NovASGrid: Novelty-Aware Smart-Grid Resilience

Abstract

Cyber–physical systems (CPS) such as smart grids increasingly encounter novel behaviors that differ from labeled anomalies and may disrupt operations if left unmanaged. We present a feasibility pipeline for handling such novelties in timeseries consumption data, combining anomaly detection, predictive residual modeling, and a simple triage illustration. Our goal is to demonstrate how existing tools can be integrated end-to-end in a pipeline that highlights challenges and opportunities for noveltyaware CPS resilience. We demonstrate the pipeline on residential consumption data from the State Grid Corporation of China (SGCC), where we inject CPS-inspired novelty patterns (e.g., drift, spike, surge, flip, outage) to stress-test detectors. We adapt off-the-shelf baselines, including density-based methods (e.g., LOF) and a predictive baseline (ELSTM), showing how they capture different novelty regimes. Finally, we illustrate how outputs may be aggregated into a novelty characterization step to highlight potential operator-facing triage. Our results suggest that this integration of passive detection, predictive residuals, and simple scoring can highlight tradeoffs between novelty regimes and motivate future, deeper studies of novelty-aware CPS resilience.