Abstract: Robust optimization is a new method for solving dynamic optimization problems with the goal of finding solutions that are still acceptable for a long time. Most of the researchers in this field try to find new robust solutions based on their future predicted fitness values. However, the error of predicting the future fitness value is often too large, which makes it difficult to find a better robust solution. To solve this problem, an algorithm framework based on problem feature change guidance (ROOT-PFCG) is proposed for dynamic robust optimization. The change of problem characteristics mainly refers to the objective function value of the solution in the current environment and the floating value of the objective function in the corresponding adjacent environment, and three important indexes are proposed. In the case of prediction and non-prediction, three different fitness decision rules are proposed to select the solution based on the index to ensure that the selected solution is less or not affected by the prediction error, so as to find a better robust solution. On this basis, a new performance evaluation index is proposed. The experimental results on benchmark problems show that the proposed algorithm can better improve the performance of robust solutions, and further analyze the impact of indicators on performance in different situations. On this basis, a better indicator combination method is analyzed.

Key words: dynamic robust optimization, particle swarm, feature change, prediction error, guide-individual

摘要： 随着时间的推移，鲁棒优化是解决动态优化问题的一种新方法，其目标是找到在很长一段时间内仍然可以接受的解决方案。该领域中大多试图根据其未来预测适应度值来寻找新的鲁棒解决方案，然而，预测未来的适应度值的误差往往偏大，对其寻求较好的鲁棒解造成较大的困难。针对这一问题，提出了一个基于问题特征变化引导的算法框架（ROOT-PFCG）来进行动态鲁棒优化。其问题特征变化情况主要参考解在当前环境下的目标函数值和相应相邻环境下的目标函数浮动值，由此提出三个重要指标。在预测和非预测的情况下，基于指标分别提出了三种不同的适应度决策规则来选解，保证其所选解受预测误差影响较小或不受影响，以此寻找更优的鲁棒解，并在此基础上提出了新的性能评价指标。在基准问题上的实验结果表明，所提出的算法能更好地提升鲁棒解的性能，并对不同情况下的指标进一步分析了其对性能的影响，在此基础上分析了更好的指标结合方法。

关键词: 动态鲁棒优化, 粒子群, 特征变化, 预测误差, 引导个体