关键词: 协同优化, 自适应, 松弛因子, 震荡, 罚函数

Abstract: To solve the problems such as the low optimization efficiency and accuracy caused by choosing the wrong relaxation factors in collaborative optimization,a new computing strategy of the adaptive relaxation factors based on inconsistency information between system level and disciplines level was presented.The optimization process was divided into three phases: in the preliminary optimization phase,decreasing the differences between system expectations and discipline design values was focused on rapidly decrease|in the shock absorption phase,a strictly decreasing function was used to reduce the shock of the value of relaxation factors and ensure it decreasing gradually|in accelerating convergence phase,system level penalty function was introduced to enhance the consistency and promote the convergence to the global optimal solution.A typical numerical example and the reducer MDO problem were adopted to test the performance of this optimization method.By comparing with the standard collaborative algorithm and constant relaxation collaborative algorithm,the result showed that the method could calculate the relaxation factors adaptively along with the optimization process and eliminate the value shocks in existing dynamic relaxation methods.Meanwhile,the presented approach was not affected by initial point selection and had satisfactory robustness with a high convergence speed.

Key words: collaborative optimization, adaptive, relaxation factor, shock, penalty function