Cross-Phase Knowledge Transfer via Variable Space Alignment for Expensive Incremental Optimization
针对昂贵增量优化问题中多阶段设计难以复用历史解的问题，提出一种基于变量空间对齐的代理辅助差分进化算法，通过Wasserstein距离筛选高质量历史解并对齐变量空间，加速当前阶段优化收敛。
One main feature of expensive incremental optimization problems (EIOPs) is the involvement of multiple costly related design phases. Based on the characteristics of EIOPs, solutions from previous design phases can be beneficial to find the optimum for the present design phase. However, extracting and reusing previous solutions is still a challenging task when solving EIOPs. To this end, this work proposes a surrogate-assisted differential evolution with variable space alignment (SADE-VSA) to effectively solve EIOPs. In SADE-VSA, a surrogate-assisted transfer strategy based on Wasserstein distance (SA-TWD) is first proposed to measure the similarity between different design phases. This strategy enables the pre-screening of high-quality solutions from previous design phases that exhibit high correlation with the present phase. Second, a surrogate-assisted variable space alignment (SA-VSA) strategy is proposed to align the variable space between previous and present design phases. Following this alignment, the highest quality solution is selected for transfer, which can help speed up the convergence of the algorithm. In order to verify the performance of the proposed SADE-VSA in solving EIOPs, six state-of-the-art methods are compared. The experimental results confirm the effectiveness of the proposed SADE-VSA in solving EIOPs.