Genetic algorithm based global optimization algorithm used to solving System Analysis of Multi-Disciplinary Optimization

摘要

Multi-Disciplinary Feasible (MDF) is a promising solving architecture for Multi-Disciplinary Optimization (MDO) problem. But traditional iteration based solving method for System Analysis (SA) of MDF has poor computational performance. The solving trouble due to SA prevents its application in complex, tight-coupled system design. The intent of this research is to improve the computational performance of SA and MDF. By analyzing the characteristics of FPI and NRI method, the reasons of computational difficulties were summarized. Then, the requirements to SA solving method were presented. To meet these requirements, original formulation of SA was changed to a Non-Linear Programming (NLP) problem. The special requirements for this NLP necessitate a new optimization algorithm. Hence, two optimization algorithms, GA and DFP, were combined in series to GA-DFP. By a test example, GA-DFP has been validated in capabilities of global search and local convergence, and meets all requirements of SA. By introducing GA-DFP into MDF, a new architecture, BO-MDF, was established. The results of typical problem show that BO-MDF has better performance than other MDO solving architectures of MDF, IDF, AAO, and CO.