A Gpu-native Approach on Tackling Grid Topology Optimization...
summary
A GPU-native DC solver runs on Jax and interfaces with PowSyBl and Pandapower to evaluate topology changes, achieving a billion load flows per second and delivering optimization results under 10 minutes on grids of ~3,000 buses. The pipeline is an end-to-end MVP with AC re‑validation, uses a quality‑diversity/random-search optimizer (map‑elites style), reports ~5 orders of magnitude speedup over commodity solvers, and already surfaces viable topologies that operators had not found.
AI-written summary · about
chapters (8)
- 0:00Introduction and project overview
- 1:47What is topology optimization
- 4:58GPU-native approach and performance
- 9:57Full pipeline and validation
- 14:30Live demo in control center
- 18:30Road ahead and future features
- 20:57Open sourcing journey
- 24:07Q&A session
related talks
12mGeneration and Load Interconnection Studies in the US Using PowSyBl
29mTowards GridFM: GENCO – a Unified Neural Solver for Steady-stat...
29mPowSyBl: TSO Cooperation Through Building Blocks
08mTowards Industrial Grid Foundation Models With PowSyBl and GridFM
34mGrid Enhancement Platform: Operationalizing Open Source for Real-Time, Risk-Aware Gri...
19mBringing the Industry Together Through LF Energy - A Power Grid Mode...
43mMore Renewable Energy Into the Power Grid with Open Source
30mBeyond the DC Walls: Building a Nationwide GPU Fabric With KubeVirt and Wide-Area…