Abstract: In view of its excellent prediction performance of support vector machine regression (SVR) model for nonlinear system, a model of residual chlorine prediction was put forward to predict changes in water supply system based on SVR. Moreover, two-order oscillating particle swarm optimization algorithm (SOPSO) was employed to optimize the SVR model parameters in order to enhance the model precision in small sample situations and improve the generalization ability of the model. This optimized model was applied to predict the residual chlorine in a water supply system, and the results showed that: in the case of limited samples, the average prediction error of the optimized SVR model is 3.86%, which is better than that of BP and ARX prediction models, and also has strong stability. This model could solve the problems of low fitting accuracy and poor efficacy of prediction which often appear by traditional models. It provides a new approach for the model construction and algorithm selection in residual chlorine prediction for water supply system.

Key words: residual chlorine, SVR, PSO, parameter optimization, water supply system