基于用电模式聚类的层级电力时序预测方法

HIERARCHICAL ELECTRICITY TIME SERIES FORECASTING METHOD BASED ON CLUSTERING OF ELECTRICITY CONSUMPTION PATTERNS

电力需求预测是城市发展和能源供应中十分重要的问题。虽然可以根据地理上的层级将其形式化为具有聚集约束的分层时间序列预测问题,但在传统的方法中,在确保聚合一致性的过程中往往会产生预测精度的损失。针对该问题,提出一种新型的基于聚类的分层电力时序预测方法。抛弃了过去直接对地理层级结构进行处理的做法,取而代之地通过聚类分析来深入探究电力消费模式,从而建立一个全新的,基于消费模式的时序层级结构。在此基础之上提出一种新的层级预测方法,大大改进了电力需求预测的效果。在真实数据场景下,大量实验证明了该方法性能显著优于传统方法,取得了最佳的精度。

Electricity demand forecasting is a very important issue in urban development and energy supply.Although it can be formalized as a hierarchical time series prediction problem with aggregation constraints according to geographical hierarchy,the prediction accuracy is often lost in the process of ensuring aggregation consistency in traditional methods.To overcome this defect,we propose a new hierarchical electricity time series forecasting method based on clustering.We abandoned the past practice of dealing directly with the geographic hierarchy structure and used cluster analysis to study the power consumption pattern in depth,so as to establish a new time series hierarchy structure based on consumption pattern.On this basis,a new hierarchical forecasting method was proposed to improve the effect of power demand forecasting greatly.In the real data scenarios,a large number of experiments show that the performance of our method is significantly better than the traditional methods,and the best accuracy is achieved.