基于主成分分析的RBF神经网络在需水预测中的应用

针对需水预测方法大都存在一定的局限性,导致预测值与实际值差别较大的问题,采用主成分分析与径向基函数(RBF)神经网络相结合的方法建立需水量预测模型。首先借助SPSS进行主成分分析,对影响因子进行降维处理,以此减少各影响因子之间原有的多重共线性;其次,选用RBF网络,运用Matlab神经网络工具箱,建立了基于主成分分析的RBF(PCA-RBF)神经网络需水预测模型;并以辽宁西部地区凌源市需水预测为例,对预测模型进行了校核。利用训练好的PCA-RBF神经网络需水预测模型对凌源市2014、2015年的总需水量进行模拟预测,预测结果与实测数据相对误差分别为2.9%、0.4%。这说明该模型可相对全面地模拟需水量变化规律,能够用于半干旱山区和材料相对较少时需水量的精准预测,为水资源规划管理提供了理论依据。

Water Demand Forecast Model of RBF Neutral Networks Based on Principle Component Analysis

Most water demand prediction methods have certain limitations and bring in large difference between predictive value and actual value. In this paper, the principal component analysis and radical basis function neural networks (RBF) are used to build up the water demand prediction model. Firstly, the principal component analysis (PCA) conducted by SPSS is used to reduce the original multi-collinearity among the influencing factors. Secondly, RBF network is chosen to predict the water demand. Finally, the water demand precondition model of PCA-RBF neural networks based on principal component analysis is built up by neural networks toolbox of Matlab. The water demand of Linyuan County in Liaoning Province is taken as an example to check the precision of PCA-RBF model. The results show that the relative errors between predicted and actual values in 2014 and 2015 are 2.9% and 0.4%, respectively. This result indicates that water demand prediction model of RBF neural networks based PCA can predict water demand precisely, which can be used on the water demand prediction in semi-arid mountainous areas with relatively few factors and provide the theoretical basis for water resources planning and management.