Efficient Global Optimization Using LS-OPT and Its Parallelization
This article presents the implementation of Kriging-based efficient global optimization (EGO) in LS- OPT, which can be used for both unconstrained and constrained optimization. Additionally it proposes a parallelization technique based on a multi-objective formulation that provides multiple sampling choices. Like any surrogate-based method, the proposed method displays some variation in the results depending on the initial DOE. Further investigation is being conducted to reduce this variation, especially for constrained optimization. The methodology for Pareto-based parallel EGO is different for constrained and unconstrained optimization in some respects. Only the methodology and results for unconstrained optimization are presented in this paper. Analytical examples with known global optima are used to demonstrate the efficacy and the efficiency of the algorithm.
https://www.lsoptsupport.com/documents/papers/optimization/2017/efficient-global-optimization-using-ls-opt-and-its-parallelization.pdf/view
https://www.lsoptsupport.com/@@site-logo/LS-Opt-Support-Logo480x80.png
Efficient Global Optimization Using LS-OPT and Its Parallelization
This article presents the implementation of Kriging-based efficient global optimization (EGO) in LS- OPT, which can be used for both unconstrained and constrained optimization. Additionally it proposes a parallelization technique based on a multi-objective formulation that provides multiple sampling choices. Like any surrogate-based method, the proposed method displays some variation in the results depending on the initial DOE. Further investigation is being conducted to reduce this variation, especially for constrained optimization. The methodology for Pareto-based parallel EGO is different for constrained and unconstrained optimization in some respects. Only the methodology and results for unconstrained optimization are presented in this paper. Analytical examples with known global optima are used to demonstrate the efficacy and the efficiency of the algorithm.