Stochastic management of pump-and-treat strategies using surrogate functions

Typical pump-and-treat (PAT) optimization problems involve design of pumping schemes, while minimizing cost and meeting a set of constraints. Due to scarcity of information about the hydrogeological system, stochastic modeling approaches can be used to assess tradeoffs between optimality and reliability. Using a stochastic approach, the constrained, single-objective problem may be turned into a multiobjective problem by substituting constraint inequalities with an additional objective function (OF) that accounts for the reliability of the PAT process. In this work, two approaches are analyzed: in one case, the additional OF consists of the probability of failure of a given remediation policy; in another, the OF additional is represented by the recourse, namely the penalty cost induced by the violation of constraints. In order to overcome the overwhelming computational cost required by stochastic simulation, surrogate forms of the OFs are introduced. In the test case under investigation, such functions are estimated by a kriging interpolation of the OF over a series of data points obtained from stochastic simulations of flow and transport, and calibrated against stochastic optimization solutions. The analysis of the two approaches for addressing the tradeoff of cost vs. reliability indicates that recourse accounts not only for the frequency of constraint violations, as the probability of failure does, but also for the intensity with which these occur. Ultimately, the recourse method allows considering less restrictive policies, although these may be highly sensitive to the choice of penalty functions.