Global Optimization Using Mixed Surrogate Models for Computation Intensive Designs

Abstract

Despite of today’s steady and continuing improvement of computation power, effective use of complex and computational intensive engineering analysis and simulation codes in design optimization remains a challenge. In this work, a new global optimization algorithm, namely Mixed Surrogate Models and Design Space Elimination Search (MSMDSES), is introduced. The approach divides the field of interest into several unimodal regions; identify and rank the regions that likely contain the global minimum; fits a Radial Basis function and Quadratic Response Surface model over each promising region with additional design experiments data points using Latin Hypercube designs; identifies its minimum and removes the processed region; and moves to the next most promising region until all regions are processed and the global optimum is identified. The new algorithm was tested using several benchmark problems for global optimization and compared with several widely used region elimination and space exploration global optimization algorithms, showing reduced computation efforts, robust performance and comparable search accuracy, making the new method an excellent tool for computation intensive, computer analysis/simulation based global design optimization problems.