Prediction Of Permanganate Index Based On BP Neural Network

The Wei river plays an important role in the economic development of Guanzhong area,and at the same time,it also become an important thoroughfare for containing the pollution.In recently years,with the fast development of regional economy,people have made serious damage to the water resources by overusing and discharging series kinds of pollution into the Wei river so that the water quality declined rapidly.As one of the most important comprehensive index of organic pollution for charging the water environmental,The COD_Mn could make judgment for the relative content of the organic matter contained in the water.This paper studied about the density of the COD_Mn around the Tongguan drawbridge area on this background.Nowadays,the predictive research of density of the COD_Mn in water mostly focused on the long-term forecast.The shortcoming of this way is that we have no idea to do with the situation when we what to add the governance of water environment into the model as a valid variable,so that the results of long-term forecast usually have great disparity with the values in actual detection,and the practical significance of those kinds of forecast are not that strong.At the same time,the short-term prediction studies of the COD_Mn are less,it caused a phenomenon that we can hardly make preparation when the water quality suddenly deteriorated.This paper will build the ARIMA time series model,BP neural network model and BP neural network model optimized by GA respectively basing on the time series analysis,BP algorithm and GA algorithm,and study about the density of the COD_Mn around the Tongguan drawbridge area in short term.With verifying the relative error between predicted value and actual detection value,finally I got a method of prediction showing remarkable accurate prediction accuracy.This paper verified the stationarity of the sequence and the first order difference by performing the time series analysis for the COD_Mn density of cross section in last three years,and built up the ARIMA time series model,which shows remarkable fit effect.The forecast results showed that the short-term accuracy is quite good.But there is also a problem that the predicted relative error will gradually increase as time went by.Build up the BP neural network model on this background,It comes a result that the relative error of the prediction consequent which comes from the BP neural network model is fairly ideal,but the problem is it's easily falling into the local minimum.Then build up the BP neural network model optimized by GA according to the shortcoming of BP neural network model.Optimized the original connection weight of neural networks with the help of genetic algorithm,and improved the network search capabilities,speeded up the convergence rate,and solved the problem of easily falling into the local minimum.It comes a result that the BP neural network model optimized by genetic algorithm works better than common BP neural network model not only in the fit training of the COD_Mn but also in terms of forecasting results,because it has less error and more suitable for short-term density prediction of the COD_Mn.