Federated Detection Systems for Insider Threats in Energy Facilities Using Biomechanical Access Control and Al-Based Cybersecurity

Abstract

Insider threats present major challenges to energy facilities, particularly where traditional centralized detection systems fall short due to privacy concerns and data variability. This research introduces a federated detection system that integrates biomechanical access controls such as pressure-sensitive floors and biometric interfaces with AI-driven behavioral analytics to enhance threat detection while safeguarding data privacy. A hierarchical federated learning architecture enables collaborative model training across multiple sites without sharing raw data, effectively addressing non-IID data issues. Using multi-modal datasets with 148 biomechanical and 83 cyber features, the system achieved 95% precision, 91% recall, and a 0.97 AUC-ROC, surpassing centralized and local models. Biomechanical authentication reached 98.9% accuracy with low false acceptance and rejection rates, offering consistent behavioral signatures. Combining cyber and physical data through attention-based deep learning boosted overall detection accuracy to over 96%, with detection latency under 150 milliseconds. Privacy was preserved using differential privacy (ε ≤ 1.2), meeting regulatory standards. Validated across diverse energy facilities, the system proves scalable, effective, and privacy-compliant. Recommendations include broader deployment, sensor optimization, and enhancing adversarial robustness.