An Efficient Message Passing Algorithm for Decentrally Controlling Complex Systems

This paper proposes a decentralised stochastic control framework for a class of large-scale and complex dynamic networks. The proposed framework describes a decentralised probabilistic control and message passing architecture of mutually interacting quasi-independent subsystems. Within this framework, the outputs of the subsystems are communicated back to topologically-connected neighbours through output probabilistic message passing. This communication approach constitutes the main contribution of the current paper and allow the achievement of the global system goal. The proposed framework reduces the amount of communication required given the low dimensionality of the output space compared to the state space. The updated knowledge through the output message passing is then applied to induce a fully probabilistic local control strategy affecting only the local subsystem.  Finally, a numerical example is presented to illustrate the effectiveness and usefulness of our novel proposed framework.