Recently, nonrevisiting genetic algorithms have demonstrated superior capabilities compared with classic genetic algorithms and other single-objective evolutionary algorithms. However, the search efficiency of nonrevisiting genetic algorithms is currently low for some complex optimisation problems. This study proposes a nonrevisiting genetic algorithm with a multi-region guided search to improve the search efficiency. The search history is stored in a binary space partition (BSP) tree, where each searched solution is assigned to a leaf node and corresponds to a search region in the search space. To fully exploit the search history, several optimal solutions in the BSP tree are archived to represent the most potential search regions and estimate the fitness landscape in the search space. Except for the conventional genetic operations, the offspring can also be generated through multi-region guided search strategy, where the current solution is first navigated to one of the candidate search regions and is further updated towards the direction of the optimal solution in the search history to speedup convergence. Thus, multi-region guided search can reduce the possibility of getting trapped in local optima when solving problems with complex landscapes. The experimental results on different types of test suites reveal the competitiveness of the proposed algorithm in comparison with several state-of-the-art methods.

最近，与经典遗传算法和其他单目标进化算法相比，非重新审视的遗传算法已显示出卓越的能力。然而，对于一些复杂的优化问题，目前非重新访问遗传算法的搜索效率很低。本研究提出了一种具有多区域引导搜索的非重访遗传算法，以提高搜索效率。搜索历史记录存储在二进制空间分区 （BSP） 树中，其中每个搜索的解决方案都分配给一个叶节点，并对应于搜索空间中的搜索区域。为了充分利用搜索历史，BSP 树中的几个最佳解决方案被存档，以表示最有潜力的搜索区域并估计搜索空间中的适应度景观。除了常规的遗传操作外，后代也可以通过多区域引导搜索策略生成，其中当前解决方案首先导航到候选搜索区域之一，然后进一步向搜索历史中最优解决方案的方向更新以加速收敛。因此，在解决复杂景观问题时，多区域引导搜索可以减少陷入局部最优的可能性。在不同类型的测试套件上的实验结果揭示了所提算法与几种最先进方法相比的竞争力。