Anomaly Detection in Smart Power Grids with Graph-Regularized MS-SVDD: a Multimodal Subspace Learning Approach

TL;DR

This work extends MS-SVDD to multimodal anomaly detection in smart power grids by introducing graph-embedded regularizers that utilize modality-specific distance information to shape a shared low-dimensional subspace for one-class classification. The framework supports multiple modalities (voltage, active power, reactive power) and allows mixed-sign gradient updates and various decision fusion strategies to study modality importance. On a synthetically generated PSML-based ED dataset, graph-regularized MS-SVDD with non-linear projections achieves high reliability and fast earliness, with GM up to about 0.79 at 0% noise and 0.78 at 10% noise, and earliness near 0.83–0.92 relative to a 100 ms CCT. The results indicate active and reactive power are more informative than voltage for multimodal anomaly detection in this setting, and the proposed approach provides a scalable, general framework for graph-regularized multimodal OCC in smart grids.

Abstract

In this paper, we address an anomaly detection problem in smart power grids using Multimodal Subspace Support Vector Data Description (MS-SVDD). This approach aims to leverage better feature relations by considering the data as coming from different modalities. These data are projected into a shared lower-dimensionality subspace which aims to preserve their inner characteristics. To supplement the previous work on this subject, we introduce novel multimodal graph-embedded regularizers that leverage graph information for every modality to enhance the training process, and we consider an improved training equation that allows us to maximize or minimize each modality according to the specified criteria. We apply this regularized graph-embedded model on a 3-modalities dataset after having generalized MS-SVDD algorithms to any number of modalities. To set up our application, we propose a whole preprocessing procedure to extract One-Class Classification training instances from time-bounded event time series that are used to evaluate both the reliability and earliness of our model for Event Detection.

Figures (6)

• Figure 1: Anomaly detection over an incomplete time series, where $\tau_1$ is the beginning of the anomaly, $\tau_2$ is the end of the anomaly, and $T$ is the current time.
• Figure 1: Preprocessing of PSML dataset for reliability evaluation.
• Figure 2: Illustration of MS-SVDD with two modalities. The shared subspace is constructed using the positive instances information.
• Figure 3: High-level illustration of graph-embedded regularizers for a modality $m$ with $D_m=2$ and $k=3$. Each plot corresponds to one type of Laplacian matrix.
• Figure 4: Examples of several time-series measured at the same bus while an event occurs somewhere in the grid. Each row corresponds to an electrical quantity (voltage, reactive power, active power), and the dotted lines are the selected event's beginnings and ends. Note that the signal still oscillates after the end of an event.