Distributed Accurate Average Consensus Algorithm and Its Application to State-of-Charge Balancing of Networked Battery Systems

The dynamic average consensus problem, a group of agents, each associated with a time-varying signal, reaching consensus at the average of these signals by their own distributed estimators that interact with each other through the communication network among the agents, finds many applications such as distributed estimation, formation control and sensor fusion. Many distributed estimators have been constructed that achieve either consensus precisely at the average of the signals or around it depending on the properties of the signals. In this thesis, we revisit the dynamic average consensus problem in both the continuous-time and discrete-time settings. By utilizing the information on the frequency components of the signals, we construct distributed estimators that achieve accurate consensus at the average of the signals. We further establish that our distributed estimators are robust to the interruption of the network connectivity in the sense that connected subgroups of agents will continue to reach consensus around the average of all signals after an interruption of the network as along as the signals are bounded and the later the interruption occurs the more accurate the consensus will be. Numerical simulation is carried out to illustrate these theoretical conclusions. We apply our proposed distributed estimators to a networked battery system to achieve accurate state-of-charge balancing while delivering the desirable total power accurately. Simulation results also verify the robustness of the battery system when the communication is interrupted.