基于多粒子群协同的动态多目标优化算法及应用

Multiple Particle Swarms Coevolutionary Algorithm for Dynamic Multi-Objective Optimization Problems and Its Application

在现实生活中大多数多目标优化问题都随时间变化,这就要求优化算法在时间约束内快速找到动态变化Pareto最优解或Pareto边界.基于此,提出一种基于多种群协同的动态多目标粒子群改进算法,旨在利用多种群竞争和协作两种模式互相配合,从而达到快速高效求解动态多目标优化问题的目的,多种群竞争模式主要任务是对解空间进行"勘探"搜索,当竞争失效后,自适应切换到协作模式对解空间进行"开采"搜索.通过对多种群协同搜索概率分析,证明多种群相比单种群具有更高的搜索效率,通过对3类动态多目标测试函数仿真,验证了改进算法的有效性;最后将该方法应用于动态系统PID控制器的参数整定上,得到了较优的控制参数,取得满意的控制效果.

Most of multi-objective optimization problems in the real-world are dynamic, so optimization algorithms are required to continuously track time-varying Pareto optimal set （POS） or Pareto optimal front （POF） rapidly with high accuracy. To meet this requirement, an improved variant based on particle swarm optimization （PSO） is proposed, in which competitive and cooperative models are combined. The competitive model is used to explore the search space, and when it fails, this model is adaptively switched to the cooperation model to exploit the search space. Co-evolution probability analysis indicates that searching solution using multiple swarms is much more efficient than using a single one. Numerical simulation also shows that the proposed algorithm is an excellent alternative for solving dynamic multi-objective optimization problems. Finally, the proposed algorithm is applied to the PID controller parameter tuning for a dynamic system and gets a satisfactory control.