数据挖掘在空气质量数值模拟中的应用研究进展  

The Application and Research Progress of Data Mining in Air Quality Numerical Simulation

作　　者：王标 彭瑜 王维[1] 屠星月 杜曙明 张亚青 冉伶 张潇天 WANG Biao;PENG Yu;WANG Wei;TU Xingyue;DU Shuming;ZHANG Yaqing;RAN Ling;ZHANG Xiaotian(Institute of Environmental Information,Chinese Research Academy of Environmental Sciences,Beijing 100012,China;Chongqing Ecological Environmental Monitoring Center,Chongqing 401120,China)

机构地区：[1]中国环境科学研究院环境信息研究所,北京100012 [2]重庆市生态环境监测中心,重庆401120

出　　处：《环境科学研究》2024年第8期1703-1713,共11页Research of Environmental Sciences

基　　金：大气重污染成因与治理攻关项目(No.DQGG2021101)。

摘　　要：近年来,数据挖掘技术被广泛用于优化空气质量数值模拟,以提升模拟精度和效率,但所用方法及其应用效果还有待梳理。因此,本文分类归纳了数据挖掘技术在优化空气质量数值模拟中的主要应用方式、优化效果和计算效率。结果表明:①此类研究根据应用环节分为三类。第一类是模拟结果优化法,对空气质量数值模拟的输出结果进行数据挖掘分析,从而优化预测结果;第二类是模拟过程优化法,利用数据挖掘技术对空气质量数值模拟过程中的输入参数、偏微分方程求解等进行优化;第三类是数据挖掘模型训练法,以数据挖掘技术替代空气质量数值模拟过程,直接预测空气质量。②数据挖掘技术有效提升了数值模型的预测效果和运行效率,尤其模拟结果优化法中的机器学习方法可实现PM_(2.5)、O_(3)预测浓度的RMSE分别降低65%~83%、24%~74%。③尽管数值模型时间复杂度为线性阶,优于数据挖掘,但数值模型的数据预处理量大,且需要计算比研究区域范围更大的嵌套模拟区域,而数据挖掘在矩阵计算、硬件加速方面更有优势,因此数值模型的实际计算开销可能高于数据挖掘。未来数据挖掘技术在空气质量预测模拟中的应用前景广阔,包括集成应用空气质量数值模拟和数据挖掘技术,支撑大气污染防控和决策;此外,还将引入可解释性人工智能技术,解释分析模型预测结果。In recent years,data mining have been widely applied to optimize air quality numerical simulation,with the aim of improving the accuracy and efficiency.However,further clarification and evaluation are needed regarding the optimization methods and their application effects.Therefore,we have classified the optimization methods,summarized predictive performance,and evaluated the computational efficiency of data mining in optimizing air quality numerical simulation.The results indicate that:(1)Previous studies can be categorized into three groups based on the simulation steps they are applied to.The first group is the Simulation Result Optimization method,which involves data mining analysis of the output results from air quality numerical simulation to optimize prediction outcomes.The second group is the Simulation Process Optimization method,which utilizes data mining to optimize input parameters and partial differential equations in the air quality numerical simulation.The third group is Data Mining Model Training method,which replaces the numerical simulation process with data mining to directly predict air quality.(2)Data mining effectively improved the predictive performance and computational efficiency of numerical models,particularly through machine learning method in Simulation Result Optimization,leading to a reduction in RMSE for predicted PM_(2.5) and O_(3) concentration by 65%-83%and 24%-74%,respectively.(3)Although the time complexity of numerical models is linear(better than that of data mining),they require extensive data preprocessing and nested simulation covering a much larger area than the study area.On the other hand,data mining offers advantages in matrix calculations and hardware acceleration,thus,the actual computational costs for data mining may be lower than numerical models.In future applications,integrating air quality numerical simulations with data mining technology holds great potential for supporting air pollution prevention and decision-making processes.Additionally,interpretable artif

关 键 词：空气质量数值模拟 数据挖掘 优化 时间复杂度 

分 类 号：X169[环境科学与工程—环境科学]