基于耦合主成分法的小波网络模型城市需水预测

Urban water demand prediction based on wavelet network coupled with principal components analysis

为了尽可能准确预测城市未来的需水量,以上海市为例,对年用水量的9个相关因子进行主成分分析,得到两个综合因子。将两综合因子与用水量的历史数据一起作为输入项,建立一种小波网络模型,以1980～2005年的数据为训练样本,采用引入了附加动量项和自适应学习率的BP算法进行模型率定,并以2006～2008年的数据对模型进行了检验。结果表明:所建模型结构简洁,收敛速度与预测精度均较为理想,在城市需水预测中有着广阔的应用前景。

Taking Shanghai City as an example,we extract two comprehensive factors by principal components analysis from 9 factors relevant with the annual water consumption,which are adopted as inputs with the historical water consumption data to establish a wavelet network model.The improved BP algorithm with additional momentum method and adaptive learning rate has been applied for model calibration using data from 1980 to 2005;and the model is verified based on the data from 2006 to 2008.The results indicate that the proposed model is superior in the structure,convergence rate and prediction precision,which has a broad application prospect in urban water demand prediction.Choosing the best hidden layers and hidden layer nodes to obtain the faster convergence rate is the focus in further study.