Abstract: On the basis of analyzing the classical methods of wastewater treatment plant effluent quality prediction,the paper puts forward a novel predictive method by Wavelet Neural Networks（WNN）. Firstly,the paper utilizes kernel principal component analysis method to realize reduce the dimension of the input vectors and orthogonalize the components of the input vectors.Then effluent quality predictive model is built using wavelet neural networks.The structure of WNN is trained using Structural Risk Minimization（SRM） based on statistical learning theory and weights of networks are optimized by adaptive orthogonal least square.The novel algorithm can ensure great probability for global optimization.The experiment result shows that the novel predictive model has higher precision and more flexibility.

Key words: Wavelet Neural Networks（WNN）, effluent quality, Kernel Principal Component Analysis（KPCA）, Structural Risk Mini-mization（SRM）

摘要： 在分析传统污水处理厂出水水质预测方法的基础上,提出一种核主元分析和小波神经网洛相结合的预测新方法。首先利用核主元分析实现输入变量的降维和去相关,然后运用小波神经网络建立预测模型。采用统计学理论的中的结构风险最小化原则为目标来训练网络的结构,采用自适应正交最小二乘法来训练网络权值,该方法最大限度地保证了网络的泛化能力。实验结果表明,该预测模型具有预测精度高,使用方便等优点。

关键词: 小波神经网络, 出水水质, KPCA核主元分析, 结构风险最小化