Prediction Of Manganese Concentration In Surface Water Based On VMD Algorithm And GA-BP Coupling Model

In recent years,water pollution events with excessive Manganese(MN)content in surface water have occurred from time to time,which have a great impact on ecological safety and human health.Since the detection of Manganese concentration is limited by labor and time costs,using water quality models to predict its concentration and development trend has become an alternative method.With the rapid development of artificial intelligence,the mathematical modeling method of Artificial Neural Network based on deep learning provides a new idea for water quality prediction.Among them,BP Neural Network has strong nonlinear fitting ability,which can build a water quality model to predict the concentration and development trend of Manganese according to the interaction between different water quality indicators.Single BP network has the disadvantages of strong randomness in the selection of initial weights and thresholds,easy to fall into local extremum,and the prediction accuracy is often not high enough.Genetic Algorithm(GA)is an adaptive heuristic global optimization algorithm,which is suitable for the optimization solution of nonlinear prediction problems,and has been widely used in the fields of function solution and optimization,and can effectively make up for the defects of BP Neural Network.The reliability of water quality data is the decisive factor for the prediction performance of the model.However,various water quality indicators with significant correlation differences and monitoring data with strong volatility bring great difficulties to water quality prediction.If there are too many water quality indicators,the key variables can be selected by correlation analysis method,and the signal decomposition method is needed to reduce the impact of water quality data volatility on the prediction accuracy of the model.Variational Mode Decomposition(VMD)can adaptively decompose the data into multiple stationary sub modes according to the discretization characteristics of hydrological time series samples,which can effectively reduce the complexity of the data and better reflect its high and low frequency characteristics,so as to effectively improve the prediction accuracy of water quality models.Based on this,in order to improve the prediction accuracy of the traditional model and realize the accurate prediction of manganese concentration in surface water,this paper combines Variational Modal Decomposition(VMD)and Genetic Algorithm(GA),and based on the excellent nonlinear fitting ability of BP Neural Network,proposes and constructs a multilayer decomposition VMD-GA-BP water quality prediction model.The main research work of this paper is as follows:(1)Based on the historical data of a typical urban surface water,the Pearson correlation coefficient method and Random Forest(RF)feature scoring algorithm are used to quantitatively calculate and classify the correlation degree between the 18 manually determined water quality parameters and the manganese index to be predicted.Seven water quality indexes with Manganese correlation coefficient greater than 0.5 and high correlation score are comprehensively selected through the two methods and used as the input variable combination of the model,It effectively reduces the workload of the model and the interference of indicators with poor correlation.(2)According to the characteristics of non-stationary,strong noise and periodicity of water quality data,the VMD algorithm is used to decompose the complex hydrological time series into four stable eigenmode functions(IMF)with step-by-step frequency change,so as to reduce the fluctuation of data and the influence of random factors,and achieve better denoising effect.(3)Considering that the prediction accuracy of single BP Neural Network is not ideal and the selection of initial weight and threshold is random,etc,a set of optimal solutions are given by using GA algorithm which can globally optimize,and the accuracy and convergence speed of BP Neural Network optimized by GA are effectively improved.(4)The multi-layer decomposition VMD-GA-BP model is trained and verified by using the monthly average monitoring data of Manganese in the study river from 1998 to2016,and its prediction results are compared with the traditional BPNN model,the time series GA-BP model with single index input,the non-linear GA-BP model considering multiple indexes,and the single-layer decomposition VMD-GA-BP model.Through the Mean Absolute Percentage Error(MAPE),Root mean square error(RMSE)and Mean Absolute Error(MAE)are used to evaluate the prediction effect of the model.The experimental results show that compared with the above models,the MAPE values of the multilayer decomposition VMD-GA-BP model are reduced by 48.52%,66.49%,35.93% and 31.06% respectively,the RMSE values are reduced by 58.38%,64.08%,36.42% and 27.08% respectively,and the MAE values are reduced by 51.63%,62.7%,39.94% and 32.62% respectively,indicating that the prediction results of the multilayer decomposition VMD-GA-BP model proposed in this paper are more accurate and effective.This method can not only accurately predict the concentration of Manganese in surface water,but also has great significance for the early warning and prevention of Manganese pollution in areas with lack of data,and can provide a reference for the prediction of the content of other pollutants in water.