Sliced-Wasserstein-based Anomaly Detection and Open Dataset for Localized Critical Peak Rebates

TL;DR

This work tackles trustworthy ML in energy demand-response by introducing a sliced-Wasserstein distance–based unsupervised outlier filter and releasing the first open dataset for localized critical peak rebates (LCPR) in Québec. The main approach combines $SW_{\|\cdot\|,t}$-driven anomaly filtering with a practical Gaussian-process benchmark for winter substation consumption, providing a simple yet robust MLOps-friendly data curation method. The contributions include the SW-based AD method, the open DR-focused LCPR dataset, and baseline benchmarking to spur robust ML research for critical-energy infrastructure. Overall, the paper advances safe, reproducible ML evaluation in high-stakes grid applications and offers a valuable resource for researchers studying localized demand-response scenarios in northern climates.

Abstract

In this work, we present a new unsupervised anomaly (outlier) detection (AD) method using the sliced-Wasserstein metric. This filtering technique is conceptually interesting for MLOps pipelines deploying machine learning models in critical sectors, e.g., energy, as it offers a conservative data selection. Additionally, we open the first dataset showcasing localized critical peak rebate demand response in a northern climate. We demonstrate the capabilities of our method on synthetic datasets as well as standard AD datasets and use it in the making of a first benchmark for our open-source localized critical peak rebate dataset.

Figures (9)

• Figure 1: Distribution of key features for each substation
• Figure 2: Correlation heatmap of key features and label for each substation
• Figure 3: Spearman coefficients of key features for each substation
• Figure 4: Shapley analysis of key features for each substation on trained XGBoost
• Figure 5: Illustration of different groups