A two-level global optimization method based on hybrid metamodel for expensive problems:

摘要

With the wide application of simulation and optimization tools in engineering problems, how to build a metamodel, which can satisfy the high accuracy requirements of real working conditions and realize fast optimization in the entire design space, becomes a hot issue. On the basis of sequential sampling optimization and updating design space optimization, a two-level global optimization method based on hybrid metamodel is developed. In this method, the hybrid metamodel is constructed using random sampling. In the first level, a space reduction strategy is proposed to reduce the design variable space and guide the search to the promising region. In the second level, Adaptive simulated annealing algorithm is integrated with metamodels to search the global optimal value in the promising region. Several global optimization problems and a real industrial design optimization example are utilized to demonstrate the superior performance of the proposed method.