Interoperability model between heterogeneous MAS platforms based on mobile agent and reinforcement learning
Noureddine El Abid Amrani, Sidi Mohamed Snineh, Mohamed Youssfi, Oum El Kheir Abra, Omar Bouattane · 2021
A model of interoperability between heterogeneous multi-agent systems (MAS) platforms is presented in this article. This model is based on a mobile agent and reinforcement learning (RL) to provide semantic interoperability and message-oriented middleware for technology interoperability. The latter concerns the technical problems of linking systems, the definition of interfaces, the data format, and the communication protocols needed to exchange messages. In this context, we propose an architecture based on the AMQP protocol. On the other hand, semantic interoperability depends on the application domain of each system and refers to the ability of two agents to exchange data while preserving their semantics, and to the ability of the receiving agent to translate or convert the information received to ensure efficient collaborative exchanges. In this context, our technique is as follows: when an agent of MAS1 receives a message from another agent of MAS2 and needs information to interpret this message, it requests the help of the mobile agent. The latter must migrate to the MAS2 equipped with a technique allowing it to extract the information necessary for the interpretation of the message. After this step, the mobile agent offers information to the first agent. If this information helps the agent to interpret the message, the mobile agent receives a positive reward from MAS1, otherwise, it receives a negative reward. The goal of the mobile agent is to maximize their rewards. The experimental results have shown the effectiveness of our approach. Indeed, according to the results, the experience of the mobile agent who has already explored the environment makes it possible to accelerate the learning process of other agents.