全局与局部模型交替辅助的差分进化算法

为求解实际复杂工程应用中的高维计算费时优化问题，提出一种全局与局部代理模型交替辅助的差分进化算法。利用历史样本训练全局和局部代理模型，通过交替搜索全局和局部代理模型得到模型最优解并对其进行真实目标函数评价，实现探索和开采的平衡以减少真实目标函数的计算次数，同时通过针对性地选择个体进行真实目标函数计算，辅助算法快速找到目标函数的较优解。在15个低维测试问题和14个高维测试问题上的实验结果表明，在有限的计算资源情况下，该算法在12个低维测试问题上相较于最优重启策略代理辅助的社会学习粒子群优化算法、基于主动学习的代理模型辅助的粒子群优化算法等表现更好，在7个高维测试问题上相较于高斯过程辅助的进化算法、代理模型辅助的分层粒子群优化算法、求解高维费时问题的代理辅助的多种群优化算法等能找到目标函数的更优解。

Differential Evolution Algorithm Alternately Assisted by Global and Local Models

To solve high-dimensional,computationally expensive optimization problems in practical complex engineering applications,a Differential Evolution(DE) algorithm assisted by the alternate optimization of global proxy and local proxy models is proposed. Therefore,a global or local proxy model is trained based on historical data and optimized alternatingly to assist the algorithm in finding the optimal solution of the original problem. A balance between exploration and exploitation can be achieved by alternatingly searching for the optimal solution of the global and local proxy models.The optimal solutions of global and local proxy models can be evaluated using the real objective function,which helps reduce the number of objective evaluations. Furthermore,it can accelerate the time required to find the optimal solution by selecting target individuals for real objective evaluation.The performance of the algorithm was tested on 15 low-dimensional and 14 high-dimensional test problems. Experimental results show that when computational resources are limited,the proposed algorithm performs better on 12 low-dimensional problems compared to the GORSSSLPSO,CAL-SAPSO and other algorithms,and can find better optimal solutions for 7 high-dimensional problems than the MGP-SLPSO,SHPSO,SAMSO and other algorithms.