基于SMOTE-GA-CatBoost算法的全国地表水水质分类评价

National surface water quality classification evaluation based on SMOTE-GA-CatBoost method

针对地表水分类评价中水污染特征空间的高冲突性以及水质类别的不均衡性等问题,以7项地表水水质指标为水质评价因子,采用SMOTE过采样技术结合遗传算法和CatBoost模型对全国主要江河和重要湖库分别进行水质分类评价,并与其他4种改进集成算法进行对比.结果表明:SMOTE预处理有效改善样本类别的不均衡性,提高CatBoost模型对少数类水质样本分类的准确性;遗传算法调参有效提高CatBoost模型的收敛速度和分类精度,优化了模型的分类性能;SMOTE-GA-CatBoost模型对江河和湖库的水质分类效果均优于其他4种改进集成分类模型,其对江河水水质分类的准确率、精确率、召回率、F1分别为97.7%、97.8%、96.1%、96.9%,对湖库水水质分类的准确率、精确率、召回率、F1分别为96.7%、96.2%、95.4%、95.8%,该模型可以实现不同水域的水质分类评价.

Aiming at the problems such as the high conflict of water pollution feature space and the imbalance of water quality categories in surface water classification evaluation,Synthetic Minority Oversampling Technique(SMOTE)which was combined with Genetic Algorithms(GA)and CatBoost model that used seven water quality indexes of surface water as water quality evaluation factors were respectively employed to evaluate the water quality of major rivers and important lakes-reservoirs in the country.The results were compared with the other four improved ensemble algorithms,which showed that the SMOTE pretreatment could effectively enhance the imbalance of sample categories and increase the accuracy of CatBoost model for the classification of minority water quality samples.The genetic algorithm parameters could effectively improve the convergence speed and classification accuracy of CatBoost model and optimize the classification performance of the model.The SMOTE-GA-CatBoost model showed higher performance of water quality classification compared with the other four improved integrated classification models.The values of accuracy,precision,recall and F1of the SMOTE-GA-CatBoost model for river water quality classification reached 97.7%,97.8%,96.1%and 96.9%,respectively.The value of accuracy,precision,recall and F1for water quality classification of lakes-reservoirs water were 96.7%,96.2%,95.4%and 95.8%,respectively.The proposed model could be used to classify and evaluate the water quality of different water areas.