多Agent主从粒子群分布式计算框架

Multi-Agent Based Distributed Computing Framework for Master-Slave Particle Swarms

面向大规模复杂优化问题,提出了一个基于并行粒子群优化的分布式Agent计算框架.框架中使用一个主群(master swarm)来演化问题的完整解,并使用一组从群(slave swarm)来并行优化一组子问题的解,主群和从群通过交替执行来提高问题的求解效率.采用异步组结构,主群/从群中的各类Agent共享一个解群,并通过相互协作,对解群进行构造、改进、修补、分解和合并等演化操作.该框架可用于求解复杂的约束多目标优化问题.通过一类典型运输问题上的实验,其结果表明,所提出的方法明显优于另外两种先进的演化算法.

To effectively solve large-scale optimization problems, the paper proposes a distributed agent computing framework based on the parallel particle swarm optimization （PSO）. The framework uses a master swarm for evolving complete solutions of the problem, and uses a set of slave swarms for evolving sub-solutions of the subproblems concurrently. The master swarm and slave swarms alternatively implement the PSO procedure to improve the problem-solving efficiency. Using the asynchronous team based agent architecture, a master/slave swarm consists of different kinds of agents, which share a population of solutions and cooperate to evolve the population, such as initializing solutions, moving particles, handling constraints, and decomposing/synthesizing sub-solutions. The framework can be used to solve complicated constained and multiobjective optimization problems efficiently. Experimental results demonstrate that this approach has significant performance advantage over two other state-of-the-art algorithms on a typical transportation problem.