基于非平稳时间序列的日现金流预测

Prediction for the daily cash flow based on nonstationary time series

电力公司每日售电实际收入的有效预测是实现国家电网强化存量资金高效运作以及现金流量预算按日排程的关键.由于居民用户的用电往往具有很明显的季节性影响,冬天和夏天空调使用频率高,导致这两个季节的用电量会相对较多;周末与非周末的缴费行为也会有明显的差异(周末效应).同时由于用户的缴费方式和行为的不同,特别是由于每月缴费时间的差异,使得所缴款项在每月的到账时间会有较大的波动,从而使得日现金流的预测变得很困难.针对上面存在的问题,本文提出了利用分段多阶差分非平稳时间序列探讨数据结构的周期性和非平稳特性.

Effective forecasting of the actual revenue from daily electricity sales of power companies is the key to strengthen the efficient operation of the stock funds and to schedule the cash flow budget on a daily basis.Because of the obvious seasonal influence of residential users’ electricity consumption,the high frequency of air conditioning usage in winter and summer will lead to relatively more electricity consumption in these two seasons;There is also a significant difference between weekend and non-weekend payment behavior(weekend effect).At the same time,due to the different payment modes and behaviors of users,especially the difference of the payment time,the arrival time of the payment will fluctuate greatly in the each month,which makes it difficult to forecast the daily cash flow.In view of the above problems,the periodic and nonstationary characteristics of data structure are discussed by using piecewise multi-order differential nonstationary time series.