Distributed online learning and dynamic robust standby dispatch for networked microgrids

Appropriate distributed dynamic standby dispatch schemes are of great importance to manage distributed energy systems and integrate large-scale renewables without the guidance of central intelligence or day-ahead forecasting. This study focuses on systems based on networked microgrids (MGs), which include self-contained medium-voltage MGs—consisting of renewable or dispatchable generators, energy storage systems (ESS), and flexible loads. An online dynamic dispatch scheme is proposed to support autonomous operation or to serve as a standby dispatch scheme in an emergency when a dispatch center is unavailable or day-ahead planning is infeasible. First, the multi-MG energy management problem is modeled into several distributed coupled subproblems. Afterwards, taking Lyapunov drift and online learning of future costs into consideration, the distributed problem is transformed into a robust long-term optimization. Moreover, a Lyapunov-based dynamic algorithm is employed to solve the problem in a fully distributed fashion. Finally, taking an actual system as an example, the superiority and feasibility of the proposed strategy are verified by simulation. The proposed distributed online standby dispatch has good performance in the long-term economy optimization and queue stability without any day-ahead forecasting and central decision-making compared with existing frameworks.