Answers are grounded in this paper and cite their sources.
Smart meters (SMs) in low-voltage distribution grid are essential for modern distribution systems, providing high-resolution data for applications like demand response, anomaly detection, and topology identification. However, SMs voltage data often suffer from missing values and outliers due to malfunctions or communication failures, particularly under nonlinear system dynamics, which compromise downstream tasks. This paper proposes a deep learning enhanced Hankel matrix data recovery method (DL-HMDR) to robustly impute missing data and correct bad data in nonlinear SMs scenarios. This method constructs a 3D short term Hankel structure from damaged multi-user time series and applies deep nonlinear projection to capture complex dynamics while avoiding fixed kernels. Subsequently, low rank constraints are enforced via learnable factor matrices with a conservative upper bound, and the network is trained with both reconstruction Mean Squared Error loss and constraint regularization to ensure accurate imputation. Furthermore, theoretical analysis is provided to guarantee Lipschitz stability and a bounded reconstruction error under noise and missing data. Experiments on synthetic and real datasets demonstrate that DL-HMDR consistently outperforms state of the art methods across varying missing rates, modes, and noise levels; the recovered data also achieves the highest F1 score in downstream topology identification, confirming its practical utility.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: