Hsin-Jung Yang, Zhanhong Jiang, Prajwal Koirala et al. · Feb 19, 2026
LexiSafe is an offline safe reinforcement learning framework that preserves safety-aligned behavior through lexicographic prioritization. It provides bounds on sample complexity guarantees and achieves reduced safety violations and improved task performance compared to constrained offline baselines. The framework can be extended to hierarchical safety requirements with multiple costs.
Why This Matters
This paper is relevant to power system engineers as it addresses the importance of safety in cyber-physical systems, particularly in grid operations where the integration of renewable energy sources poses new challenges for ensuring grid stability and reliability. The proposed offline safe reinforcement learning framework, LexiSafe, can help reduce safety violations and improve task performance in grid operation and planning, which are critical applications for power system engineers.