Reproduce “Bayesian State Estimation for Unobservable Distribution Systems via Deep Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kursat Rasim Mestav, Jaime Luengo-Rozas, Lang Tong
A deep learning approach to Bayesian state estimation, called BSEdnn, accurately estimates states in unobservable distribution systems by learning the Minimum Mean Squared Error estimator. This method achieves 1-2 orders of magnitude lower error and is five orders of magnitude faster in online computation compared to traditional pseudo-measurement techniques, while also offering robust bad-data detection.
Key insights
- Deep neural networks can efficiently approximate the computationally complex Minimum Mean Squared Error (MMSE) estimator for Bayesian state estimation, overcoming a long-standing barrier.
- Integrating slow-timescale smart meter data for learning the underlying probability distributions of power injections provides a robust way to generate realistic training data, addressing resolution mismatch.
- A 'bulging middle' architecture in the deep neural network, along with optimal depth, was identified to improve estimation accuracy and regularization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.