Research On Water Quality Prediction Based On Support Vector Regression

Water quality prediction is a basic work in water resource management and pollution control. Building mathematic model to predict water quality accurately is an effective means of water pollution management. At present, although there are many existing methods for water quality prediction, few general and reliable models can be found. There remain quantities of arbitrariness when selecting models for predicting water quality. SVR (Support Vector Regression) is a new machine learning method based on VC-dimension and SRM (structural risk minimization) principle which demonstrates much superiority in solving small samples, nonlinear and multi-dimensional pattern recognition.Against the complexity of water environment, this paper attempts to utilize SVR algorithm for water quality prediction, and engage in application analysis which enriches the theory and methods of water prediction, and provides scientific technical support for water environment detection and management, with the result of practical application value.The water quality monitoring data of some section in Xiaoqing River Basin in Shandong Province has been considered as the research object. Building SVR time series prediction model is to calculate COD concentration and NH3-N concentration in a certain period of time. SVR prediction model can be established to search for the most optimized parameter to predict water quality through adjusting input variables to select different kinds of model parameters to optimize parameter and analyzing the effect of penalty coefficient, insensitive coefficient and coefficient of RBF kernel function on the model accuracy combined with experiment. It can be found out from the comparison between prediction result and BP neural network that the average predictive error of SVR model is0.39%which is smaller than the BP neural network whose average predictive error is0.85%. From the overall predictive effect, SVR predictive performance exhibits more superiorities than BP neural network. SVR model could reflect the real situation of water quality better and predict water pollution accidents more accurately.In this thesis, SVR predictive model has been applied in automatic water monitoring and early warning system. We bring in water warning technology and construct an automatic water monitoring and early warning system on the basis of water automatic monitoring management system of Shandong province. After analyzing the difference and relation between water quality prediction and early warning, we introduce SVR predictive model into early warning of water quality and preliminary discuss the implement of water quality warning technology based on SVR predictive model.