基于MCMC方法的城区有毒气体扩散源反演  

Source inversion of toxic gas dispersion in urban areas based on the MCMC method

作　　者：郭少冬[1,2] 杨锐[1] 翁文国[1] 

机构地区：[1]清华大学工程物理系,公共安全研究中心,北京100084 [2]清华大学航天航空学院,工程力学系,北京100084

出　　处：《清华大学学报（自然科学版）》2009年第5期629-634,共6页Journal of Tsinghua University(Science and Technology)

摘　　要：城区中有毒气体突发性泄漏时,需要快速对泄漏源进行定位和识别,以便科学预测气体的蔓延及其影响范围。利用基于Bayes推断理论的MCMC(Markov chain Monte Carlo)抽样方法,根据城市中分布的传感器测量信息和气体扩散数值计算模型,构造似然函数,对泄漏源的位置、强度进行反演。计算了这些参数和空间各点浓度的相关统计量,表明反演结果与泄漏源的真实参数十分吻合。此外,还讨论了传感器测量误差的概率分布对结果的影响。结果表明,误差概率会显著影响计算效果,概率分布越平坦,泄漏源反演信息的不确定度越大。When toxic gas accidentally releases in the urban environment, accurately and rapidly locating and identifying the source is an important issue to predict and confirm the gas dispersion and the affected areas. With the observations of the sensors distributed over the urban areas and the concentrations predicted by an atmospheric dispersion model, a likelihood [unction was assigned, with which Markov chain Monte Carlo （MCMC） sampling based on Bayesian inference was used to invert the parameters, including the source location and the dispersion strength. The probability distributions of the parameters were then calculated which agreed well with the actual results. Analyses show that the probability distribution of the sensor error significantly affects the calculational results and that flatter and broader probability distribution of the sensor error leads to source inversion with larger uncertainty.

关 键 词：有毒气体泄漏 反演 BAYES推断 似然函数 后验概率 

分 类 号：X928.9[环境科学与工程—安全科学]