Hydrosystem optimization on a budget: Investigating the potential of surrogate based optimization techniques

Development of uncertainty-aware operational rules for multi-reservoir systems is a demanding and challenging task due to the complexity of the system dynamics, the number of decision variables and the hydrological uncertainty. In order to overcome this issue the parsimonious parameterization-simulation-optimization (PSO) framework is employed coupled with stochastically generated hydrological time-series. However, when the simulation model requires long computational time this coupling imposes a computational barrier to the framework. The purpose of this paper is threefold: a) Investigate the potential of Efficient Global Optimization (EGO) algorithm (and its variants) which is capable of reaching global optima within a few simulation model evaluations (~500 or less). b) Extend the capabilities of WEAP21 water resources management model by using it within PSO framework (named WEAP21-PSO) and c) Validate and compare the results of WEAP21-PSO using the well-known hydrosystem management model Hydronomeas coupled with Evolutionary Annealing Simplex (EAS) optimization algorithm. Results confirm that EGO has the potential and the capabilities to handle computationally demanding problems and furthermore is capable of locating the optimal solution within few simulation model evaluations and that the WEAP21-PSO framework performs well at the task at hand.