基于贝叶斯推理与改进的MCMC方法反演地下水污染源释放历史  

Reconstructing the release history of groundwater contamination sources based on the Bayesian inference and improved MCMC method

作　　者：顾文龙[1,2] 卢文喜[1,2] 张宇[1,2] 肖传宁 

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]吉林大学环境与资源学院,吉林长春130021

出　　处：《水利学报》2016年第6期772-779,共8页Journal of Hydraulic Engineering

基　　金：中国地调局项目(1212011140027;12120114027401);吉林大学研究生创新基金项目(2015026)

摘　　要：有效识别地下水污染源信息既是设计合理修复方案的基础,也是依法治污明确责权的依据。本文将污染源反演过程转化为贝叶斯推断过程,并与克里格替代模型相结合,提出了一种反演地下水污染源释放历史的新思路,同时针对求解过程中采用的Metropolis抽样算法提出改进方案。算例结果表明:(1)该方法能够有效识别地下水污染源释放历史,反演结果的平均相对误差为3.45%;(2)在500次迭代条件下,改进的Metropolis算法将反演结果的平均相对误差从57.41%降低至3.45%,有效提高了反演效率与精度;(3)在污染源释放速率有较大差异且存在扰动的条件下,反演结果并未出现大幅偏离或波动的异常,效果稳定。Reconstructing the information of groundwater contamination sources effectively,is not only thefoundation of designing a reasonable remediation project,but also the basis of governing pollution in accor-dance with the law and dividing the responsibility. In this paper,a promising approach was presented,ac-cording to which the recovering approach was considered as a Bayesian approach and combined with Krig-ing surrogate model. In addition,an improvement plan was proposed based on the Metropolis sampling algo-rithm. According to the results：（1）the new method can recover the release history of groundwater contami-nant sources efficiently,whose results’ average relative error is 3.45 %;（2）the improved Metropolis algo-rithm enhances the efficiency and accuracy of the inversion results obviously,which can decrease the aver-age relative error from 57.41 % to 3.45 %,with the condition of 500 iterations;（3）the final results arestable,while the disturbance and difference between magnitude during different periods exist.

关 键 词：污染源反演 贝叶斯推理 替代模型 改进的Metropolis算法 释放历史 

分 类 号：P641[天文地球—地质矿产勘探]