混合元模型自适应空间探索优化方法及应用

To improve the design quality and optimization efficiency for complicated engineering systems, such as blended wing body underwater gliders, a hybrid metamodel-based adaptive space exploration optimization method is proposed in this paper. Three classical metamodels, including polynomial response surfaces, radial basis functions, and kriging, are constructed using the initial sample points generated by the enhanced stochastic evolutionary method. The hybrid metamodels can be constructed using the weighted superposition of the weight factors of each metamodel based on the predicted root mean square error. The adaptive space exploration method is used to identify the promising design subspace according to the known information. Then, new sample points are selected in this promising design subspace. Thus, the prediction accuracy in the region of interest can be gradually enhanced until the optimization convergence. Based on the testing results using eight benchmark optimization functions and blended wing body underwater glider shape design optimization problem, the newly proposed optimization method shows improved capability in terms of the optimization efficiency and global optimum identification.