基于贝叶斯-微分进化算法的污染源识别反问题

针对污染源瞬时排放的河流水污染事件反问题,通过贝叶斯统计方法和二维水质对流-扩散方程,建立水体污染识别模型,得到关于污染源强度、污染源位置和污染源排放时刻3个未知参数的后验概率密度函数。运用最大似然估计的思想,采用微分进化算法,求解使后验概率密度函数达到最大值的参数,作为模型未知参数的估计值。算例表明:运用贝叶斯微分进化算法,3个未知参数估计值迭代50次时可以达到稳定,当迭代次数达到280次时,可与真值完全重合;与贝叶斯-蒙特卡洛法相比,贝叶斯-微分进化算法可使3个未知参数估计值达到稳定时的迭代次数降低97.5%,均值误差分别减少1.69%、2.12%和4.03%,具有收敛快、精度高的特点。

Inverse problems of pollution source identification based on Bayesian-DE

Aiming at the inverse problem of water pollution of river with pollutant source discharged instantaneously, a methodical model was constructed based on Bayesian statistical method and two-dimensional water quality model. The posterior probability distribution of unknown parameters including source's position, intensity and discharging time was deduced. The parameters made the posterior probability density function reach the maximum value by the idea of maximum likelihood estimate and differential evolution algorithm(DE), which were viewed as the estimates of model parameters. The example showed that three estimate parameters could reach the stable state based on Bayesian-DE after 50 iterations, and correspond with truth values after 280 iterations. Compared with Bayesian-Markov chain Monte Carlo simulation(Bayesian-MCMC), 97.5% iterations of three estimate parameters reaching the stable state could be reduced, and the mean errors were decreased by 1.69%, 2.12% and 4.03% through the use of Bayesian-DE featuring rapid convergence and high accuracy.