Application Of Improved Grey Neural Network Model To Water Quality Prediction

Scientific water quality prediction, which has important practical significance to improve water environment protection and sustainable utilization, is the basis of water environment planning, evaluation and management. Water quality forecast has already became a subject that was widely concerned by environmental workers. Grey prediction technology and neural network prediction technology now are the usual method of water quality prediction.According to the monitoring data of total phosphorus concentration about Jialing River measured at Ciqikou in 2006, a traditional model based on the grey prediction technology and an improved one were put up. As the conclusions have shown, the best accuracy of prediction was not always resulted from the improved model. As a matter of fact, different quantity data and ways that the models were built in lead to different outcomes. This result reveals that there are limitations and risks in the application for the outcomes resulted from single method are not stable.The grey model and neural network have their own advantages and shortcomings. Grey model applys to short-term water quality prediction which has obvious trend and less fluctuation. In this way, results with accuracy are acquirable when data are few. Neural network can forecast time series which is disorderly and fluctuant, so that it has advantage in long-term water quality prediction. When original data are abundant, neural network can get accurate results by tracing changing data of water quality. But when data are few, bad results appear. By combining advantages of both method, a grey neural network prediction model based on the monitoring data about the Jialing River was built up, Average-Relative-Error of the prediction is 2.8%. As the conclusion has shown, the combination of grey prediction and neural network in the issue of water quality prediction can obviously improve the accuracy.In this paper, the traditional gray neural network model was further improved. Improved grey neural network combined model was combined three categories of improved grey model and neural network. In which the predicting results of three improved grey prediction models were used as the neural network's inputs, and the original sequence was used as the output of the neural network. The neural network was trained to get the optimal structure, weights and thresholds. The combined model was applied to predict the total phosphorus concentration of section of Ciqikou Jialing river. Results indicated that the combined model obtained highly precise forecast result,with relative errors below 5%.The average relative errors of prediction of WPGM(1,1),p GM(1,1),C GM(1,1),traditional gray neural network model and improved gray neural network model is respectively 7.19%,6.07%,9.65%,9.44%,2.85% and 2.59%, its adaptive capacity,ability of promotion and precision were better than only a single predictive method of grey and grey neural network model.