Quantifying Energy Overhead in SCADA Cybersecurity Using AI-Based Threat Detection Models

Abstract

The increasing reliance on AI-driven cybersecurity solutions in SCADA environments has introduced significant computational demands, raising concerns about energy consumption in critical industrial systems. This research addresses the gap in existing studies which focus on security effectiveness while overlooking the energy implications of deploying deep learning-based intrusion detection mechanisms. The study presents an empirical analysis quantifying the energy footprint of AI-based cybersecurity approaches including convolutional recurrent networks, generative adversarial networks, and adaptive hybrid models within a realistic SCADA testbed. Energy profiling is conducted using high-resolution hardware instrumentation and software-based power monitoring techniques capturing variations in power usage across different AI models and deployment strategies. The findings demonstrate that while advanced AI models enhance security detection capabilities, they incur substantial energy costs which can be mitigated through model optimization techniques such as pruning quantization and knowledge distillation. Adaptive execution strategies further reduce power consumption by dynamically modulating AI processing complexity based on real-time threat assessments. The study establishes a foundation for developing energy-efficient cybersecurity frameworks that balance security resilience with operational sustainability. These insights inform industry best practices and contribute to future research on low-power AI models, decentralized security architectures, and energy-conscious industrial cybersecurity solutions ensuring effective protection of critical infrastructure without excessive resource burdens.