Using Many-Objective Optimization and Robust Decision Making to Identify Robust Regional Water Resource System Plans

Abstract:

Water resource system planning regulations are increasingly requiring potential plans to be robust, i.e., perform well over a wide range of possible future conditions. Robust Decision Making (RDM) has shown success in aiding the development of robust plans under conditions of ‘deep’ uncertainty. Under RDM, decision makers iteratively improve the robustness of a candidate plan (or plans) by quantifying its vulnerabilities to future uncertain inputs and proposing ameliorations. RDM requires planners to have an initial candidate plan. However, if the initial plan is far from robust, it may take several iterations before planners are satisfied with its performance across the wide range of conditions. Identifying an initial candidate plan is further complicated if many possible alternative plans exist and if performance is assessed against multiple conflicting criteria. Planners may benefit from considering a plan that already balances multiple performance criteria and provides some level of robustness before the first RDM iteration. In this study we use many-objective evolutionary optimization to identify promising plans before undertaking RDM. This is done for a very large regional planning problem spanning the service area of four major water utilities in East England. The five-objective optimization is performed under an ensemble of twelve uncertainty scenarios to ensure the Pareto-approximate plans exhibit an initial level of robustness. New supply interventions include two reservoirs, one aquifer recharge and recovery scheme, two transfers from an existing reservoir, five reuse and five desalination schemes. Each option can potentially supply multiple demands at varying capacities resulting in 38 unique decisions. Four candidate portfolios were selected using trade-off visualization with the involved utilities. The performance of these plans was compared under a wider range of possible scenarios. The most balanced plan was then submitted into the vulnerability analysis step of RDM. This strategy was shown to perform significantly better in all metrics at moderate costs as compared to the baseline strategy.