Study Of Water Quality Assessment And Parameter Prediction Based On Support Vector Machine

Ground water quality assessment is one of the most important factors among all the ground water resources assessment. Its objectif is to analyse the level of utility of ground water according to the comparison between the physicochemical value of the major components and the water quality standard and so that it is the scientific basis for the exploitation, utilization, planning and management of water ground water resourcesAt present, there are many mathematic models for water quality assessment such as indices method, fuzzy mathematic method, gray-clustering method etc. However these traditional methods have not resolved the non linear relation problematic between the assessment factor and water quality classes. The bias is given manually which is the constraint of the generalization of assessment pattern and also conducts the influence of the reliability of the result. Support machine vector, new techniques of the statistical learning theory proposed by Vapnik in 1995 and developed from the Structural Risk Minimization (SRM) theory, provides a good solution in order to carry out the objective above. After the study of SVM theory, we performed the main work as follows:In this paper, we studied SVM as regression techniques for water quality parameters prediction. The result of the experiment shows that the method based on SVM is more performance compared to the two ANN approaches: BP Neural Network and RBF Neural Network.We studied also the usability of SVM as classification techniques for water quality assessment in Jinan City. The three SVMs with different kernel functions are used to process the water monitoring datas. The result of experiment shows that this approach is feasible, with the advantage of independence on the type of kernel functions.With analyzing the result of this assessment approach we also compared this result with the traditional approaches: single factor method, fuzzy method, BP neural network method. It shows that SVM method is in accord with the objective reality.Upon the work above, a new model based is proposed on the SVM classification and SVM regression, that is hybride SVM Model.