Advancing surrogate-based optimization of time-expensive environmental problems through adaptive multi-model search. (April 2023)

• Record Type: Journal Article
• Authors: Tsattalios, Spyridon; Tsoukalas, Ioannis; Dimas, Panagiotis; Kossieris, Panagiotis; Efstratiadis, Andreas; Makropoulos, Christos
• Is Part Of: Environmental modelling & software. Volume 162(2023)
• Publication Date: 2023-04
• Subjects: Surrogate modeling -- Machine learning -- High-dimensional expensive black-box (HEB) problems -- Evolutionary annealing-simplex -- Test functions -- Hydraulic design; Environmental monitoring -- Computer programs -- Periodicals; Ecology -- Computer simulation -- Periodicals; Digital computer simulation -- Periodicals; Computer software -- Periodicals; Environmental Monitoring -- Periodicals; Computer Simulation -- Periodicals; Environnement -- Surveillance -- Logiciels -- Périodiques; Écologie -- Simulation, Méthodes de -- Périodiques; Simulation par ordinateur -- Périodiques; Logiciels -- Périodiques; Computer software; Digital computer simulation; Ecology -- Computer simulation; Environmental monitoring -- Computer programs; Periodicals; Electronic journals; 363.70015118
• DOI: 10.1016/j.envsoft.2023.105639
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: Complex environmental optimization problems often require computationally expensive simulation models to assess candidate solutions. However, the complexity of response surfaces necessitates multiple such assessments and thus renders the search procedure too laborious. Surrogate-based optimization is a powerful approach for accelerating convergence towards promising solutions. Here we introduce the Adaptive Multi-Surrogate Enhanced Evolutionary Annealing Simplex (AMSEEAS) algorithm, as an extension of its precursor SEEAS, which is a single-surrogate-based optimization method. AMSEEAS exploits the strengths of multiple surrogate models that are combined via a roulette-type mechanism, for selecting a specific metamodel to be activated in every iteration. AMSEEAS proves its robustness and efficiency via extensive benchmarking against SEEAS and other state-of-the-art surrogate-based global optimization methods in both theoretical mathematical problems and in a computationally demanding real-world hydraulic design application. The latter seeks for cost-effective sizing of levees along a drainage channel to minimize flood inundation, calculated by the time-expensive hydrodynamic model HEC-RAS. Highlights: AMSEEAS embeds multiple surrogate models that evolve and cooperate as a group. A virtual roulette decides which metamodel is activated in every iteration. The adaptive multimodel approach ensures flexibility against varying geometries. Benchmarking is employed via theoretical functions and a hydraulic design study. In all problems, AMSEEAS outperforms the contrasting global optimization methods.