An Improved Flocking Algorithm For Multi-agent Systems Via Local Topological Structure Optimization

The ideas of multi-agent systems come from biological group who can emerge complexand orderly crowed behavior only through the communication between individuals. Thephenomenon has attracted continuous attention and research of many scholars in differentdisciplines. For multi-agent systems, how to achieve flocking control of multi-agent systemsonly through the communication between individuals and self-organization behavior in thecase of lack of centralized control and the global model, has become the focus of currentresearch. This dissertation mainly focuses on the flocking control of multi-agent systemsbased on studying and summarizing the related research results and the ideas of communitydetection.The main research contents and contributions are as follows:1) Some rudimentary knowledge and algorithms related research contents areintroduced, such as matrix theory and algebraic graph theory, basic flocking controlalgorithms, flocking control algorithms with virtual leader, pinning control algorithms, PSOalgorithm and Community partition algorithms.2) In the view of most flocking control algorithms applied to large-scale multi-systemsachieve flocking slowly, inspired by the ideas of community structure, the topologicalstructure of multi-agent systems is analyzed by the community partition algorithms. Then themulti-agent systems be divided into a quantity of communities by FN algorithms based onEuclidean distance, these communities have a common nature, namely the communityinternal communication connection between agents are close, agents communicationconnection between different communities are sparse. The ideas change the traditionalflocking control problems into the flocking control problems of these communities. Finally,the simulation results prove the effectively improve the speed of flocking.3) On the basis of the classics flocking control algorithms with virtual leader whichwere presented by Olfati-Saber, this paper does the further study on the relationship betweenthe speed of virtual leader and the connectivity of multi-agent systems. Then, three groups ofsimulation verify that the speed of virtual leader is too big or too small are not conducive tocontrol of multi-agent systems. Next, combined with PSO algorithms to optimize the speed ofthe virtual leader, calculate the speed of virtual leader when the algebraic connectivity ofmulti-agent systems achieves maximum value by means of the fitness function which arecomposed of algebraic connectivity of multi-agent system. Finally, the proposed optimizationstrategy is verified by simulation experiments on the effectiveness of the multi-agent systemscontrol strategy.