Research On Water Quality Prediction Model Of Qinghe Reservoir Based On Neural Network

Water quality prediction is a trend that predicts the future changes of water quality by predicting the status of river water quality,the characteristics of pollutant migration and the discharge of pollutants,and plays an active role in the field of water resources protection.Reasonable water resources management and environmental planning for water pollution prevention are important tasks for water environmental protection,and water quality prediction is an important factor for the smooth realization of these tasks,and is an indispensable basic work with universal significance.By reading the relevant literature and materials constructed by water quality models at home and abroad,the research status of related water quality model construction at home and abroad was studied,and the construction methods of water quality models were summarized.At the same time,for the study area,collect and organize the data to determine the research method.Taking the water quality of Qinghe Reservoir as the research object,based on the analysis of the water quality model construction methods at home and abroad,combined with the water quality status and monitoring data of Qinghe Reservoir,BP neural network and NAR neural network were used to establish the water quality prediction model of Qinghe Reservoir respectively.The main pollution index of the reservoir is simulated and predicted.The main research results are as follows:(1)BP neural network water quality prediction model based on grey relational analysisWhen the BP neural network was used to construct the water quality prediction model,the gray relation analysis method was used to process the water quality monitoring index and the model prediction index of the study area,and the model input node of the BP neural network water quality model was optimized to avoid water quality monitoring.The interaction between the indicators,thereby optimizing the network structure of the model,shortening the training time of the model,and achieving the purpose of reducing the simulation error of the water quality model.From the simulation of the model,the fitted values of the predicted values and the monitored values are compared.High,very few months,the error is larger.The regression R values of the constructed GRA-BP model in the three parts of training,test and test are 0.918,0.961,0.900,respectively.The total regression R value of the model is 0.921,which is close to 1,indicating that the BP model input is performed using the gray correlation degree.There is a strong correlation between the influence factor obtained by the optimization of the node and the total nitrogen concentration,indicating the good performance of the model.When using the constructed GRA-BP total nitrogen prediction model for prediction,the maximum absolute error is-4.83%,the minimum absolute error is-2.91%,and the average absolute error is 2.05%;indicating that the model has better generalization ability.At the same time,it has better predictive performance.(2)Water quality prediction model based on NAR neural networkAccording to the research object,the water quality prediction model was constructed by NAR neural network technology.In order to make the NAR model more practicable,the forecasting object is selected as the monitoring index total nitrogen concentration of the second-grade standard of the annual super-standard water in the water quality monitoring index of Qinghe Reservoir.In order to better determine the delay order of the NAR neural network model,first-order difference is made to the total nitrogen monitoring data sequence to construct a stationary sequence,autocorrelation detection and partial correlation detection;through the selection of empirical formulas,and the trial and error method,The number of hidden layer neurons in the NAR neural network is determined.The training algorithm suitable for this model is determined by analyzing the sample data.Finally,the autocorrelation error analysis of the NAR neural network model and the response analysis and error analysis of the model are analyzed.To complete the training of the model,it can provide a new method for the construction of the Qinghe Reservoir water quality prediction model.