摘要
采用主成分分析法筛选出显著的驱动因子,结合灰色关联分析将筛选出的驱动因子进行灰色聚类和优势分析.以用水量驱动因子为基础,利用SPSS建立多层感知器网络,利用矩阵实验室建立GRNN神经网络和BP神经网络.将诱导有序加权平均算子(IOWA)应用到水量预测模型中,构建基于IOWA算子的MLP-GRNN-BP组合用水量预测模型,最后建立由平方和误差(SSE)、均方误差(MSE)、平均绝对误差(MAE)、平均绝对百分比误差(MAPE)、均方百分比误差(MSPE)和Theil系数(μ)组成的预测效果评价体系,评价预测模型的预测效果,最后以重庆市的用水量预测为例,验证以上方法的可行性.结果表明:经过主成分分析及灰色关联分析,可将用水量驱动因子由31个降为12个,12个驱动因子可综合为4个聚类,可确定4种用水量的各自驱动因子的重要性排序;BP,MLP,GRNN和MLP-GRNN-BP组合模型预测结果的MAPE,MSPE和Theil系数均在5%以内.
Urban water consumption prediction is full of complexity due to the different influencing factors and the uncertainty of the statistics database. A comprehensive water consumption prediction model was de- veloped to slove this problem. Multiple approaches were integrated into this model. Specifically, the signifi- cant influencing factors of water consumption were selected by principal component analysis： then, the se- lected influencing factors were further classified by gray cluster analysis and gray relational analysis. Based on the evaluation of the significant influencing factors, a multilayer perceptron network was established by SPSS software, GRNN and BP neural networks were established by Matlab software. IOWA operator was also applied to the water prediction model. Consequently, a MLP -GRNN -BP comprehensive water demand consumption prediction model was developed based on the IOWA operator. The evaluation system,using the sum of squares error （SSE）, the mean square error （MSE）, mean absolute error （MAE）, mean absolute percentage error （MAPE）, mean square percentage error （MSPE） and Theil coefficient （/x） , was estab- lished to evaluate the performance of the predictive models. The proposed models were applied to the city of Chongqing for the municipal water consumption prediction to verify the feasibility of these methods. The re- suits show that the number of significant influencing factors for water consumption in Chongqing city can be reduced from 31 to 12, and the 12 significant influencing factors can be further classified into four clusters, and four kinds of the influencing factors can be analyzed and ranked respectively. MAPE, MSPE and Theil coefficient values are within 5%, when the BP, MLP, GRNN and MLP - GRNN - BP model are used in the case study, indication a good prediction.
出处
《排灌机械工程学报》
EI
北大核心
2014年第12期1051-1056,共6页
Journal of Drainage and Irrigation Machinery Engineering
基金
重庆市科技计划项目(CSTC2006AB7020)
关键词
用水量驱动因子
用水量预测
神经网络
IOWA算子
主成分分析
influencing factors of water consumption
water consumption
neural network
IOWA operator
principal component analysis
