基于机器学习模型的水质预测评估

Evaluation of Water Quality Prediction Based on Machine Learning Models

摘要: 随着城市化、工业化推进及人类活动强度增加，河流水环境问题日益突出，水质污染呈现多源性、时空异质性和复杂驱动性，水质预测成为水环境管理的重要需求。本研究为分析与预测主要地表水水质指标的时空分布特征，构建了随机森林(RF)、极端梯度提升(XGBoost)、决策树(DT)和自适应增强(AdaBoost)四种机器学习模型，并以决定系数(R²)、均方根误差(RMSE)和平均绝对误差(MAE)作为评估指标。结果显示，四种模型的预测性能排序为RF > XGBoost > DT > AdaBoost，其中RF模型表现最优，其R²、RMSE、MAE分别为0.985、0.042、0.024，具备极强的预测精度、稳定性及泛化能力，能有效捕捉水质指标间的复杂非线性关系。研究表明集成学习模型在处理非线性关系和抑制过拟合方面优势显著，RF模型可作为水质空间预测分析的最优模型，为精准开展水质动态模拟与水环境管理提供支撑。

Abstract: With the advancement of urbanization and industrialization and the intensification of human activities, river water environments have been increasingly threatened. Water quality pollution is characterized by multiple sources, pronounced spatiotemporal heterogeneity, and complex driving mechanisms, making water quality prediction a critical requirement for effective water environment management. To analyze and predict the spatiotemporal distribution patterns of key surface water quality indicators, this study developed four machine learning models, including Random Forest (RF), Extreme Gradient Boosting (XGBoost), Decision Tree (DT), and Adaptive Boosting (AdaBoost). The coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE) were employed to evaluate model performance. The results indicate that the predictive performance of the four models followed the order: RF > XGBoost > DT > AdaBoost. Among them, the RF model exhibited the best performance, with R², RMSE, and MAE values of 0.985, 0.042, and 0.024, respectively, demonstrating superior prediction accuracy, stability, and generalization ability. The RF model effectively captured the complex nonlinear relationships among water quality indicators. The findings highlight the advantages of ensemble learning models in handling nonlinear processes and mitigating overfitting, suggesting that the RF model represents an optimal approach for spatial prediction and assessment of river water quality, thereby providing robust support for dynamic water quality simulation and water environment management.