Abstract: To accurately evaluate cooperation agents, it's necessary to use the expericnccbased and reputation-based models existed in MAS. Therefore, a learning mechanism of MAS according to the learning result of active parameter was proposed. MAS using this mechanism gives different trust model the corresponding weight, dynamically calculates the reliability and evaluates the cooperation objects to make the optimal rewards. Simulation shows the efficiency of the learning mechanism.

Key words: MAS,Experience-based model,Reputation-based model,Dynamic learning mechanism