Deep-Learning-Based Predictive Control of Battery Management for Frequency Regulation

This paper proposes a deep-learning-based optimal battery management scheme for frequency regulation (FR) by integrating model predictive control (MPC), supervised learning (SL), reinforcement (RL), and high-fidelity models. By taking advantage of deep neural networks (DNNs), the derived DNN-approximated policy is computationally efficient in online implementation. The design procedure proposed consists two sequential processes: (1) SL process, which we first run simulation with an MPC embedding low-fidelity to generate training data set, then, based on generated optimize using algorithms; (2) RL utilize algorithms improve performance balancing short-term economic incentives long-term degradation. process speeds up subsequent providing good initialization. utilizing algorithms, one prominent property that it can learn from simulating FR simulator adjust policy, originally initialized model. A case study real-world signals prices performed. Simulation results show that, compared conventional schemes, effectively achieve higher benefits participation while maintaining lower computational cost.