Distributed optimization of energy flows in microgrids based on dual decomposition

In this paper, a transactive control technique, dual decomposition (DD), combined with a reinforcement learning algorithm, fitted Q-iteration (FQI), is proposed to control the power injected into the main utility grid by a microgrid. To maintain a stable grid connection and provide flexibility services to the distribution system operator, the microgrid manager can control the power injected to the grid by microgrid users with a price signal computed using DD. This work focuses on how batteries located at the homes of microgrid users can be used to minimize the amount of power injected to the grid. The operation of the battery is controlled using FQI. The effect of battery sizes and charging rates on the amount of flexibility available is also analysed. Our approach is evaluated by considering a microgrid with 7 residential users each containing a battery, a photovoltaic installation and a non-controllable base load. Data used for the simulations is obtained from Belgian residential users. Simulation results show that with DD, a 10% increase in power injected to the grid is obtained compared to a theoretical optimal benchmark. These results show the performance of the proposed method while preserving user privacy: inaccessible user preferences/devices. Also, the simulation results show that batteries can be used as flexibility providers to control the amount of power injected to the grid despite their operational constraints. Moreover, these results provide an insight for considering DD combined with FQI in scenarios where several community microgrids interact with each other and the main utility grid.