Privacy Preserving Anomaly Detection System for DER Solar Inverters
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
Reliable operation of solar inverters depends on maintaining a stable frequency. Recent cyber attacks on solar DERs are concerning and increase the likelihood of such stealthy attacks leading to anomalies in DER. Therefore, a robust anomaly detection system (ADS) is crucial for solar inverters used in distributed energy resources (DERs), enabling timely detection and correction of frequency anomalies. Additionally, preserving data privacy is essential for the security and reliability of the power grid. This paper proposes a privacy-preserving anomaly detection system (PP-ADS) based on a multi-stage hybrid machine learning (MSHML) model specifically designed for solar inverter data in DER environments. The MSHML model combines Random Forest and K- Nearest Neighbors regression algorithms. The proposed PP-ADS operates on edge devices installed near the inverter, eliminating the need to transmit raw data to a central server and thereby preserving data privacy. This novel approach employs regression models and a Mean Absolute Error (MAE)-based thresholding mechanism for anomaly detection. One month of real-time operational data from two Fronius Primo 15.0-1 208-240 single-phase solar inverters was used to train the model. The evaluation was performed using datasets with synthetically injected anomalies. The PP-ADS achieved a recall rate of 71.1%, demonstrating its effectiveness in identifying critical faults. The proposed PP-ADS offers a lightweight, adaptive, and edge-compatible solution, enhancing both reliability and cybersecurity in smart grids integrated with DERs.
Fields of Study & Subjects
Smart Grid Security and Resilience
Anomaly Detection Techniques and Applications
Islanding Detection in Power Systems
Publication details
- Journal / source
- UND Scholarly Commons (University of North Dakota)
- Publication date
- 5 March 2026
- Language
- EN