The Establishment And Application Of The Neural Network Operating Model Of Urban Sewage Treatment Plants

Along with the development of high speed computing of computers, neural network models are concerned for the application in management of dynamic environment especially for the prediction and evaluation of the sewage treatment plants based on operation and management. As for the objective and scientific weights, the powerful ability to deal with multivariate in the way of nonlinear fitting, and the strong features in robustness, memory capacity and nonlinear fitting, it is a better way to evaluate and manage the dynamic multi-variable system. Therefore, neural network models has more advantages in the prediction and evaluation of the sewage treatment plants based on operation and management.Considering the water quality indicators, environmental indicators, economic indicators and indicators of supervision and management, which based on the existing evaluation system of sewage treatment plants in Shenzhen, it obtained relevant data and processed those data in the way of regression analysis and residual analysis to gradually establish and optimize the evaluation index system; then the BP neural network models were established based on historical data of Shajing sewage treatment plant and the optimized evaluation index system; next the performance of the BP neural network models were analyzed by models training, models testing and models predicting based on the other history data; according to adjusting the numbers of hidden layers and the nodes, the optimal model was ultimately choiced.Checking the stability of optimal model, obtaining the optimal model of high stability: [6,6,7], getting the weight matrix from the optimal model of high stability and analyzing the weight matrix by means of the significant analysis. It could be concluded as follows: the weights of unit sewage flocculant consumption and operating load rate reached the maximum, which were 0.132, 0.128; yet the weights of COD in effluent and influent flow, which had larger impact on the actual operation and management of the sewage treatment plants, were very small. In order to obtain the maximum of the weights of COD in effluent and influent flow, it needed to increase or decrease the numbers of index variables, hidden layers and nodes of the optimal model of high stability; through checking the stability of the model, it finally obtained the depth optimization modelof high stability: [10,5,8,7].Finally, the depth optimization modelof high stability was selected as the comprehensive evaluation model by testing the predictive capability and accuracy of models. Comparing the weights between the comprehensive evaluation model and operation and evaluation model of sewage treatment plant in Shenzhen, it provided related policy recommendations in operational management, energy conservation as well as economic costs to Shenzhen municipal government and the managers of Shajing sewage treatment plant.