Time-Varying Causal Graph Learning for Distribution Network Scheduling
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
The increasing integration of distributed energy resources (DERs) in active distribution networks necessitates efficient multi-timescale scheduling strategies to balance economic operation and system stability. However, existing deep reinforcement learning (DRL) approaches often lack interpretability and struggle to capture dynamic physical dependencies among DERs. This paper proposes a novel Causal Graph-Guided Dual-Timescale DDPG framework for explainable and adaptive scheduling. An online learning algorithm is developed to construct time-varying causal graphs that capture the evolving control-response sensitivities (i.e., physical causality) among DERs. Building upon this structure, a causal graph-guided attention mechanism enables selective feature aggregation, while an adaptive dual-timescale critic network dynamically balances short-term safety constraints and long-term economic objectives. Simulation results on the IEEE 33-bus system demonstrate that the proposed method significantly reduces the voltage violation rate to 35.4 %, outperforming state-of-the-art baselines including Soft Actor-Critic (SAC). Crucially, a physical feasibility analysis reveals that 88.2 % of the residual violations are physically unavoidable due to system capacity limits, indicating that the agent achieves near-optimal control with a gap of only 4.2 percentage points to the theoretical physical limit. Furthermore, the learned causal graphs align with the grid topology, providing intuitive interpretability for practical deployment.
Fields of Study & Subjects
Optimal Power Flow Distribution
Smart Grid Energy Management
Microgrid Control and Optimization
Publication details
- Journal / source
- International Conference on Electrical Automation and Artificial Intelligence (ICEAAI)
- Pages
- 182–188
- Publication date
- 16 January 2026
- Language
- EN