DAE-Embedded Verification Ensures Safe Neural Control for Ship Microgrids
A formal method to prevent control spikes in shipboard microgrids during transient shocks
Neural controllers offer strong potential for managing highly nonlinear dynamics in shipboard microgrids (SMGs), but their black-box nature can cause abrupt control spikes and actuator saturation during initial transient shocks. To address this, researchers Fei Feng, Lizhi Wang, and Ziqian Liu present a formal verification method tailored for SMG neural controllers. Their approach includes two key innovations: a set-based differential-algebraic equation (DAE) model compatible with set propagation, and a DAE-embedded bound propagation technique that computes tight envelopes of all possible neural control outputs. This enables engineers to formally certify that the controller will stay within safe bounds even under uncertain disturbances.
Extensive case studies validate the method's effectiveness in certifying SMG control performance across a range of transient shock scenarios. The work bridges the gap between black-box neural network control and the rigorous safety requirements of maritime power systems. By providing provable guarantees against dangerous control spikes, this method could accelerate the adoption of AI-driven control in critical infrastructure like shipboard power grids, where reliability and safety are paramount.
- Introduces a set-based DAE model for shipboard microgrids that enables formal set propagation for neural control verification.
- DAE-embedded bound propagation computes tight envelopes of all possible neural control outputs under transient shocks.
- Case studies demonstrate formal certification of control performance, preventing actuator saturation and spikes.
Why It Matters
Enables safe deployment of neural controllers in safety-critical maritime power systems under unpredictable disturbances.