不确定性条件下地下水污染监测井网优化设计——基于XGBoost替代模型

在地下水污染监测井网优化设计中,应用模拟优化方法时客观存在的模型参数不确定性往往会影响到设计监测井网的可靠性.针对该问题,重点考虑了渗透系数和污染源释放强度的不确定性,应用模拟优化方法和蒙特卡罗方法求解上述不确定性参数影响下的最优监测井布设方案.为缓解蒙特卡罗方法多次调用模拟模型所产生的巨大计算负荷,本研究建立了XGBoost替代模型,代替模拟模型与优化模型进行耦合.为提高监测井网对实际污染羽的监测精度,污染监测井网优化模型以监测空间矩误差极小化为优化目标.此外,本次研究还考虑了监测井网设计中污染源释放强度的动态变化过程.最后,以抚顺市某煤矸石堆放场地为基础建立假想例子,验证所提方法的有效性.结果表明:1.XGBoost能够有效近似模拟模型的输入输出关系,显著降低了计算负荷.2.空间矩能够有效评估监测井网插值污染羽和实际污染羽的逼近程度,优化设计后的监测井网能够较为准确地捕捉到实际污染羽的状态.3.模拟优化方法结合蒙特卡罗方法能有效求解不确定性条件下最优监测井网的设计问题.本文为地下水污染监测井网设计提供了一种稳定可靠的方法.

Optimal design of groundwater pollution monitoring network under uncertainty

When applying the simulation-optimization method, objective parameter uncertainty will usually affect the reliability of the design result of groundwater pollution monitoring network. For this problem, the study simultaneously considered the uncertainty of hydraulic conductivity and emission intensity of pollution source, applying Monte Carlo method to design the optimal monitoring wells scheme under the influence of model uncertainty. But Monte Carlo method need to invoke the simulation model many times which will cause a huge amount of calculation. To reduce the calculation load, the study proposed to use Extreme Gradient Boosting (XGBoost) method to construct the surrogate model replacing the simulation model to couple the optimization model in the optimal design of GPMN. To sufficiently improve the monitoring precision of GPMN, the optimization model applied error of spatial moment as objective function. Besides, the dynamic change of emission intensity of pollution source was also considered. Finally, we proposed a hypothetical example based on a coal gangue pile in Fushun City to verify the validity of the method. The results are demonstrated: 1.the XGBoost surrogate method can fit the input-output relationship of the simulation model to a high degree with less computation. 2.spatial moment can effectively assess the approximation degree between interpolation pollution plume of GPMN and actual pollution plume, through which the optimized monitoring network can accurately depict actual pollution plume 3.the simulation-optimization method combines Monte Carlo method can solve the problem of the design of GPMN under uncertainty. In conclusion, the paper provides a stable and reliable method for the design of GPMN.