Research On The Modified SOM And Its Application In Water Quality Evaluation

Along with great-leap-forward development of economic, water resource environment is facing tremendous pressure, which has brought people's lives and property safety and social stability in crisis. The Realization of objective, accurate and proper evaluation of water quality can protect people's lives and property and achieve sustainable socio-economic development, therefore is of great economic and social significance.Based on water quality evaluation, the methods and significance of a comprehensive evaluation of water quality are introduced in this paper as well as the status quo of water quality evaluation. With above-mentioned works, methods for water quality based on NN (Neural Network) and PCACE (Principal Component Analysis Comprehensive Evaluation) are both studied. Because water environment is a complex nonlinear system, many correlative characteristics are often existing in evaluation index system at the same time in the evaluation process. These characteristics have a different weight respectively in the comprehensive evaluation of the sample. Therefore, this paper studies on the available combination of the NN and PCACE according to their respective advantages and disadvantages to enhance the capability of NN in processing multi-dimensional data with strong correlation.The proposed PCA-weighted SOM (Self-organizing Map Neural Network) model, implants statistics into SOM Neural Network and converges the advantages of SOM and PCACE, is a comprehensive evaluating model based on SOM and PCACE that sets self-organization, self-learning and high-dimensional nonlinear information processing as a whole. The principal characteristics component of evaluating target is extracted using PCA, and these extracted principal components are inputted into the NN instead of original data. Moreover, the variance contributes of the principal component are introduced to Euclidean distance as weights. Hence, the structure of PCA-weighted SOM and its algorithm are studied. The comparison of evaluating results between SOM and PCA-weighted SOM shows the superior performance of the latter approach, such as in converge rate, error control, etc.The proposed PCA-weighted SOM is implemented to analyze and evaluate the water quality in the main river in Chongqing municipal area. The research focuses on the selecting of the characteristic and the constructing of the evaluating system in this process. Applying the 2-dimensioned PCA-weighted SOM in water quality evaluation, the PCA-weighted SOM model is constituted with SPSS and MATLAB. The model is tested with the river water quality monitoring data. The results show that the application of the proposed model in water quality evaluation makes the evaluating results abstain from subjective factors, that is, achieves a reasonable, credible result. It avoids a large number of complicated calculations and achieves an effective evaluation.