Abstract
This paper presents a multi-fidelity RBF (radial basis function) surrogate-based optimization framework (MRSO) for computationally expensive multi-modal optimization problems when multi-fidelity (high-fidelity (HF) and low-fidelity (LF)) models are available. The HF model is expensive and accurate while the LF model is cheaper to compute but less accurate. To exploit the correlation between the LF and HF models and improve algorithm efficiency, in MRSO, we first apply the DYCORS (dynamic coordinate search algorithm using response surface) algorithm to search on the LF model and then employ a potential area detection procedure to identify the promising points from the LF model. The promising points serve as the initial start points when we further search for the optimal solution based on the HF model. The performance of MRSO is compared with 6 other surrogate-based optimization methods (4 are using a single-fidelity surrogate and the rest 2 are using multi-fidelity surrogates). The comparisons are conducted on a multi-fidelity optimization test suite containing 10 problems with 10 and 30 dimensions. Besides the benchmark functions, we also apply the proposed algorithm to a practical and computationally expensive capacity planning problem in manufacturing systems which involves discrete event simulations. The experimental results demonstrate that MRSO outperforms all the compared methods.
Highlights
Expensive multi-modal black-box optimization problems arise in many science and engineering areas
We propose MRSO, a multi-fidelity RBF surrogatebased optimization framework that utilizes the original RBF surrogate-assisted dynamic coordinate search algorithm (DYCORS) (Regis and Shoemaker 2013), in which the surrogate is built on single-fidelity models
(2) To link the search on LF and HF models, we innovatively develop a potential area detection procedure to identify the guesses for local optimal locations, which serve as the starting points for the HF run
Summary
Expensive multi-modal black-box optimization problems arise in many science and engineering areas. Where x = (x1, x2, · · · , xd ) is the decision vector, d is the number of decision variables, and xi (1 ≤ i ≤ d) is the ith decision variable. Expensive multimodal black-box optimization problems have the following challenges: (1) Every single evaluation of the objective function is expensive, which may take minutes, hours, or even days. In these cases only a relatively small number of evaluations are possible; (2) no derivative information can be obtained from the problem. (3) The problem may have multiple local minima The gradient-based optimization methods cannot be directly applied. (3) The problem may have multiple local minima
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
