Modeling Water Quality Prediction for Aquaculture Waste water Treatment by Isolation Forest Algorithm-Optimized XGBoost
In order to solve the data distortion problem in the process of water quality soft measurement,this study adopted the isolated forest(IF)algorithm to process the abnormal value in the online monitoring data of water quality sensors,optimized the selection of model variables using recursive feature elimination(RFE).XGBoost algorithm was used to construct the water quality prediction model for predicting chemical oxygen demand(CODCr),total phosphorus(TP)and total nitrogen(TN)in the tailwater effluent of the treated farmed fish ponds.The experiments showed that the water quality prediction model for CODCr,TN and TP of the bio-purification pond constructed by the XGBoost algorithm had good prediction performance,and the coefficient of determination(R2)of each model reached 0.837,0.804 and 0.878,respectively,the MAE was 0.679,0.087 and 0.036,and the RMSE was 0.700,0.105 and 0.044,respectively.Meanwhile,after using IF algorithm to identify and remove the outliers of the collected data,the R2 of the model was improved to 0.875,0.866 and 0.926,the MAE decreased to 0.658,0.077 and 0.028,and the RMSE decreased to 0.681,0.099 and 0.035.This study has an important guiding value for the development of intelligent soft sensing technology of water quality.
machine learningisolation forestabnormal value detectionaquaculture wastewaterwater quality prediction
万方数据
维普
