摘要
快速低成本评估污染场地的环境风险对于场地污染详查和污染治理具有重要意义.本研究基于场地污染源、迁移途径等业务数据,改进了LightGBM的特征选择方法,构建了风险筛查模型,并以广州制造业及电力、热力、燃气及水生产和供应业场地作为典型案例进行应用评估.结果表明:(1)基于因果增强改进的LightGBM的特征选择方法能够有效识别场地污染关键因子,减少特征维度的同时提升模型预测精度;(2)分别基于不同数量的关键指标,构建了初始LightGBM与因果增强改进的LightGBM模型.在使用44个关键指标时,模型达到最高的预准确度(91.17%),表明特征选择与模型优化相结合能够显著提高风险筛查分值的预测能力;(3)88个典型场地中,模型成功识别出79个风险场地,准确率达89.8%,实际应用效果良好.本研究改进后的风险筛查分值模型在污染场地风险评估中具有较高的预测精度,能够为场地污染快速识别、污染治理优先级提供决策支持.
Rapid and cost-effective risk assessment of contaminated sites is essential for detailed site investigations and effective remediation planning.In this study,we propose an enhanced feature selection method for Light Gradient Boosting Machine(LightGBM)by integrating business data on pollution sources,migration pathways,and related site characteristics,and construct a risk screening model.The model was further validated through typical case studies of manufacturing and utility sites(electricity,heat,gas,and water production and supply)in Guangzhou.Results indicated that:①The causally enhanced LightGBM feature selection approach can effectively identify key factors contributing to site contamination,reducing feature dimensionality while improving the model's prediction accuracy;②When constructed with different sets of key indicators,the causally enhanced LightGBM consistently outperformed the baseline model,achieving its best accuracy with 44 key indicators(91.17%),demonstrating that combining feature selection with model optimization can substantially improve risk screening performance;③In the evaluation of 88 representative sites,the model successfully identified 79 risky sites,with an accuracy of 89.8%,indicating strong practical applicability.Overall,the proposed model achieves high predictive accuracy for contaminated site risk assessment and provides valuable decision support for rapid site identification of priority sites and the prioritization of remediation efforts.
作者
陈蓉
吴志杰
刘娜
何炜琪
CHEN Rong;WU Zhijie;LIU Na;HE Weiqi(Scientific Research Center for Environmental Big Data,Research Institute for Environmental Innovation(Suzhou)Tsinghua,Suzhou 215163;Suzhou Data Innovation and Application Laboratory,Research Institute for Environmental Innovation(Suzhou)Tsinghua,Suzhou 215163)
出处
《环境科学学报》
北大核心
2026年第1期499-506,共8页
Acta Scientiae Circumstantiae
基金
国家重点研发计划(No.2024YFC3711803)。
