论文部分内容阅读
In this paper,two artificial neural networks,BP neural network and Fuzzy Neural Network (FNN),were employed to simulate the water qualities in municipal water supply networks of Suzhou.Residual chlorine concentration was chosen as the index for simulating.Residual chlorine concentrations and water yield data of three waterworks in Suzhou,as well as the on-line residual chlorine concentrations of Jin Men Road,where located one of the on-line water quality monitoring sites and the water quality at there was thought to be affected by all the three waterworks,were adopted to establish the artificial neural network models.The simulating results show that both of the two neural network models can analog the concentration of residual chlorine in water supply pipe at Jin Men Road.The maximal relative error of BP network model is 52.7% for the 615 groups of training data,and the average root mean square error (RMSE) is 0.113 for the 44 groups of validating data.Meanwhile,the maximal relative error of FNN model is 39.4%,when using the same training data groups as the BP model and the average RMSE is 0.076 for the same validating data groups,which indicates the FNN model is more accurate than the BP model.The FNN model is more suitable for simulating the residual chlorine concentration in municipal water supply pipe networks in Suzhou and it can meet the requirements of waterworks.Moreover,the FNN model has the advantages of fuzzy control and neural network,in accordance with the dynamic adjustment of the implementation of variable controller to adjust the dynamic parameters,and needs less regulating parameters,which is convenient for application by Adaptive Neural Fuzzy Inference System.Therefore,the FNN model has certain reference valuesand can be used on the simulation of water quality and the prediction of the urban water supply systems.