Anomaly Detection for Sensor Data in Cyber-Physical Energy Systems Based on Time-Series Prediction and Heuristic Segmentation

The increasing complexity and interconnectivity of cyber-physical energy systems highlight the critical need for effective anomaly detection methods, especially given the vast volumes of sensor data generated in these environments. This paper proposes a novel anomaly detection method for sensor data in cyber-physical energy systems, termed HS-TSP (Heuristic Segmentation and Time-Series Prediction). HS-TSP integrates heuristic segmentation to divide sensor datasets into statistically homogeneous segments, combined with advanced time-series prediction models to enable accurate forecasting within each segment. Anomalies are detected by analyzing the residuals between predicted and observed values, assessed against a dynamically adaptive threshold that reflects the unique operational characteristics of energy systems. Extensive evaluations on real-world datasets from energy infrastructures demonstrate that HS-TSP effectively identifies diverse anomaly types, achieving high accuracy and low latency. This method offers a robust and computationally efficient solution for anomaly detection in cyber- physical energy systems, contributing to enhanced operational reliability and performance.