面向多源数据的AQP区域大气污染精准溯源研究  

Accurate traceability of air pollution in AQP area for multi-source data

作　　者：廖世凯 董红召[1] 杨强 夏阳 林盈盈[1] LIAO Shikai;DONG Hongzhao;YANG Qiang;XIA Yang;LIN Yingying(Joint Institute of Intelligent Transportation(Smart City),Zhejiang University of Technology,Hangzhou 310014;Hangzhou Huanyan Technology Co.,Ltd.,Hangzhou 311122;Hangzhou Institute of Ecological and Environmental Sciences,Hangzhou 310005)

机构地区：[1]浙江工业大学,智能交通(智慧城市)联合研究所,杭州310014 [2]杭州环研科技有限公司,杭州311122 [3]杭州市生态环境科学研究院,杭州310005

出　　处：《环境科学学报》2023年第4期131-141,共11页Acta Scientiae Circumstantiae

基　　金：杭州市科技发展计划项目(No.20201203B158);浙江省公益技术研究项目(No.LGF20F030001)。

摘　　要：空气质量优先保障(AQP, Air Quality Priority)区域包括城市居住区和亚运会、世博会等重大活动的举办地等.为了解决AQP区域PM_(2.5)浓度超标而无法准确追溯污染源头的问题,提出一种新的面向多源数据的AQP区域大气污染精准溯源方法 .由于直接溯源方法无法量化不同污染源区域对复合污染的贡献大小,建立一种融合气象数据、污染源区域排放数据、污染物浓度数据的机器学习模型LightGBM-PSO,以捕捉污染源排放和大气污染物浓度之间的非线性响应,模型输出AQP及相邻区域各污染源监测区域排放输入的特征重要程度(FI),由此可得各污染源区域的贡献度排名,并通过贡献度划分污染源站点等级,结合不同等级污染源站点的空间分布确定溯源结果.以2022年1月1日—4月15日杭州市滨江区、上城区、西湖区和萧山区的气象数据、污染源区域排放数据和大气污染物数据进行实验验证.结果表明:相比于贝叶斯优化算法,PSO对LightGBM模型超参数具有更好的优化效果,分别在RMSE、MAE和R^(2)指标上高出7%、3%和3%;相比于SVR、LSTM和CNNLSTM模型,提出的LightGBM-PSO模型具有更高的预测精度和稳定性;萧山区、上城区、西湖区对AQP区域PM_(2.5)污染的贡献比例分别为48%、24%、12%;AQP区域内的(双)江陵路/滨盛路站监测范围内存在污染排放源,外部污染排放源位于其北部和东南方.结合巡查人员历史污染事件统计和现场污染监控可知智能大气污染溯源结果具有可靠性,可取代低效的人工巡查方法 .AQP(Air Quality Priority)areas include urban residential areas and venues for major events such as the Asian Games and World Expo.The source of PM_(2.5)pollution in AQP region cannot be traced accurately,in this way,a new method for accurate traceability of air pollution in AQP region with multi-source data was proposed.Since the direct traceability method cannot quantify the contribution of different sources to the composite pollution,a machine learning model LightGBM-PSO,which integrated meteorological data,source area emission data and pollutant concentration data,was developed to capture the nonlinear response between source emissions and air pollutant concentrations.In particular,the model outputted the FI(Feature Importance)of the emission characteristics of the sources which were consisted of the emission monitoring data in AQP and its adjacent areas.We divided the rank of emission monitoring sites according to FI.Finally,we determined the location of pollution sources according to their spatial distribution.Meteorological data,emission data and air pollutant data of Binjiang,Shangcheng,Xihu and Xiaoshan districts of Hangzhou from January 1 to April 15,2022,were used for experimental validation.The results show that PSO has better optimization effect on the hyperparameters of LightGBM model with 7%,3%and 3%improvement in RMSE,MAE and R^(2) indexes,compared with the Bayesian optimization algorithm;LightGBMPSO model has higher prediction accuracy and stability compared with SVR,LSTM and CNN-LSTM models.Xiaoshan,Shangcheng and Xihu districts contributed 48%,24%and 12%respectively to the AQP regional PM_(2.5)pollution.There are pollution emission sources within the monitoring area of the(double)Jiangling Road/Binsheng Road Station in the AQP area,and the external sources are located to the north and southeast of it.The air pollution traceability results are reliable and can replace the inefficient manual inspection methods by combining the historical pollution event statistics of inspectors and on-site polluti

关 键 词：大气污染溯源 AQP区域 大气污染防治 LightGBM PSO 

分 类 号：X51[环境科学与工程—环境工程]