基于数据驱动的湘潭市空气质量时空动态分析与预测

本研究选取湘潭市的雨湖区和岳塘区2020年全年的空气质量监测数据与气象数据作为研究对象，运用统计描述、曼肯德尔趋势检验、动态时间规整、层次聚类、主成分分析与克里金空间插值的方法完成了对污染指标间时空关联性的探索，并构建了LSTM-CNN的深度学习模型对重点污染物进行浓度预测。结果表明大气污染物间存在非常复杂的非线性响应关系，且均具有明显的时间分布与空间分布特征。湘潭市的PM_2.5、NO₂、SO₂、CO与O₃成拮抗作用，PM_2.5与PM₁₀以及O₃与SO₂存在最相似的污染过程，主城区存在机动车排放、工业排放、重点区域排放的三大类污染模式。模型预测的结果表明，基于数据驱动的方法对包含复杂反应关系的数据具有较好的预测精度，最低均方误差低至0.02，相比基于传输与化学反应动力学的数值方法有显著优势。该模型对不同污染物、不同站点的预测准确度差异不大，对数据分布较为特殊的污染物依然有不错表现。分别以单站点7项污染物、单站点12项污染物与气象因子、双站点24项污染物与气象因子作为输入参数，模型精度依次提升，表明在预测过程中，部分参数不能代替完整参数的作用，基于数据驱动的方法能最大程度地保留原始数据的特征。

This study focuses on the correlation between air pollution monitoring indicators and the prediction model that bases on data-driven approaches. Yuhu district and Yuetang district in Xiangtan city are selected as research objects. Air quality data as well as meteorological data from 2020 are included. Mainly use statistical description, Mann Kendall trend test, dynamic time wrapping, hierarchical cluster, principal component analysis and the kriging methods to achieve the exploration of spatio-temporal correlation between pollution indicators. The LSTM – CNN deep learning model is also built to predict the concentration of key pollutants. The results show that there are very complex nonlinear response relationships among air pollutants, and all of them have obvious temporal and spatial distribution characteristics. In Xiangtan city, PM_2.5, NO, SO₂, CO have antagonistic effect with O₃, PM_2.5 versus PM₁₀ and O₃ versus SO₂ have the most similar pollution process. There are three types of pollution modes in main urban areas: motor vehicle emission, industrial emission and key area emission. The prediction results of the model show that the data-driven method has good prediction accuracy for the data containing complex reaction relations, and the lowest mean square error is as low as 0.02, which can greatly reduce the difficulty of the numerical method. There is little difference in the prediction accuracy of the model for different pollutants and different monitoring sites, and it still performs well for pollutants with special data distribution. Respectively in a single site 7 pollutants, single 12 pollutants and meteorological factors, double site 24 pollutants and meteorological factors as input parameters, the model accuracy in ascension. The reduced dimension parameters cannot replace complete parameters, and the data-driven method can preserve the features of the original data’s features to the greatest extent.

[1] LELIEVELD J, KLINGMüLLER K, POZZER A, et al. Effects of fossil fuel and totalanthropogenic emission removal on public health and climate[J]. Proceedings of theNational Academy of Sciences, 2019, 116(15):7192-7197.
[2] CHEN FL, CHEN ZF. Cost of economic growth: Air pollution and healthexpenditure[J]. Science of The Total Environment, 2021, 755.
[3] ZHANG X, OU X, YANG X, et al. Socioeconomic burden of air pollut ion in China:Province-level analysis based on energy economic model[J]. Energy Economics, 2017,(68):478-489.
[4] 王文兴,柴发合,任阵海,等.新中国成立70 年来我国大气污染防治历程、成就与经验[J].环境科学研究,2019,32(10):1621-1635.
[5] 钟美芳,田俊泰,叶代启.“十四五”我国VOCs 排放总量控制方案研究与建议[J].环境影响评价,2021, 43(02):1-6.
[6] 刘双霜.长沙株洲湘潭大气污染防治的相关问题及解决对策[J].皮革制作与环保科技,2020,1(08):33-34.
[7] 周作明,荆国华,徐欣.湘潭市环境空气质量变化趋势分析及对策[J].四川环境,2005,(05):27-29.
[8] 罗岳平,刘孟佳,甘杰,等.长株潭城市环境空气中PM2.5 和O3 质量浓度的相关性研究[J].安全与环境学报,2015,15(04):313-317.
[9] 王跃思,姚利,刘子锐,等.京津冀大气霾污染及控制策略思考[J].中国科学院院刊,2013,28(03): 353-363.
[10] 宁淼,孙亚梅,杨金田.国内外区域大气污染联防联控管理模式分析[J].环境与可持续发展,2012,37(05):11-18.
[11] 汪伟全. 空气污染的跨域合作治理研究— — 以北京地区为例[J]. 公共管理学报,2014,11(01):55-64.
[12] 张新民,范西彩,赵文娟,等.关于O3 和PM2.5 协同控制的一些思考[J].环境影响评价,2021,43(02):25-29.
[13] 陈婉.VOCs 和NOx 协同减排是实现PM2.5 和臭氧协同治理的关键[J].环境经济,2021,(02):43-47.
[14] ZHANG Y, BOCQUET M, MALLET V, et al. Real-time air quality forecasting, partI: History, techniques, and current status[J]. Atmospheric Environment, 2012, (60):632-655.
[15] 卢亚灵,李勃,范朝阳,等.空气质量预测模拟技术演变与发展研究[J].中国环境管理,2021,13(04):84-92.参考文献55
[16] 谭成好,陈昕,赵天良,等.空气质量数值模型的构建及应用研究进展[J].环境监控与预警,2014,6(06):1-7.
[17] LV B, COBOURN W G, BAI Y. Development of nonlinear empirical models toforecast daily PM2.5 and ozone levels in three large Chinese cities[J]. AtmosphericEnvironment, 2016, (147):209-230.
[18] COBOURN W G. An enhanced PM2.5 air quality forecast model based on nonlinearregression and back-trajectory concentrations[J]. Atmospheric Environment, 2010,44(25):3015-3023.
[19] 卢学强,梁雪慧,卢学军.神经网络方法及其在非线性时间序列预测中的应用[J].系统工程理论与实践,1997,(06):98-100.
[20] 梁艳春,王政,周春光.模糊神经网络在时间序列预测中的应用[J].计算机研究与发展,1998,(07):88-92.
[21] 王玉涛,夏靖波,周建常,等.基于神经网络模型的时间序列预测算法及其应用[J].信息与控制,1998,(06):14-18.
[22] 熊生龙.贵阳空气中PM10 浓度的神经网络模拟预测与运用[D].浙江大学,2006.
[23] 张怡文,胡静宜,王冉.基于神经网络的PM2.5 预测模型[J].江苏师范大学学报(自然科学版),2015,33(01):63-65.
[24] 张亚群,王潇,李珉.基于BP 网络的兰州市PM10 的污染预测[J].洛阳理工学院学报(自然科学版),2014,24(03):62-72.
[25] 艾洪福, 石莹. 基于BP 人工神经网络的雾霾天气预测研究[J]. 计算机仿真,2015,32(01):402-405.
[26] 王敏,邹滨,郭宇,等.基于BP 人工神经网络的城市PM2.5 浓度空间预测[J].环境污染与防治,2013,35(09):63-66.
[27] 段大高,赵振东,梁少虎,等. 基于LSTM 的PM2.5 浓度预测模型[J].计算机测量与控制,2019,27(03):215-219.
[28] WEN C, LIU S, YAO X, et al. A novel spatiotemporal convolutional long short -termneural network for air pollution prediction[J]. Science of The Total Environment,2019, (654):1091-1099.
[29] 于伸庭.基于长短期记忆网络和卷积神经网络(LSTM-CNN)的PM2.5 浓度预测研究[D].上海交通大学,2020.
[30] 汤静,王春林,谭浩波,等.利用PCA-KNN 方法改进广州市空气质量模式PM2.5 预报[J].热带气象学报,2019,35(01):125-134.
[31] KUO YH, KUSIAK A. From data to big data in production research: the past and futuretrends[J]. International Journal of Production Research, 2019, 57(15-16):4828-4253.
[32] 王宏志, 梁志宇, 李建中, 等. 工业大数据分析综述: 模型与算法[J]. 大数据,2018,4(05):62-79.参考文献56
[33] BYUN D, SCHERE K L. Review of the governing equations, computationalalgorithms, and other components of the models -3 Community Multiscale Air Quality(CMAQ) modeling system[J]. Applied Mechanics Reviews, 2006, 59(1-6):51-77.
[34] 王占山,李晓倩,王宗爽,等.空气质量模型CMAQ 的国内外研究现状[J].环境科学与技术,2013,36(S1):386-391.
[35] 中国信息通信研究院.大数据白皮书(2022 年)[R].北京:中国信息通信研究院, 2020.
[36] 黄珊.数据驱动模型下的郑州大气污染预测[D].华北水利水电大学,2020.
[37] 常丽娜,王颖俐,王瑶.基于K-均值聚类和贝叶斯判别的城市空气质量等级分类及预测[J].太原师范学院学报(自然科学版),2021,20(02):41-46.
[38] 刘琴,葛梅梅.基于聚类分析和因子分析对我国31 个主要城市空气质量的研究[J].西昌学院学报(自然科学版),2021,35(01):62-65.
[39] 陈颖,张仲伍.基于聚类分析和主成分分析的城市空气质量评价——以山西省11 个地级市为例[J].山西师范大学学报(自然科学版),2020,34(04):72-78.
[40] 薛志钢,郝吉明,陈复,等.国外大气污染控制经验[J].重庆环境科学,2003,(11):159-161.
[41] TANG XB, CHEN XH, TIAN Y. Chemical composition and source apportionment ofPM2.5: A case study from one year continuous sampling in the Chang-Zhu-Tan urbanagglomeration[J]. Atmospheric Pollution Research, 2017, 8(5):885-899.
[42] 天津双允环保科技有限公司,湘潭市环境保护监测站,北京南科大蓝色科技有限公司,等.湘潭市秋冬季大气PM2.5 来源解析报告[R].天津:南开大学国家环境保护城市空气颗粒物污染防治重点实验室,2019.
[43] 柏玲,姜磊,刘耀彬.长江中游城市群环境压力的时空特征——以工业SO2 排放为例[J].经济地理,2017,37(03):174-181.
[44] 国家统计局,国务院第七次全国人口普查领导小组办公室.第七次全国人口普查公报（第七号）——城乡人口和流动人口情况[J].中国统计,2021,(05):13.
[45] 章诞武,丛振涛,倪广恒.基于中国气象资料的趋势检验方法对比分析[J].水科学进展,2013,24(04):490-496.
[46] MANN H B. Nonparametric tests against trend[J]. Econometrica, 1945, 13(3):245-259.
[47] 魏凤英.现代气候统计诊断与预测技术[M].北京:气象出版社,1999.
[48] ALYOUSIFI Y, IBRAHIM K, ZIN W Z W, et al. Trend analysis and change pointdetection of air pollution index in Malaysia[J]. International Journal of EnvironmentalScience and Technology, 2021.
[49] JAISWAL A, SAMUEL C, KADABGAON V M. Statistical trend analysis and forecastmodeling of air pollutants[J]. Global Journal of Environmental Science andManagement-Gjesm, 2018, 4(4):427-438.
[50] 鲁凤,钱鹏,胡秀芳,等. 基于小波分析与Mann-Kendall 法的上海市近12 年空气质量变化[J].长江流域资源与环境,2013,22(12):1614-1620.
[51] ALHATHLOUL S H, KHAN A A, MISHRA A K. Trend analysis and change pointdetection of annual and seasonal horizontal visibility trends in Saudi Arabia[J].Theoretical and Applied Climatology, 2021, 144(1-2):127-146.参考文献57
[52] 曲歌.我国空气质量区域联动性及典型地区影响因素探究[D].首都经济贸易大学,2018.
[53] JEONG Y S, JEONG M K, OMITAOMU O A. Weighted dynamic time warping fortime series classification[J]. Pattern Recognition, 2011, 44(9):2231-2340.
[54] 孙吉贵,刘杰,赵连宇.聚类算法研究[J].软件学报,2008,(01):48-61.
[55] 李瑞,李清,徐健,等.秋冬季区域性大气污染过程对长三角北部典型城市的影响[J].环境科学,2020,41(04):1520-1534.
[56] JOLLIFFE I T, CADIMA J. Principal component analysis: a review and recentdevelopments[J]. Philosophical Transactions Of The Royal Society A-MathematicalPhysical and Engineering Sciences, 2016, 374(2065).
[57] HARROU F, KADRI F, KHADRAOUI S, et al. Ozone measurements monitoring usingdata-based approach[J]. Process Safety and Environmental Protection, 2016,(100):220-231.
[58] AZID A, JUAHIR H, EZANI E, et al. Identification Source of Variation on RegionalImpact of Air Quality Pattern Using Chemometric[J]. Aerosol and Air QualityResearch, 2015, 15(4):1545-1558.
[59] 张晓雨.中国中东部地区典型城市大气细颗粒物中化学组成特征及来源解析研究[D].南京大学,2017.
[60] 姜新华,薛河儒,张存厚,等.基于主成分分析的呼和浩特市空气质量影响因素研究[J].安全与环境工程,2016,23(01):75-79.
[61] 孟琛琛,王丽涛,苏捷,等.邯郸市PM2.5 化学组成特征及来源解析[J].环境科学与技术,2016,39(02):57-64.
[62] SHAFIE S H M, MAHMUD M, MOHAMAD S, et al. Influence of urban air pollutionon the population in the Klang Valley, Malaysia: a spatial approach[J]. EcologicalProcesses, 2022, 11(1).
[63] KAMARUZZAMAN A, SAUDI A M, AZID A, et al. Assessment on air quality pattern:a case study in putrajaya, malaysia[J]. Journal of Fundamental and Applied Sciences,2017, (9):789-800.
[64] MISHRA D, GOYAL P. Development of artificial intelligence based NO2 forecastingmodels at Taj Mahal, Agra[J]. Atmospheric Pollution Research, 2015, 6(1):99-106.
[65] YANG RJ, HU XQ, HE LJ, et al. Prediction of Shanghai air quality index based on BPneural network optimized by genetic algorithm[C]. 2020 13th InternationalSymposium on Computational Intelligence and Design (ISCID 2020), 2020.
[66] KUMAR A, GOYAL P. Forecasting of Air Quality Index in Delhi Using NeuralNetwork Based on Principal Component Analysis[J]. Pure and Applied Geophysics,2013, 170(4):711-722.
[67] 李璇,王雪松,刘中,等.宁波人为源VOC 清单及重点工业行业贡献分析[J].环境科学,2014,35(07):2497-2502.参考文献58
[68] 陈优良,李亚倩.长三角PM2.5 和O3 变化特征及与气象要素的关系[J].长江流域资源与环境,2021,30(02):382-396.
[69] 李郅瑾,郝彦斌.多城市PM2.5、PM10、O3 浓度与气象条件关联性的Meta 分析[J].绿色科技,2022,24(02):110-113.
[70] LI K, JACOB D J, LIAO H, et al. Anthropogenic drivers of 2013-2017 trends insummer surface ozone in China[J]. Proceedings of The National Academy of Sciencesof The United States of America, 2019, 116(2):422-427.
[71] 朱珠,谭成好,吴惬,等.大气污染特征研究及臭氧污染个例分析——以深圳市龙华区为例[J].环境保护科学,2020,46(04):80-86.
[72] 冯凝,唐梦雪,李孟林,等.深圳市城区VOCs 对PM2.5 和O3 耦合生成影响研究[J].中国环境科学,2021,41(01):11-17.
[73] 翁佳烽,梁晓媛,邓开强,等.不同季节肇庆市PM2.5 和O3 污染特征及潜在源区分析[J].环境科学研究,2021,34(06):1306-1317.
[74] 刘长焕,邓雪娇,朱彬,等.近10 年中国三大经济区太阳总辐射特征及其与O3、PM2.5的关系[J].中国环境科学,2018,38(08):2820-2829.
[75] 王占山,张大伟,李云婷,等.北京市夏季不同O3 和PM2.5 污染状况研究[J].环境科学,2016,37(03):807-815.
[76] 梅梅,徐大海,朱蓉,等.减排措施与气象因子对2013—2019 年中国大陆地区PM2.5浓度变化的贡献[J].环境科学学报,2021,41(07):2519-2529.
[77] 周亚端,朱宽广,黄凡,等.新冠肺炎疫情期间湖北省大气污染物减排效果评估[J].环境科学与技术,2020,43(03):228-236.
[78] 吴波,刘春琼,张娇,等.COVID-19 期间区域大气高污染发生的非线性动力机制[J].中国环境科学,2021,41(05):2028-2039.
[79] 李贝睿.长株潭区域大气污染物排放源清单研究[D].湘潭大学,2016.
[80] 刘妍月,李军成.长沙市大气中PM2.5 浓度分布的空间插值方法比较[J].环境监测管理与技术,2016,28(02):14-18.
[81] 周体鹏.基于克里金插值法的昆明市PM2.5 预测[D].云南大学, 2016.
[82] ZHOU Y L, CHANG F J, CHANG L C, et al. Explore a deep learning multi -outputneural network for regional multi-step-ahead air quality forecasts[J]. Journal ofCleaner Production, 2019, (209):134-145.
[83] SHEN X, TIAN X M, LIU T L, et al. Continuous Dropout [J]. Ieee Transactions onNeural Networks and Learning Systems, 2018, 29(9):3926-3937.
[84] JIN HF, SONG QQ, HU X, et al. Auto-Keras: An Efficient Neural Architecture SearchSystem[M]. Kdd'19:Proceedings of The 25th Acm Sigkdd International Conferencceon Knowledge Discovery and Data Mining. 2019:1946-1956.
[85] SEBASTIANI F. Machine learning in automated text categorization[J]. AcmComputing Surveys, 2002, 34(1):1-47.
[86] JAIN A K, DUIN R P W, MAO J C. Statistical pattern recognition: A review[J]. Ieee参考文献59Transactions on Pattern Analysis and Machine Intelligence, 2000, 22(1):4-37.
[87] CRAWFORD T M, DUCHON C E. An improved parameterization for estimatingeffective atmospheric emissivity for use in calculating daytime downwelling longwaveradiation[J]. Journal of Applied Meteorology, 1999, 38(4):474-480.
[88] 刘媛媛. 融合CNN-LSTM 和注意力机制的空气质量指数预测[J]. 计算机时代,2022, (01) 58-60.
[89] 刘蕾.基于Tensorflow 的循环神经网络模型在上海市空气质量预测中的应用[D].上海师范大学,2019.