Objectives:1. A support vector regression machine based forecasting model was established for predicting the monthly average concentration of permanganate index and aecal coliforms of water quality, with the water quality information from 2003 to 2005 in Yuzhong District of Chongqing. The forecasting performance were compared with BP Neutral Network (BPNN)and radial basis function neural network(RBFNN).2. A support vector regression machine based forecasting model was established for predicting the influence of the water quality on population health, based on the number of water-based infectious diseases inpatients from 2003 to 2005 in Yuzhong District of Chongqing. The forecasting performance were compared with BP Neutral Network (BPNN) and radial basis function Neural Network (RBFNN).3. These results can provide reference to water management, pollution control and the water-based infectious diseases prediction and control.Methods:SVR,BPNN and RBFNN were used to forecast the water quality and the number of water-based infectious diseases inpatients from 2003 to 2005 in Yuzhong District of Chongqing. Prediction results were compared by some statistical indexes.Results:RMSE and MAPE were used to evaluate the forecasting results. It indicates that the precision of support vector regression machine is superior to BP Neutral Network and RBF Neural Network.Conclusions:1. On study of small-sample data, SVM based on structural risk minimization principle has better generalization performance than BPNN and RBFNN which based on empirical risk minimization principle. 2. It is a practical and beneficial exploration in the water pollution control and the water-based infectious diseases prediction and control.
|