Efficient parallel surrogate optimization algorithm and framework with application to parameter calibration of computationally expensive three-dimensional hydrodynamic lake PDE models

Improving the speed of global parallel optimization on PDE models with processor affinity scheduling

“…,P. Prior research on parallel optimization for expensive simulations like PDE mainly focused on improving optimization efficiency by designing faster parallel optimization algorithms that can generate a larger number of effective evaluation points per iteration (to reduce wall clock time [WCT]), and that can solve problems with a smaller number of objective function evaluations (to reduce the total computing time). An example is PODS (parallel optimization with dynamic coordinate search using surrogates; Xia et al, 2021), the efficient parallel surrogate optimization method we used in this study.…”

“…Gaussian process-based and radial basis function-based methods are the two most popular surrogate-based optimization algorithms. Surrogate methods were used to solve various real-world expensive problems, including groundwater problems (Christelis et al, 2018;Mugunthan et al, 2005), lake hydrodynamic problems (Xia et al, 2021), lake water quality problems (Xia & Shoemaker, 2020), aerodynamic regional airliner wing design problem (Sóbester & Forrester, 2014), and air quality modeling problem (Carnevale et al, 2012).…”

“…The global parallel surrogate algorithm used in this study, PODS (Xia et al, 2021), does parallel optimization with dynamic coordinate search scrategy and the use of surrogates. PODS is a parallelized adaption of an earlier serial surrogate method, dynamic coordinate search using response surface models (DYCORS; Regis & Shoemaker, 2013).…”

“…PODS showed superior performance over these prior methods. The impact of memory on speedup and efficiency of PODS was not investigated in the study of Xia et al (2021).…”

“…In this study, two PDE models, PDE-FLOW (from Xia et al, 2021) and PDE-WAQ (from Xia & Shoemaker, 2020), were applied for hydrodynamic and water quality simulation of a tropical reservoir in Singapore, respectively. PDE-FLOW simulates the time-varying spatial distribution of the current (velocity) and temperature distribution of the reservoir by solving a system of shallow equations (including the continuity equation, motion equation, and transport equation) that are derived from the Navier-Stokes equations under the shallow water and the Boussinesq assumptions (Xia et al, 2021). PDE-WAQ simulates the complex water quality substance (e.g., algae, nutrients) behavior by solving a set of advection-diffusion equations (Xia & Shoemaker, 2020).…”