Machine Learning and Genetic Algorithms for Cyber-Physical Anomaly Detection in Smart Grids
A recent study published on arXiv (2605.22749) introduces a novel approach that integrates machine learning with genetic-algorithm-driven feature selection to identify anomalies in cyber-physical systems within IoT-enabled smart grids. Utilizing the MSU/ORNL Power System Attack Dataset, this method seeks to differentiate between physical events, such as faults, and malicious activities like false data injection. The research assesses various baseline models, including logistic regression, RBF-SVM, XGBoost, Random Forest, and Extra Trees, concluding that tree-based ensemble models yield the highest performance. The study aims to ascertain whether a streamlined set of physically relevant PMU/IED measurements can facilitate dependable detection, addressing the heightened vulnerabilities of contemporary smart grids due to their complex measurement and communication frameworks.
Key facts
- arXiv paper 2605.22749 proposes cyber-physical anomaly detection for smart grids.
- Method combines machine learning with genetic-algorithm-based feature selection.
- Uses the MSU/ORNL Power System Attack Dataset.
- Goal: distinguish physical incidents from malicious actions.
- Evaluates logistic regression, RBF-SVM, XGBoost, Random Forest, and Extra Trees.
- Tree-based ensemble models show best results.
- Aims to use reduced set of PMU/IED measurements for detection.
- Addresses vulnerability from dense measurement infrastructures and communication links.
Entities
Institutions
- arXiv
- MSU/ORNL Power System Attack Dataset