Research On Optimization Of Network Cognition Based Cooperative Interaction In Decentralized System

Decentralized coordination systems have been widely used in modern information systems, in which the individuals are typically deployed in complex environments. These individuals coordinate with each other to achieve a common goal. While when the team scales up, how to design the interaction policy among individuals to improve the efficiency of the system becomes a bottleneck due to the partial observability of the team state. This problem has been a NEXP-COMPELTE hard problem, it is barely impossible to optimize the behavior of all team members. The current researches usually focused on the communication strategies, and ignored the impact of network structure or complex network effects on the team performance. There have been lots of researches on complex network, however, there is very little research using or changing the network structure to improve the team performance.The interactive ways of nodes in decentralized system are similar to human beings, nodes make their own decisions based on their location as well as the current state of the team. Nodes can make rational decisions based on their recognition of the network structure, or by changing the structure to make it more profitable to interact with others.In this paper, we focused on the improvement of performance of two typical types of decentralized systems: sensor network or multi-robot system in which nodes communicate with each other via broadcast, and multi-agent based coordination in which nodes communicate by P2 P. As in broadcasting network, the connection among nodes relies on the physical deployment, the change of network structure will consume a lot of energy of the nodes, which will shorten the life of them, as well as the network, so we do not changing the network structure, and take advantage of the complex network effects. While in the P2P-based communication network, connections among nodes are logical, we can changing them easily. Therefore, by alter the connections among nodes, we can adjust the network structure and make the network more conducive to collaborative interactions.We explore how the complex network effects can be used in the team coordination, or how the network structure affect the team performance by simulation, including average distance, node degree distribution and clustering coefficient, betweenness centrality. Then we explore the local observation model to complex network and the algorithm of complex network attributes maintenance as well as the updating function. Based on the complex network updating algorithm, we build an integrated algorithm for information rebroadcast decision making from the estimated complex attributes value in broadcasting network, and the heuristic online network structure adjusting algorithms corresponding to the beneficial social effects in P2 P network 。 The simulations demonstrate the effectiveness of the two heuristic approaches.