多Agent自动协商中的改进遗传算法

针对多Agent自动协商中所存在的协商时间长、复杂性高、协商结果无法保证等问题,文中提出了一个基于Pareto最优解的改进遗传算法。该算法在评价函数、种群规模与分布、交叉和变异操作等方面都对传统遗传算法进行了改进,以提高具有高维度多Agent自动协商中算法的性能和效率。通过买卖Agent之间的多对多自动协商典型算例表明,这种算法在保证不同利益群体Agent之间协商取得Pareto最优解的稳定性和使得协商参与者达到协同进化等方面都具有明显的优势。

Improving Genetic Algorithm(GA) Needed for Automated Negotiations among Multi-Agents

To deal with such problems as long negotiation time,high complexity and occasionally deadlocked negotiation in automated negotiations among multi-agents,we put forward an improved GA in the full paper.We explain our improved GA in some detail in the first section of the full paper.In this abstract,we just add some pertinent remarks to naming the three subsections: automated negotiations among multi-agents and the Pareto optimal solution(subsection 1.1),our improved GA(subsection 1.2) and the performance evaluation of our improved GA(subsection 1.3).In subsection 1.2,we do four things:(1) we describe the evaluation function of our improved GA,using the Pareto optimal solutions;(2) to avoid computing complexity and premature convergence,we select that the population scale between 1.5 to 2 times the coding length of chromosomes' strings;(3) to improve search efficiency,we divide the initial scales of population into two groups,one group for the interests of buyers and the other group for those of sellers;(4) to improve the stability and convergence of the GA,we perform the operation of crossover in population with higher utilities and that of mutation in population with lower utilities.Finally,to compare our improved GA with traditional GAs,we did experiments on many-to-many multi-criteria negotiations in electronic markets.The Pareto optimal solutions from the experiments,discussed in subsections 2.2 and 2.3,show preliminarily that:(1) our improved GA is more effective and efficient than the traditional GA in Ref.1(authored by Oliveira et al) in that our GA obtains the Pareto optimal solutions every time;(2) the average nondimensional generation distance is 0.35,much shorter than that of the traditional GA in Ref.2(authored by Niu).