Early termination strategies with asynchronous parallel optimization in application to automatic calibration of groundwater PDE models. (January 2022)

• Record Type: Journal Article
• Authors: Pang, Min; Shoemaker, Christine Ann; Bindel, David
• Is Part Of: Environmental modelling & software. Volume 147(2022)
• Publication Date: 2022-01
• Subjects: Asynchronous parallel -- Knowledge-based early-truncation -- Surrogate-based optimization -- Automatic calibration -- Computationally expensive groundwater model; 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.2021.105237
• Languages: English
• ISSNs: 1364-8152

Abstract

Abstract: Automatic calibration is widely used to estimate parameters in hydrological models. The main idea is to use optimization algorithms to minimize the discrepancy between field data and simulation prediction. This process involves iterative exchanges of a parameter set chosen by the optimization and simulation. However, for computationally expensive models, such as groundwater flow and transport models, the calibration process may be extremely computationally demanding. In this study, we introduce and demonstrate a new asynchronous parallel surrogate-assisted optimization algorithm with an early truncation feature and different knowledge extraction strategies (SO-AET-k). Results show that our asynchronous algorithm performs significantly better than the synchronous analogue (SO-SP), requiring only 40%–70% of the computation time to achieve the same averaged results. In addition, various knowledge extraction strategies of the asynchronous SO-AET-k are tested. In comparisons with two other algorithms, SCE-UA and APPSPACK, asynchronous SO-AET-k shows significantly better performance in terms of both efficiency and robustness. Highlights: Asynchronous parallel with dynamic early truncation strategies are proposed to solve calibration problems. Knowledge extraction rules used in optimization are investigated to exploit and explore the search of optimal solution. The reduction of computational time in finding optimal solution is demonstrated by early stopping expensive objective.