An Adaptive Dynamic Kriging Surrogate Model for Application to the Optimal Remediation of Contaminated Groundwater

When using the simulation–optimization model to optimize groundwater extraction-treatment schemes, constructing a surrogate model for the numerical simulation model is an effective tool for overcoming the large computational load. However, the construction of a one-shot static-surrogate model has disadvantages, such as a large sample size, low accuracy, and the problem of losing the optimal solution. A construction strategy for a batch locally optimal solution-based adaptive dynamic kriging surrogate model is proposed here and applied to the optimal remediation of contaminated groundwater. First, the preliminary kriging surrogate model is established by the kriging method. Second, the adaptive dynamic kriging surrogate model is updated based on the batch locally optimal solutions method. Finally, when the accuracy of the adaptive dynamic kriging surrogate model reaches the convergence criterion, the update stops to obtain the convergent adaptive kriging surrogate model and the optimal remediation scheme. The results show that the optimal pumping wells based on the convergent adaptive kriging surrogate model are well 5, well 6, and well 9, with a remediation cost of ¥ 44,336.16. All pollutant concentrations meet the limit (6 mg/L). This remediation scheme has better effects and less costs than that based on the preliminary kriging surrogate model. Therefore, the batch locally optimal solution-based convergent adaptive kriging surrogate model can effectively avoid the risk of losing the optimal solution, which is of great importance for improving the computational efficiency and accuracy of solving the simulation–optimization model.