A Barrier-Certificated Reinforcement Learning Approach for Enhancing Power System Transient Stability

A Survey on Physics Informed Reinforcement Learning: Review and Open Problems

“…A control barrier function is then employed to devise a control law that keeps the states within the safety set. In certain scenarios barrier functions are represented as NNs and learned through data driven approaches [141,140]. In above control theoretic approaches the system dynamics either partial or learnable and safety sets represent the primary physical information.…”

“…Figure 11: Example of action regulation, using physics priors. In [141], a barrier certification system receives RL control policy generated control actions and refines them sequentially using a barrier certificate to satisfy operational constraints.…”

“…For safe power system control [141] proposes a framework for learning a stabilizing controller that satisfies predefined safety regions, see Fig. 11.…”

“…For example, in [68] the designed reward incorporates IMU sensor data, imbibing inertial constraints, while in [75] the physics informed reward is designed to satisfy explicit operational targets. To ensure safe exploration during training and deployment, works such as [133,141] learn a data-driven barrier certificate based on physical property-based losses and a set of unsafe state vectors.…”

“…Some approaches were developed to improve simulators for sample efficiency and better sim to real transfer [1,81]. Carefully selecting task-specific state representations [59,51], reward functions [13,14], and action spaces [124,141] has been shown to improve both the time to convergence and performance. To sum it up, integrating underlying physics about the learning task structure has been found to improve performance and accelerate convergence.…”