How To Orchestrate RL Agents for Topological Remedial Actions Using PowSyBl’s Ope...
summary
Combines PowSyBl security analysis with reinforcement learning to produce curative topology reconfigurations for a Simbench EHV network in node‑breaker form, training one PPO policy per contingency ID and running about 81 lines worth of agents. Agents observe voltages, line loadings and switch states, output grouped breaker/disconnector actions enforced by invalid action masks, export trained policies via an ONNX runtime, and the demo shows consistent reductions in maximum line loading during live Security Analysis runs.
AI-written summary · about
chapters (14)
- 0:00Introduction and agenda
- 1:04Need for topological measures
- 2:07Reinforcement learning approach
- 3:34Security analysis context
- 4:18Training environment details
- 5:46Action space design
- 7:56Reward function structure
- 8:41Real-time security analysis
- 9:45Training pipeline
- 10:47Implementation workflow
- 12:13Deployment with Onyx runtime
- 12:57Operational demonstration
- 15:06Software design principles
- 16:59Q&A session
related talks
14mConnecting Reinforcement Learning to Simulation Software
29mLFE Project PowSyBl: Grid Modeling at the Core of Europe’s Power Grid Planification...
21mScalable agent-based models for optimized policy design: applications to the ec...
18mKeynote: AI for Financial Services
26mTowards an AI Assistant for Tertiary Voltage Control
23mOpenROAD and the Era of Agents
25mIntelligent Kubernetes Workload Optimization
21mFrom Reactive to Proactive: How AI is Redefining Resilience in Modern Networks