Abstract:

With the large-scale integration of renewable energy into power grids, system frequency regulation faces unprecedented challenges. This study proposes an adaptive frequency regulation method for hybrid energy storage systems based on quantum-enhanced deep reinforcement learning and spatiotemporal graph neural networks (QE-DRL-ST-GNN), aiming to improve grid frequency regulation performance across multiple timescales. By integrating quantum computing with deep reinforcement learning and graph neural networks, the method overcomes limitations of traditional approaches in handling high-dimensional state spaces and complex spatiotemporal dependencies. QE-DRL-ST-GNN utilizes quantum state encoding to represent system states, extracts spatiotemporal features via convolutional operations on quantum graphs, and optimizes the reinforcement learning strategy through quantum variational algorithms. Furthermore, an adaptive quantum circuit generation mechanism is designed to automatically adjust the quantum circuit structure in response to dynamic system characteristics. Case analyses show that, compared with the traditional quantum-enhanced deep reinforcement learning (QE-DRL) method, QE-DRL-ST-GNN maintains frequency deviation within 0.05 Hz under extreme conditions, whereas the traditional deep reinforcement learning (DRL) method allows a deviation of 0.15 Hz, indicating an improvement of 66.67%. In terms of regulation time, QE-DRL-ST-GNN requires only 1.67 s in complex scenarios, representing a 47% reduction compared with the traditional DRL method. Additionally, in extreme conditions, the performance improves by 13 percentage points over the 83% achieved by the traditional DRL method.

Key words: hybrid energy storage frequency modulation, quantum enhanced learning, adaptive control, multi-timescale coordination, graph neural networks, hybrid quantum-classical control