Prediction Of Water Quality Index In Lam Tsuen River Based On Precision Weighted FA-Stacking

In recent years,emerging technologies such as the Internet of Things(Io T)and artificial intelligence(AI)have been widely applied in various industries.In this context,China’s water conservancy construction is developing towards informationization and automation.However,with the rapid development of the economy,water pollution has become an increasingly serious problem,which hinders the construction of China’s ecological civilization.Water Quality Index(WQI)prediction is the core of water quality field construction,and it is important to accurately and efficiently issue alerts for water resource pollution and anomalies,and to improve the corresponding water resource protection capabilities.Traditional WQI calculation is time-consuming,and when parameters are missing,it may not produce correct results,resulting in unnecessary economic losses.Therefore,using non-physical methods to predict WQI accurately is a current research focus.This study combines machine learning to conduct WQI prediction research and the construction of a water quality management prediction system,which can quickly and accurately conduct water quality evaluation,solve the problems of complex modeling,long training cycles,and difficult practical application in traditional water quality prediction field,and provide timely and efficient technical support for water pollution control decision-making.The research work of this paper is as follows:First,feature engineering was conducted on the water quality index dataset of the Lam Tsuen River in Hong Kong to complete data cleaning.In the feature selection stage,the sensitivity of the input feature column was sorted based on the idea of fake nearest neighbor points combined with Gamma Test(GT criterion).Eleven different input combinations were constructed based on the size of the sensitivity value,and six basic learners,including Support Vector Regression(SVR),Random Forest(RF),Extreme Tree Regression(ETR),Gradient Boosting Decision Tree(GBDT),Extreme Gradient Boosting(XGBoost),and Multilayer Perceptron Regression(MLPRegressor),were used to select the optimal input combination for subsequent algorithm research.Secondly,in order to improve the prediction accuracy of the Stacking model on the regression problem,this paper optimizes the Stacking model in the traditional K-fold cross-verification.When the training set of the second layer meta-learner is generated,the traditional Stacking model uses the average value of K children of the base learner,but ignores the accuracy of the children.To this end,this paper optimized the Stacking strategy and proposed a stacking model based on learner precision weighting.Then the firefly algorithm(FA)is used to select the optimal values of the key parameters of the 6 base learning algorithms selected in this study.In the process of building the improved Stacking model,the optimal combination of the first base learning tools(SVR,ETR,and XGBoost)is selected using the iterative strategy.The second layer of meta-learner uses RF algorithm with strong nonlinear fitting ability to construct the final precision weighted FA-Stacking model.Based on the data set of Lam Tsuen River,the evaluation index analysis with six optimized base learning algorithms and the traditional FA-Stacking model shows that the precision weighted FA-Stacking model has the optimal performance R~2=0.987,RMSE=4.30,MAE=2.21.Finally,the water quality management forecasting system of Lam Tsuen River monitoring station is implemented.In the process of system design,I completed the system architecture design,physical architecture design and functional design from the perspective of system security,stability and subsequent expansion.The overall function of the system is divided into hardware management function,water quality sampling function,water quality prediction function,system management function and soft gateway five modules.