净水技术2026，Vol.45Issue(3)：14-24,11.DOI:10.15890/j.cnki.jsjs.2026.03.002
基于可解释机器学习的洱海北部流域溶解氧驱动因素分析与改善策略
Explainable Machine Learning-Based Analysis of Driving Factors and Improvement Solutions for DO in Northern Erhai Basin
摘要
Abstract
[Objective]Dissolved oxygen(DO)is a key indicator for evaluating water environmental quality,low DO concentration in rivers disrupts ecological balance and threatens aquatic ecosystem health.This paper models and predicts DO concentrations,identifies the key factors driving its variations,and proposes essential measures for regulating water quality and protecting aquatic ecosystems.[Methods]Based on water quality and quantity monitoring datas from the northern Erhai Lake Basin from 2011 to 2020,this paper constructed an integrated machine learning framework,established and selected the optimal model-light gradient boosting machine model(LightGBM)for predicting changes in DO.Furthermore,this paper employed Shapley additive explanations(SHAP)to quantify the contributions of different features to the variation in DO concentration.[Results]The coefficient of determination(R2)for the LightGBM model improved by 11.2％compared to the baseline model,demonstrating superior predictive performance[root mean square error(RMSE)=0.284 mg/L,mean absolute error(MAE)=0.226 mg/L,R2=0.912].SHAP analysis revealed that flow rates emerged as the dominant influencing factor(35.5％),followed by chemical oxygen demand(COD)(17.2％).DO levels initially increased and then decreased with rising flow rates:moderate flow rates enhanced oxygenation,while excessive flow rates introduced a large amount of oxygen-consuming pollutants,thus reducing DO.Suitable flow rates and low COD were identified as key conditions for maintaining high DO concentrations,a flow rate of 0.2 m3/s was recommended,and achieving the lowest possible COD level was crucial.[Conclusion]In the study areas,both high summer flows and low winter flows(or flow interruption)lead to decreased DO concentration.Regulating water storage and purification using reservoirs and ponds along the basin can stabilize flow rates,reduce runoff pollution,and sustain DO levels.This paper enhances understanding of flow-dependent DO dynamics and provides a scientific basis for ecological conservation in plateau river systems.关键词
轻量梯度提升/夏普利加性解释(SHAP)/溶解氧预测/流量作用/库塘调蓄Key words
light gradient boosting/Shapley additive explanation(SHAP)/DO prediction/flow rate effect/reservoir storage and regulation分类
资源环境引用本文复制引用
袁站站,魏卿,陈沛沛,徐祖信..基于可解释机器学习的洱海北部流域溶解氧驱动因素分析与改善策略[J].净水技术,2026,45(3):14-24,11.基金项目
云南省科技厅顶尖团队项目(202505AT350002-4) （202505AT350002-4）
