Reachability-based Time-domain Distance Protection

What the math gives to ML

The paper replaces pointwise fault detection with reachable-set membership: a measurement is accepted only if it lies in the set of trajectories consistent with an RLC network, source uncertainty, and a candidate fault model. Its transferable mechanism is the constructive reduction of a high-dimensional dynamical system to a low-dimensional apparent-state model, enabling fast set-based state estimation without solving the full differential equations. In neural networks, this can provide a cheap monitor for recurrent or state-space hidden dynamics, detecting hidden-state excursions caused by instability, distribution shift, or corrupted inputs. The central falsifiable prediction is that violations of the propagated reachable tube should occur near the local expansion boundary and should precede loss explosion or long-horizon prediction failure.