基于替代模型的地下水DNAPLs污染源反演识别

应用基于核极限学习机替代模型的模拟-优化理论和方法研究解决了地下水DNAPLs污染源及含水层参数的同步反演识别问题.结果表明：1）核极限学习机替代模型对模拟模型有较高的逼近精度，能够识别并模仿模拟模型的输入-输出关系，绝大部分相对误差小于5%，平均相对误差仅有2.98%；2）以替代模型代替模拟模型，大幅度地减小了模拟-优化过程的计算负荷，将反演识别时间由传统方法的83天减少到3小时，并能够保持较高的计算精度；3）应用基于模拟退火的粒子群优化算法求解优化模型，能够以较快的速度搜寻到全局最优，同时避免搜索过程陷于局部极小解.

Surrogate-based source identification of DNAPLs-contaminated groundwater

Groundwater contamination source identification (GCSI) is critical for taking effective actions in designing remediation strategies, estimating risks, and confirming responsibility. Surrogate-based simulation-optimization technique was applied to source identification and parameter estimation of DNAPLs-contaminated aquifer in this article. The results showed that:1) kernel extreme learning machines (KELM) surrogate model approximated the simulation model accurately. It could simulate the input/output relationship of the simulation model with most of the relative errors less than 5%, and the mean relative error was only 2.98%; 2) Replacing the simulation model with a KELM model considerably reduced the computational burden of the simulation-optimization process and maintained high computation accuracy, the identification time was reduced to 3hours from 83days; 3) Simulated annealing-based particle swarm optimization algorithm is efficient in searching the global optimal solution of the nonlinear programming optimization model, and avoiding the optimization process trapping into local optimum.