| Water quality prediction is the important content of water pollution control technology. The prediction results can provide important basis for environmental protection department. At present, most study is still limited to mechanism model, the non-deterministic and highly nonlinear of water pollution process makes the mechanism model can't accurately simulate the water quality contaminants complex process. ANN is the frontier of complex non-linear science and artificial intelligence science, its simulation of complex nonlinear system can obtain good prediction results. But its applied research in water quality forecasting is still in its infancy. Our group in previous work used ANN to research water quality prediction model of Nansi Lake and Moshui River, the results showed that ANN can obtain better forecasting results.This paper taked Xinxue River constructed wetland of Nansi Lake as the area for studying. First, the main influent and effluent pollutant concentration of the constructed wetland were monitored and researched the wetland purification effect to the river, and provided data base for the establishment of wetland water quality model. Then, from the perspective of systems theory, researched the wetland water quality model based on BP neural network, and applyed the model to predict the largest influent loads under the restrained conditions of theâ…£standard of"Surface Water Environment Quality Standard". The water quality prediction results formed the basis of the research of water quality and quantity scheduling scheme.The main findings were as follows:1)By making research on Xinxue River constructed wetland, monitored the diversion channel influent CODCr, ammonia, TN and TP concentration, as well as the processing unit effluent CODCr and ammonia concentration. The pollutant removal effect were studied, and obtained the relationship of pollutants purifying loading and organic polluting loading. Research showed that wetland had good pollutants purification capacity. The construction of wetlands greatly improved the water quality of Xinxue River.2)Based on fully study mechanism of ANN and research status, established prediction model for wetland water quality based on BP neural network. In the process of model building, in order to enhance the network generalization ability and improve training efficiency and precision, made a series of optimal design. Then by continuous trail calculation, analyze the performance of the network model with different hidden nodes and different training methods to select a reasonable network structure and training methods. Finally the network topology structure was determined as follows: the three nerve network model of one input layer, one hidden layer, one output layer. The hidden nodes of summer model was 9, training function was trainscg. The hidden nodes of winter model was 14 and its training function was traingda.3)After the model was established, selected the first 11 sets of data as training samples respectively in wetland summer and winter sample data, and the 12 sets of data as test samples. Test results showed that the relative prediction error of CODCr, ammonia, TN, TP in summer model were 1.5%, 2.4%, 7.7%, 8.3%, respectively. And in winter model they were 11.7%, 30.6 %, 3.3%, 27.3%,respectively. The network prediction error was under allowed band, the network training is successful with satisfactory prediction, its performance could meet the requirements of practical application.4)The established wetland water quality prediction model had been applied. By the prediction we could see, in the assurance that the effluent of Xinxue River wetland met theâ…¢standard of"Surface Water Environment Quality Standard", the predictive value of constructed wetland summer model of CODCr and ammonia should be less than 48.8 mg/L and 2.23 mg/L respectively. And the winter model should be less than 26.1mg/L and 1.17mg/L respectively. |
PDF Full Text Request