Model-Free Disturbance Observer with Online Modification: Listening to MFDOOM
arXiv:2607.07082
2026
Architecture
1 ideas extracted · analyzed Aug 30, 2026
What the math gives to ML
The paper provides a constructive residual-dynamics observer: prediction errors are stored in a continuously updated Hankel matrix, and a coefficient vector reconstructs future error trajectories from their recent history. The transferable asset is not DeePC itself, but model-free identification of low-dimensional autonomous disturbance dynamics without training a recurrent model. A neural forecaster can therefore be augmented with a cheap online linear residual predictor, yielding causal multi-step corrections for streaming time series under regime shifts or periodic disturbances.
Ideas from this paper
Unverified
2026
Attach a model-free residual-dynamics observer to a neural multi-step forecaster. Instead of asking the network to relearn persistent periodic or autoregressive disturbances, maintain a Hankel dictionary of recent forecast errors and use ridge reconstruction to predict the next residual sequence online. Add the predicted residual to the network forecast with a confidence-dependent correction gain.
Useful6/10
Difficulty4/10
Novelty7/10