Abstract: In order to solve the black-box problem which is commonly existed in engineering applications such as robots, an efficient and stable hybrid global optimization (HGO) algorithm based on genetic algorithm, non-uniform Kriging metamodel and gradient projection method is proposed. In the proposed algorithm, non-uniform Kriging metamodel is used to evaluate the objective function, which can ensure the accuracy of the optimization process without demanding the global accuracy of the approximate model and save a lot of computation. Moreover, gradient projection method is used to mutate the population of genetic algorithm, which can improve the convergence efficiency of optimization and ensure the optimization constraints to avoid using the non-strict penalty function method to deal with constraints. To validate its effectiveness and superiority, the proposed algorithm is applied to two mathematical test examples and a modular manipulator optimization example, then compared with other optimization algorithms. The results show that the proposed algorithm can balance the accuracy of the results, the optimization efficiency and the stability of the algorithm to achieve a better comprehensive performance, so as to achieve a global optimization design for engineering problems.

Key words: black-box problem, genetic algorithm, global optimization, gradient projection method, Kriging metamodel