Based on Reinforcement Learning for Power Command Allocation of Battery Energy Storage Stations

At present, battery energy storage is gradually becoming an important part of secondary frequency regulation in power grids. To improve the operational efficiency of battery energy storage stations in participating in secondary frequency regulation, it is necessary to reasonably allocate frequency regulation power commands to each energy storage unit. A two-level power decomposition algorithm based on reinforcement learning is proposed, which can quickly process nonlinear solution problems through reinforcement learning. Firstly, an upper-level combination model based on reinforcement learning is constructed based on the fitting curve of energy storage unit power loss, with minimum power loss as the goal to calculate the power command allocation combination. Secondly, a lower-level arrangement model based on sorting algorithms is constructed, with the goal of balancing the state of charge (SoC) to adjust the order of power command allocation and issue corresponding commands to the energy storage units.