Predictive control of a Smart Grid: A distributed optimization algorithm with centralized performance properties

The authors recently proposed several model predictive control (MPC) approaches to manage residential level energy generation and storage, including centralized, distributed, and decentralized schemes. As expected, the distributed and decentralized schemes result in a loss of performance but are scalable and more flexible with regards to network topology. In this paper we present a distributed optimization approach which asymptotically recovers the performance of the centralized optimization problem performed in MPC at each time step. Simulations using data from an Australian electricity distribution company, Ausgrid, are provided showing the benefit of a variable step size in the algorithm and the impact of an increasing number of participating residential energy systems. Furthermore, when used in a receding horizon scheme, simulations indicate that terminating the iterative distributed optimization algorithm before convergence does not result in a significant loss of performance.