Exploration For Water Quality Prediction Based On Neural Networks And Gray System

Water quality prediction is the basis of water environmental planning, evaluation and management. On the basis of analysing the research in water quality prediction, and learning the basic principles and the optimal algorithm of artificial neural network(ANN), fuzzy mathematics and gray system, this dissertation gives an application of compensative fuzzy neural network(CFNN)and gray dynamic model group approach to water quality prediction. The main research is to do some exploratory work of water quality prediction in order to improve the accuracy of water quality prediction.This paper describes the principle, network structure and algorithm of CFNN, establish a CFNN model, which has fast learning algorithm and can execute compensatory fuzzy reasoning to overcome the technical difficulties of conventional fuzzy neural network(FNN). The CFNN model is used to predict sewage treatment plant influent water quality. Found that CFNN model has higher forecast accuracy, better fitting and more reliable forecast than conventional FNN. In addition, the study also found that CFNN model can give a very accurate prediction in the case of greater sewage index values and flatter numerical fluctuations, but in the opposite case, the result is less satisfactory prediction. So if CFNN model is applied to predict the accurate indicators of water quality, may be it could not appropriately exhibit its advantage.This paper describes the principle of gray system and introduces the modeling mechanism of gray dynamic model group which is used for water quality prediction. On the basis of the modeling mechanism, this dissertation studies the actual water quality monitoring data of Zhuozhang river, established the corresponding gray dynamic model to make an initial verification of the model performance. Then the gray dynamic model group predicts the same data which has been predicted by CFNN. The study found that, in the case of the training samples which have larger values and good stability, CFNN model can show their good learning performance and reliable predictions, while in the opposite case, gray dynamic model group shows a fairly dynamic advantage to give more accurate predictions. Both methods have advantages, thus we can make choices under different circumstances.