A new approach for water quality assessment based on multivariate statistical analysis and Radial Basis Function Neural Networks
基于多元统计分析和RBFNNs的水质评价方法

Aimed to supplement previous researches, this paper proposed a new assessment method of surface water quality based on multivariate statistical analysis and Radial Basis Function Neural Networks, which was useful for the large-scale and long-term monitoring. The main procedures of this approach include: (1) analyzing the temporal and spatial differences of independent samples according to analysis of variance (ANOVA), and recognizing the samples which were statistically significantly different between each others; (2) grouping the former samples into clusters on the basis of similarities within a cluster and dissimilarities between different clusters based on hierarchical cluster analysis (HCA); (3) modeling the appropriate Radial Basis Function neural networks to evaluate the surface water quality of each class, then feeding back this results to every original samples. Moreover, its particular characteristics were that it could reduce the workload in assessment and comprehensively represent both holistic condition and individual's, and its result was objective and discriminative. The proposed method was applied to water quality assessment of Taizi river in Liaoning Province, China. The 144 original samples of six monitoring sites from 2001 to 2003 were divided into 74 significantly different samples and then into 9 clusters, and their results of water quality assessment were 2.7394, 4.4306, 4.0994, 2.777, 4.2192, 4.1214, 4.4129, 4.4259, and 4.4359, which was basically consistent with traditional simple index method. Besides, the water quality condition of each monitoring sites in Taizi River was mostly worse than Class IV.