A Contribution-Based Decomposition Method for Multidisciplinary Engineering Optimization
Shuiwei Xie, John F. Milthorpe, Warren Smith, Riyu Wei · 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization Conference · 2004
The paper reports a novel decomposition method that we refer to as Contribution-based Decomposition. Based on the contributions of variables to the objective functions, a design problem is decomposed into a number of subproblems so that the variables have similar contributions within a subproblem. To cope with the variation of variables' contributions during the optimization process, a Dynamic Contribution-based Decomposition is applied. A subspace optimization method has been employed to solve decomposed subproblems without explicit coordination. The lack of global search ability of the subspace optimization can be overcome by a random decomposition working in conjunction with the Dynamic Contribution-based Decomposition to maintain compulsory re-decompositions. All these decomposition methods have been tested through a number of numerical examples and engineering design problems. The results indicate that decomposition methods based on contribution information can benefit the computing in terms of convergence stability, reliability, accuracy of final results, and reduction of computing iterations.