Design And Implementation Of Multi Agent Communication System Based On Cognitive Consistency

In recent years,multi-agent reinforcement learning has been widely applied in fields such as robot cooperation and resource allocation as an extremely appropriate technology for modeling the real world.This thesis conducts research on multi-agent reinforcement learning communication based on cognitive consistency to improve the robustness and efficiency of communication.The main work is as follows:(1)To address the issue that existing multi-agent communication work requires sufficient bandwidth resources and stable communication channels,a highly efficient and robust communication algorithm is proposed.In this thesis,random noise is added during the training process and combined with a gating mechanism to improve the robustness and efficiency of communication.Based on the classical multi-agent environment,the proposed algorithm can discard nearly 80%of redundant messages while maintaining performance,which is better than other baseline algorithms by 5%to 20%.Additionally,the proposed algorithm exhibits significant advantages in communication robustness.The experimental results show that the proposed model has significant efficiency and robustness in multi-agent communication.(2)To address the issue of introducing too many assumptions in multi-agent communication in previous work,the concept of cognitive consistency is introduced,which allows the model to avoid making too many unreasonable assumptions.Based on the classical multi-agent environment,the proposed algorithm is 5%to 20%better than baseline algorithms in terms of cooperative performance,and the Euclidean distance between the recovered messages and the true messages shows that over 80%of the messages have a distance of less than 0.1.The experimental results show that the proposed algorithm achieves cognitive consistency in multi-agent systems.(3)Based on the task of verifying the efficiency and robustness of multi-agent communication,a front-end and back-end separated multi-agent communication system is designed and implemented in detail.The system has three major modules:the user interaction module,the result display module,and the back-end service module.Firstly,the system requirements and functions are analyzed,and then the system is designed in detail according to the requirements analysis,and the appropriate programming language is selected to implement the system.The user interaction module and the result display module of the system are constructed based on HTML language,and the back-end service module of the system is built based on the Python Django framework.The system supports the selection of various models trained in different scenarios and with different parameters,supports the setting of basic hyperparameters and the selection of corresponding experimental environments,and finally displays the experimental data in a visual form,so that users can easily view the experimental results corresponding to the selected models and parameters,and better verify the performance of the algorithm proposed in this thesis in terms of the efficiency and robustness of multi-agent communication.Considering the communication process between the client and the server,the communication time between the client and the server in this system is no more than 1 second,the client restart time is no more than 10 seconds,and multi-core processors are used to improve processing performance.Moreover,multi-process technology is introduced to ensure that the front-end and back-end services provided by the system are stable and robust.