Genetic Algorithms based Predication in an Adaptive Multi-Agent System for Automated Negotiation

A model of cognitive agents, acting in an open environment, which capture the most relevant elements of agents' behavior related to negotiation with other agents, is developed in this paper. Different negotiation strategies used for reaching an agreement between buyers and sellers, modeled as software agents, are described. The article develops the design and the evaluation of different negotiation strategies, where the counter-offers are computed based on genetic algorithms. The agents are capable to make correct decisions, based on their learning capabilities and on the interactions with other agents from the system. The agents' behaviour is mainly motivated by the gain they can obtain, while fulfilling their goals and negotiating, but their behaviour can change during negotiation, according to previous interactions with other agents in the system. The agents are using different strategies to negotiate and several models to adjust their decision during negotiation. They are capable of increasing their performance with the experience, by adapting to the environment conditions. The proposed model is tested in a virtual market environment, where several buyer and seller agents are acting. Using a different set of experiments, the agents show the possibility to improve in time their negotiation strategy, as more negotiations are taking place in the environment.