量分季度预报中的应用证明优于原方法。 This paper improves Han Zhigang’s “multi-level forecasting method”, which includes the selection of slack factor, initial value determination, non-stationary sequence processing and parameter modeling, and also gives the confidence interval of forecast. The improved method has been verified and the application of quarterly forecast of electricity consumption in Shanghai proves to be superior to the original method.