Study Of Prediction And Evaluation Method Based On SVM For Water Quality In Three Gorges Reservoir Area

With the rapid development of China’s economy, water environment has becomeone of the nation’s concerns. So it is of great significance that whether or not theenvironment of relevant waters in the Three Gorges Reservoir Area, the national keywaters, is in good condition. As one of the basic work of water environment informationgovernance, water quality prediction and assessment play a critical role in the treatmentof water environment.On the basis of The Water Environment Risk Assessment and Pre-warningDemonstration Platform Construction of the Three Gorges Reservoir Area, a national“11th Five-Year” water sub-subject, we conduct research related to the prediction andassessment methods of the water quality in the Three Gorges Reservoir Area. We firstintroduce the research status about the prediction and assessment methods of waterquality, and then propose the SVM-based prediction and evaluation models about thewater quality in the Three Gorges Reservoir Area according to the characteristics of thewater quality data therein. We have the following findings:①As for SVM-based modeling, parameter selection is one of the most criticallinks. We optimize the data of SVM-based modeling by genetic algorithm, but thetraditional genetic algorithm tends to encounter prematurely, local convergence and otherproblems. So we propose to optimize the data of SVM-based modeling by adaptivegenetic algorithm (AGA).Through the similation, and compare the parameteroptimization efficiency of the SVM-based modeling by the grid method, traditionalgenetic algorithm, AGA and AGA with varying population size, which proves AGAwith varying population size to be the most efficient.②Due to periodic, volatile and random water quality data missing in some watersof the Three Gorges Reservoir Area, we propose a complex nonlinear water quality dataprediction modeling. We adopt the improved exponential smoothing to supplement thesingle-point missing and continuous multi-point missing of the initial data, predictlinear part of the water quality data by ARIMA, predict the non-linear part of waterquality data by SVM-based modeling, conduct SVM-based parameter optimization byAGA with varying population size, and conclude the final prediction result by addingthe prediction results mentioned above. The experimental analysis shows that theprediction modeling proposed here has higher prediction accuracy than other common modeling.③Due to the complex water quality, multiple pollution source types and the highevaluation index correlation in the Three Gorges Reservoir Area, we proposePCA-SVM water quality assessment modeling. We first conduct data compression andinformation extraction on primary indicator variables by PCA, then remove thecorrelation between these indicator variables, establish the assessment modeling for theextracted PCA components by SVM, and then conduct SVM-based parameteroptimization by AGA with varying population size. The simulation results show that theSVM-based assessment method of the water quality in the Three Gorges Reservoir Areahas higher assessment efficiency.