多Agent系统中信任预测的SRL模型.pdfVIP

2010 12 Dec. 2010 33 6 Journal of Beijing University of Posts andTelecommunications Vol. 33 o. 6 : 2010) Agent SRL 李小勇, 周 锋, 杨旭东, 倪 晖 ( , 100876) : Agent(MAS) , Sarsa( SRL) MAS Agent. Agent, , SRL, . SRL , . , ,. : Agent; ; Sarsa : TP393 : A SRLBased TrustPredictingModelUsed inMultiAgent Systems LIX iaoyong, ZHOU Feng, YA G Xudong, IHui ( BeijingKey Laboratory of IntelligentT elecommunications Software andMultmi edia, BeijingU niversity ofPosts and Telecommunications, Beijing 100876, China) Abstract: Focusing on the requirement of trustmanagement inmultiAgent systems, the sarsa reinforce ment learning ( SRL) theory is applied to construct trust prediction model formultiAgent system s based on Agent s behavior. First, basic form aldescription is conducted for trustdecision, and behavior state space structure is constructed based on tmi estamp according the interaction tmi e sequence bewt een net work Agents. W ith SRL algorithm, overall trust relationship predictingmodelbased on direct trust degree and feedback trustdegree is proposed. Themodelmakes fulluse of the advantages of the strong dynam ic adaptive capacity of the SRL algorithm, brakes away from the inadequate dynam ic adaptive capacity in the traditional software trustmodeling process. Smi ulation in cumulative errors shows that, compared to the existing models the new modelhas remarkable enhancements in the trust decision accuracy. Key words: multiAgent system s; trustmode;l Sarsa reinforcement learning [ 45] MAS Agent, , SRL , . Agent