Abstract: In order to solve the problem of searching and tracing the Pareto Optimal Front （POF） and Pareto Optimal Set （POS）, two strategies are investigated. The adaptive immigration strategy is designed to improve the diversity of the population by adaptively inserting the immigrations according to the changed environments, thus can improve the adaptability to the environments. The prediction strategy is used to quickly trace POF by the prediction population which is established by the time series and some disturbances. The two strategies are introduced into differential evolution to solve the dynamic multi-objective problems. The experimental results show that the adaptive and prediction strategies based differential evolution shows great ability to adapt to the changed environments and can find POS quickly.

Key words: dynamic multi-objective optimization, adaptive immigration strategy, prediction strategy, differential evolution

摘要： 针对动态多目标优化环境下寻找并跟踪变化的Pareto最优前沿和Pareto最优解集的难题，提出两个策略：自适应迁移策略和预测策略。自适应迁移策略是根据环境的变化自适应地插入迁移个体来提高算法种群的多样性，从而提高算法对动态环境的适应能力。预测策略是通过时间序列并加上一定的扰动来产生预测种群，来预测环境变化之后的Pareto最优解集，以达到对其快速跟踪的目的。通过两个策略在多目标差分演化算法上的应用来解决动态多目标优化问题。实验过程中，通过平均最优解集分布均匀度和平均决策空间世代距离等指标表明，基于自适应迁移策略和预测策略的多目标差分演化算法能够很好适应变化的环境，并能够快速找到Pareto最优解集。

关键词: 动态多目标优化, 自适应迁移策略, 预测策略, 差分演化算法