摘要
针对城市用水特点和需求,提出了一种基于小波分解与随机森林模型、ARMA模型结合的短期用水量预测的方法。采用小波分解算法将用水量时间序列分解成若干子序列,最高频子序列具有数据波动剧烈、变化速率快的特点,在实例中对比了随机森林模型、BP神经网络、Logistic回归模型、ARMA模型对于最高频子序列的拟合能力,选定ARMA模型对于高频子序列进行预测;对低频分量与部分高频分量进行预测时结合实时气象数据、时间信息、节假日信息利用随机森林回归算法进行预测,最后将各预测结果进行等权相加得到最终预测结果。实例中的数据为东南沿海城市的历史用水量数据,经实际验证,小波组合模型能明显提高预测精度,满足供水调度运行实际需求。
According to the characteristics and demands of urban water use,a method of water supply load forecasting based on wavelet decomposition,stochastic forest model and ARMA model is proposed.Wavelet decomposition algorithm is used to decompose water consumption time series into several sub-sequences.Combining with real-time meteorological data,time information and holiday information,random forest regression algorithm and ARMA algorithm are used to forecast different sub-sequences of load.Finally,the predicted results are equal-weighted to get the final prediction results.The data in the example is the historical water consumption data of small towns along the southeast coast.
作者
刘志壮
吕谋
周国升
Liu Zhizhuang;LV Mou;Zhou Guosheng(School of Emvironmental and Munici pal Engineering,Qingdao University of Technology,Qingdao 266000,China)
出处
《给水排水》
CSCD
北大核心
2020年第10期110-114,131,共6页
Water & Wastewater Engineering
基金
国家自然科学基金(51778307)
山东省重点研发项目(GG201809260435)。
关键词
小波分解
随机森林
ARMA
短期用水量预测
Wavelet analysis
Random forests
ARMA
Short term water supply load forecast
