Multiscale consensus for decentralized estimation and its application to building systems

Multiscale approaches to accelerate the convergence of decentralized consensus problems are introduced. Consecutive consensus iterations are executed on several scales to achieve fast convergence for networks with poor connectivity. As an example the proposed algorithm is applied to the decentralized Kalman filtering problem for estimation of contaminants in building systems. Two conventional observers are designed and convergence is compared with respect to the number of communications necessary, which is an effective measure of system complexity. It is demonstrated that the proposed multiscale scheme substantially accelerates the decentralized consensus. Future extentions and directions are briefly summarized.