不同建议分布MCMC算法在地下水污染源反演识别中的对比研究  

Comparative study of different proposal distribution MCMC algorithms in groundwater pollution source identification

作　　者：李雪利 罗建男[1,2] 刘勇 LI Xue-li;LUO Jian-nan;LIU Yong(Key Laboratory of Groundwater Resources and Environment,Ministry of Education,Jilin University,Changchun 130021,China;College of New Energy and Environment,Jilin University,Changchun 130021,China)

机构地区：[1]吉林大学地下水资源与环境教育部重点实验室,吉林长春130021 [2]吉林大学新能源与环境学院,吉林长春130021

出　　处：《中国环境科学》2023年第4期1646-1654,共9页China Environmental Science

基　　金：国家自然科学基金资助项目(42072279);吉林省教育厅科学研究项目(JJKH20211108KJ)。

摘　　要：为了探究正态分布、均匀分布和随机游走三种不同建议分布的MCMC算法对地下水污染源反演识别结果的影响,并遴选出最优的建议分布.本文根据假想案例建立了污染质运移模拟模型及基于Kriging方法的替代模型,并建立了基于正态分布、均匀分布和随机游走三种建议分布的MCMC算法以反演识别地下水污染源的释放历史.结果表明,针对本文算例,以均匀分布为建议分布的MCMC算法具有反演精度高、稳定性好、收敛速度快等优点,最为适合作为本研究案例的建议分布进行地下水污染源反演识别研究.In order to obtain the optimal proposal distribution by exploring the influences of normal distribution,uniform distribution and random walk of the MCMC algorithms on the inversion identification results of groundwater pollution source,this paper established a simulation model of pollutant transport and a surrogate model using Kriging method with a hypothetical case,and the MCMC algorithm based on normal distribution,uniform distribution and random walk was developed to identify the release history of groundwater pollution sources.The results show that the inversion algorithm with uniform distribution as the proposal distribution has the advantages of high inversion accuracy,good stability and fast convergence speed,which proves to be the most suitable proposal distribution for groundwater pollution sources inversion.

关 键 词：地下水污染源识别 贝叶斯理论 MCMC算法 建议分布 

分 类 号：X523[环境科学与工程—环境工程]