Adaptive Control of a Microgrid - Application to the Lebanese Case

The main objective of our work is to find an adaptive master controller responsible for controlling the power flow in a microgrid. Due to the different types of microgrids, and the variables that changes on a daily basis throughout the year, we chose Deep Reinforcement Learning to be the control strategy behind our master controller. This type of algorithm will be able to adapt to any given situation and could be used in any microgrid when it reaches a sufficient level of training. We only need to accurately model the microgrid which will be the environment where the Reinforcement Learning (RL) agent can perform the training before it could be applied to a real-world microgrid. The results show that the RL controller gives better performance than a simple human-designed algorithm, even if using limited computing power.