Research On Information Evolution Network Towards Distributed Cooperative Decision Making

Decenterlized multi-agent coordination is a classic hot research topic of distributed artificial intelligence.However,as a typical interdiscipline frontier theoretical research aera,there remain many technical issues to be solved.In the current research,the dynamic characters and partial observable restrictions of the agent's nature makes it impossible to observe or predict the global state information in the overall system,and it leads to the incapacitation of the traditional classic offline planning methods.The research in this dissertation narrows the coordination perspective from the large-scale multi-agent system to the micro-level,that is,focuses on the two agents case,and studies the phenomena of association,interaction and coordination among them,and the construction and characteristics analysis under time evolution scale.This dissertation starts with a novel graph-based information evolution network model,and the typical network structure and evolution characteristic of the proposed model is given in the form of derivation,and corresponding numerical experiments.Then,the abstract model is further instantiated into an associated information graph model among agents,and an association analysis between agents is revealed.Based on this,a multi-cadidate model is analyzed.Then,according to the partial observable constraints,a fuzzy state model is bulit,and which evaluation is based on the fuzzy multi-attribute decision.On this basis,a new local interaction coordination model named DLI-MDPs is constructed.By introducing the feature of local interactive behavior between agents,an optimization algorithm HDLI is propsed.In the simulation part,three typical scenarios are choosed to comparing and analyzing the effiency of the HDLI algorithm.The work in this dissertation reveals the essence of multi-agent coordination decision-making,the analyzing of the integrating and evolving process of information between agents.It provides an effective theoretical basis for large-scale multi-agent coordination and complex analysis of higher-level semantic requirements.