Research On Prediction Of Residual Chlorine At The End Of Pipe Network Based On Dynamic Neural Network Optimization Model

Chlorine is the most widely used disinfectant in urban drinking water.It can control the growth of bacteria in water and prevent diseases from spreading through water.Therefore,in order to maintain a continuous disinfection capacity in the pipe network,China's"Sanitary Standard for Drinking Water"?GB5749-2006?requires that the factory ClO₂ should be greater than or equal to 0.1mg/L,and the residual chlorine in the pipe network should be greater than or equal to 0.02mg/L.Due to the instability of chlorine,nonlinear attenuation occurs in the pipe network as transportation time increases.Therefore,it is difficult to control the dosage of chlorine.The insufficient dosage can not inhibit the microbial stability in the pipe network,and the excessive chlorine content is easy to react with the organic matter in the water to produce disinfection by-products,which is harmful to the health of users.Therefore,the precise control of the amount of chlorine added to the effluent is of great significance for the enterprise to control the deterioration of water quality and reduce the disinfection by-products to improve resource utilization.This paper collects730 data of pH value,turbidity,chlorination amount and residual chlorine content of three end detection points in a two-year continuous water plant in a water plant in southern Yunnan Province to establish a dynamic neural network pipe network residual chlorine prediction model.The research results and conclusions are as follows:?1?Determine the chaotic characteristics of the residual chlorine time series.In this paper,the mutual information method,FNN algorithm and C-C method are used to determine the delay time and embedding dimension.Finally,the delay time of 1#-3#detection points is 5,3,and 5 respectively,and the embedding dimensions are 6,7,and 5 respectively.The pHase space reconstruction of the three groups of residual chlorine sequences is carried out by using the calculated values to restore the folding law,and it is convenient to find the mapping relationship between similar law numbers.The maximum Lyapunov exponent is used to judge the chaotic characteristics of the residual chlorine sequence.The results show that the maximum Lyapunov exponents of the three detection points are larger than0,which has obvious chaotic characteristics,which proves the predictability of the time series.?2?Establish a residual chlorine prediction model for CS-Elman neural network.The Elman neural network is trained by using the residual chlorine data of the detection points and predicting the water output data of the corresponding water plants.However,since the implicit layer of the Elman neural network is determined to be subjective,the initial weight and threshold determination are random,which seriously affects the network prediction accuracy.Therefore,the adaptive iterative method is used to find the optimal number of hidden layers.The cuckoo algorithm?CS algorithm?is used to optimize the initial weights and thresholds,and an adaptive CS-Elman residual chlorine prediction model is established.The results show that the average MSE of the CS-Elman residual chlorine prediction model in the prediction accuracy analysis of the three detection points is5.61×10-4,R² is 0.91,MAPE is 8.17,the average absolute error is 1.34,and the maximum relative error is 18.78.The average MSE of the unoptimized Elman neural network model decreased by 72.50%,R²increased by 23.1%,MAPE decreased by 65.05%,the average absolute error decreased by 64.56%,and the maximum relative error decreased by 57.99%,indicating that the CS algorithm can be largely Reduce the prediction error of Elman neural network and improve the prediction accuracy of the model.?3?Establish a GA-NARX neural network residual chlorine prediction model.In this paper,the above data is used to train the NARX neural network.Since the NARX neural network uses the gradient descent method for learning training,it is easy to fall into the local optimal value,and the combination of the hidden layer number and the delay time is required to be artificially predicted.The value accuracy is not high.Therefore,the adaptive iterative method is used to determine the optimal delay time and the combined number of hidden layers.The genetic algorithm?GA algorithm?is used to optimize the NARX neural network,and the prediction accuracy of the two different structures NARX neural network before and after optimization is compared..The results show:?1?the OGA-NARX model is higher.The average MSE of the ONARX neural network is reduced by 98.10%,the R² is increased by 17.60%,the MAPE is decreased by91.74%,the average absolute error is decreased by 91.67%,and the maximum relative error is reduced.It fell by 67.96%.?2?Comparing the two different structural optimization models CGA-NARX and OGA-NARX,it is found that the predicted values??of OGA-NARX neural network and CGA-NARX neural network are not much different,but the CGA-NARX neural network is more complex,and the overall accuracy of prediction is slightly lower.The average MSE of OGA-NARX was 21.55%lower,R² was increased by 0.34%,MAPE was 22.03%lower,the average absolute error was 22.91%lower,and the maximum relative error was 2.36%higher than CGA-NARX.It is proved that the genetic algorithm can greatly reduce the error of NARX and improve the learning efficiency of NARX.?4?Multiple nonlinear regression was used to fit the predicted results of the neural network,and the amount of chlorine added in the factory was calculated reversely.Comparing the prediction results of different models in the first two chapters,it is found that the prediction effect of the OGA-NARX model is better.However,because the neural network has the"black box characteristics",although the accurate prediction of residual chlorine can be realized,it is difficult to understand the correlation between the influencing factors and the residual chlorine.Therefore,the fitting performance of the multiple linear regression model is used in this paper.Fitting the network prediction curve,deriving the relevant regression equation,establishing the functional relationship between the factors and the residual chlorine content of the pipe network,and guiding the control of the chlorine content of the water plant.The results show that:?1?residual chlorine is inversely proportional to pH and turbidity,and proportional to the amount of chlorine added.?2?In the case that the amount of residual chlorine in the first few days of the pipe network is met,when the pH value is between 6.88 and 8.39,the effluent meets the turbidity of 1 NTU,the amount of chlorine added is 0.01 mg/L,or the effluent turbidity is 3 NTU,and the amount of chlorine added is 0.06.The mg/L can ensure that the residual chlorine in the three detection points at the end of the pipe network reaches the standard.When the turbidity is 7NTU in the limit condition,increasing the chlorine content by 0.25mg/L can ensure that the residual chlorine content of the three detection points is up to standard.This paper proposes a residual neural prediction model based on dynamic neural network,and uses multivariate nonlinear analysis to further analyze the influencing factors of residual chlorine,and proposes a new and efficient method for water plant residual chlorine prediction,which can adjust the water supply for water supply enterprises.The indicators give the most direct reference and also provide a new way of thinking for research in other industries.