Study On Cooperative Control And Optimization Of Complex Multi-agent Networks

The problems of cooperative control and optimization in complex multi-agent networks have received considerable attention and interest from the control community. The multi-agent network will emerge a kind of collective asymptotic behavior under certain conditions. Many industrial, military and consumer applications call for the cooperation of complex multi-agent networks. The transitions in traffic flow networks depend on the network topol-ogy, transmission medium, routing strategies, the performance of agents and so on. For complex multi-agent networks, how to achieve optimal traffic flow on the networks and how to design meaningful and reasonable distribution optimization algorithms are all challenging problems.In the dissertation, based on previous research works of the others, some essential issues of cooperative control and optimization for complex networks of agents are discussed from the point of view on control theory and optimization by virtue of stochastic analysis techniques, algebraic graph theory, matrix theory and the method of convex analysis. Some new results are given as to these basic problems.The hybrid control problem for global synchronization of uncertain Markovian jump stochastic complex networks with time-varying delay and sector-bounded nonlinearities is studied. The network under investigation is quite general to reflect the reality. Some novel theoretical results for the network synchronization are proposed. By introducing the tuning parameter, the proposed control scheme is less conservative than those existing works in general. An example is given to illustrate the effectiveness of the proposed control design approach.Taking into account both additive and multiplicative noise effects, the issue of almost sure consensus for stochastic multi-agent systems with general directed topology is studied. Some novel general criteria for almost sure global consensus of multi-agent are established and convergence results are provided. These results improve and extend some known rele-vant conclusions from the literature. One example with numerical simulation is worked out for illustrationStochastic consensus seeking problem is considered for networked agents on directed graphs with intrinsic deterministic nonlinear perturbation and stochastic noisy measurement. To attenuate the measurement noises, stochastic approximation type algorithm with decreas-ing step size is employed so that the individual states converge in mean square to the same limit. A sufficient criterion is obtained to guarantee mean square global asymptotical con-sensus of all agents. A simulation example is provided to demonstrate the validity of the proposed scheme.The problem of optimal capacity allocation on heterogeneous complex transport net-works is studied. Network heterogeneity of degree distribution and processing delay of the traffic are considered. An explicit analytical solution is provided, which is based on M/M/1 queueing theory and Karush-Kuhn-Tucker optimization principle. Furthermore, op-timal trade-off between efficiency and capacity for heterogeneous complex transportation networks is considered. An order parameter simulation example by comparing results with those obtained via simple capacity allocation in large Barabd si-Albert (BA) scale-free net-work is provided to illustrate the effectiveness of the theoretical results.Asynchronous algorithm for distributed constraint optimization problem in networks of agents is designed. The iterative algorithm maintains local estimate at each node and de-pends on local projected (sub)gradient updates in combination with a consensus policy. An agent will end up averaging its own value with an outdated value of another processor. The asynchronous scheme does not require that agents exchange state information frequently. Therefore, it is energy-efficient and more realistic than the synchronous one. The con-vergence of algorithms is established under usual conditions. Furthermore, the proposed algorithm is applied to distributed regression with robustness to outliers in sensor networks. Finally, Monte Carlo simulation results are provided to demonstrate the superiority and ef-fectiveness of the proposed scheme.Finally, a summary is done for all discussions in the dissertation. The research works further study are presented.