A Novel Adaptive Region-Based Global Optimization Method for High Dimensional Problem

Surrogate models are widely used in simulation-based engineering design and optimization to save the computational cost. In this work, an adaptive region-based global optimization method is suggested. During the sampling process the approach uses hyperrectangles to partition the design space and construct local surrogates according to existing sample points. The sizes of hyperrectangles are adaptively generated by the maximum distance of the centered sample point between other points. The large size of the hyperrectangle indicates that the constructed local surrogate might be low accuracy, and the extended size of the hyperrectangles indicates that new sample points should be sampled in this sub-region. On the other hand, considering the exploration of the design space, an uncertainty predicted by using the Kriging model is integrated with the local surrogate strategy and applied to the global optimization method. Finally, comparative results with several global optimization methods demonstrates that the proposed approach is simple, robust, and efficient.