A Comparative Study of SVMs Model Optimized by Machine Learning Methods in Water Quality Assessment of Dongting Lake

This paper presents 15 SVM models optimized by machine learning methods in water quality assessment. Through comparative analysis, we found that all the models reached reasonable accuracy during the training and testing process. In particular, SAA-SVM-PI, GWO-SVM-PI, GWO-SVM-TN, PSO-SVM-TP models have the highest accuracy, the highest squared correlation and the lowest mean squared error when testing the three indexes of PI, TN and TP. In order to study the spatial distribution characteristics of water quality in Dongting Lake, GWO-SVM-PI, GWO-SVM-TN and PSO-SVM-TP models were respectively used for single-factor water quality classification. The results showed that: As far as the PI is concerned, the water quality is mainly in grade I and II; As far as TN is concerned, the water quality is mainly of grade IV and V, with the worst water quality in the East Dongting Lake and the outlet profile; In terms of TP, the water quality is mainly of grade III, IV and V, with the best inlet water quality, the worst water quality in the South Dongting Lake.