基于优化-自适应稀疏网格替代模型的地下水模拟参数不确定性分析

Uncertainty Analysis of Groundwater Simulation Parameters Based on Optimized-Adaptive Sparse Grid Surrogate Model

地下水模型常用的参数识别方法通常需要多次调用原始模型,从而导致计算负荷问题。替代模型是常用来解决模型计算负荷的有效方法,其中稀疏网格(SG)替代技术已被广泛应用。为了进一步提高稀疏网格替代模型的效率,将斥力粒子群优化算法(RPSO)和维数-局部自适应稀疏网格(DA-LA-SG)结合,提出了优化-自适应稀疏网格(O-DA-LA-SG)替代模型技术。基于一个解析案例和纳米颗粒物运移数值案例,对提出的O-DA-LA-SG替代模型技术进行验证,并利用马尔科夫链蒙特卡洛(MCMC)方法识别纳米颗粒物运移模型参数的后验分布。结果表明:在纳米颗粒物运移数值案例中,O-DA-LA-SG的优化过程总共节省了1598个网格节点,占DA-LA-SG插值节点总数的24.6%;通过MCMC方法直接调用替代模型,减少了参数的不确定性。

Commonly used parameter identification methods of groundwater models often need to run the original model many times,which brings the problem of calculation burden.The surrogate models are commonly used to solve the problem of calculation burden effectively,in which the sparse grid(SG)surrogate technique has been widely used.In order to further improve the efficiency of the sparse grid surrogate model,an optimized-adaptive sparse grid(O-DA-LA-SG)surrogate model technique is proposed by combining repulsive particle swarm optimization(RPSO)and dimension adaptive-local adaptive sparse grid(DA-LA-SG).The proposed O-DA-LA-SG surrogate model technique is verified by an analytical case and a numerical case of nanoparticle transport,and the Markov chain Monte Carlo method(MCMC)is used to identify the parameters of the nanoparticle transport model.The results show that in the numerical case of nanoparticle transport,the optimization process of O-DA-LA-SG saves a total of 1598 grid nodes,accounting for 24.6%of the total number of interpolation nodes of DA-LA-SG;Thus,the uncertainty of parameters is reduced by using the MCMC method to run the surrogate model.