Abstract:
When Multi-Agent System(MAS) is with limited resource, environmental information is unknown, and task is randomly generated in turn, considering the case of sequential task generated randomly, a new fitness function based on the balance of residual resource is designed, and an improved Binary Particle Swarm Optimization(BPSO) algorithm is proposed. The unbalance of system residual resource is defined by introducing penalty coefficient. The new fitness function considers not only the system profit, but also the balance of system residual resource, and makes a compromise between them by adjusting the penalty coefficient. The improved BPSO algorithm is used to optimize the coalition, redefine the particle velocity and position update formula. The particle divergent is effectively controlled, and the local search ability of the algorithm is improved. Simulation results show that, MAS using proposed fitness function can execute more tasks than using traditional fitness functions. The proposed algorithm has better performance compared with BPSO and Genetic Algorithm(GA) in terms of quality, convergence speed and stability of solutions.
Key words:
Multi-Agent System(MAS),
unknown environment,
multi-task,
discrete particle swarm,
coalition generation
摘要： 针对多Agent系统(MAS)资源有限、环境信息未知、任务依次随机产生的情况，通过引入惩罚系数，基于剩余资源平衡定义一种新的适应度函数，并提出改进的二进制离散粒子群优化(BPSO)算法。新的适应度函数不仅考虑系统收益，同时还考虑系统剩余资源的平衡性，并通过调整惩罚系数在两者之间做出折衷。利用改进的BPSO算法对联盟进行优化，给出粒子速度和位置的更新公式，从而控制粒子的发散性，提高算法的局部搜索能力。仿真结果表明，新的适应度函数可使MAS执行更多的任务。与基本BPSO和遗传算法相比，改进算法在解的质量、收敛速度和稳定性方面具有更好的性能。
关键词:
多Agent系统,
未知环境,
多任务,
离散粒子群,
联盟生成
CLC Number: