Reinforcement Learning-Based Optimal Power Flow Control in Islanded Microgrids

One of the popular paradigms for control problems in different applications is Reinforcement Learning. In this paper, we develop an RL-based technique for OPF control in islanded micro grids. Micro grids are small power systems for remote or rural regions that can function independently of the national grid and serve as a dependable alternative. The dependence on their decentralized power generation over DERs is a major complication for micro grid system stability and noteworthy optimization of reactive power flow. The operation of micro grids is both dynamic and stochastic; as a result, operational models based on traditional control methods at the model level or rule-based conflicts with optimized controllers that need to work in an environment where the future long-term forecasts remain uncertain. RL is suggested as a data-driven and adaptive control method to address these issues and optimize power flow in islanded micro grids. The RL-based control proposed in this paper uses realtime data from the micro grid to learn an optimal policy. The presented micro grid environment is for the reinforcement learning agent with which it interacts, and on success or failure of its actions, it will be given a reward signal to learn. The RL agent, through its interactions and learning process by taking actions during the operation of islanded micro grids, can optimize power flow in a distributed manner given DERs availability, load demand dynamics, and network constraints.