Optimal Scheduling of Virtual Power Plant based on Twin Delayed Deep Deterministic Policy Gradient Algorithm

초록

The virtual power plant (VPP) serves as an important market entity that aggregates demand-side distributed resources to participate in electricity market transactions. However, the coordination of heterogeneous distributed resources within a VPP often suffers from low efficiency of numerical optimization methods, while heuristic optimization methods typically incur excessive computational time. To address these challenges, this paper proposes a VPP energy optimal scheduling strategy based on the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm. First, a market framework for VPP scheduling is established. Then, a TD3-based optimization model is developed, which avoids the reliance of conventional optimization methods on fine-grained resource modeling and commercial solvers. Case studies and simulation analyses demonstrate that the proposed TD3-based strategy achieves faster response in the decision-making phase compared with numerical and heuristic optimization methods, even without detailed modeling of diverse distributed resources. These results highlight the potential of the proposed TD3-based deep reinforcement learning approach for scalable and efficient energy management in large-scale practical VPP applications. © 2025 IEEE.

키워드