Real-Time Power Dispatch in Grid-Connected Microgrids using Deep Q-Learning

This study presents a methodology to optimize the operation of an Energy Storage System (ESS) to reduce overall costs and maintain network stability through the application of the Deep Q Network (DQN) algorithm. DQN is trained to make energy trading decisions based on real-time data, including energy load, pricing and solar generation. The Prioritized Experience Replay strategy is employed to train the DQN, which in the presence of solar energy must respond to real-time prices and store and release energy at appropriate times to minimize overall costs. In a non-solar environment, the ESS must also manage solar energy within the microgrid while reducing overall costs. The algorithm is evaluated through simulations. Simulation results indicate that the RL-based approach is effective in reducing overall costs and network stability. In the presence of solar energy, the average daily cost with ESS is lower than without ESS, resulting in a net benefit of US$294,886.57. In the absence of solar energy, the algorithm still shows potential to reduce overall costs, with a net benefit of US$615,332.50, highlighting the algorithm's potential to reduce overall costs.