ENHANCING THE PHYSICAL PROTECTION OF CRITICAL FACILITIES THROUGH THE INTEGRATION OF PHYSICAL PROCESS MODELS AND MACHINE LEARNING

This paper substantiates a hybrid approach to enhancing the physical protection of critical facilities through the integration of physical process models and machine learning methods. It is shown that conventional deterministic and probabilistic models of physical protection provide high explainability and regulatory compliance, yet exhibit limited adaptability under conditions of a dynamically changing operational environment, sensor data instability, and parameter uncertainty. The proposed methodology combines the preservation of causal logic inherent in physics-based models with the adaptive capabilities of machine learning, primarily for updating the probability of detection using operational and historical data. A structured integration framework is formulated, incorporating a physical process model, a machine learning module, and an alignment block that enforces physical and regulatory constraints. The paper presents a mathematical formulation of protection effectiveness as a joint function of detection, delay, and response components, as well as an adaptive scheme for operational risk updating and the selection of protective measures. The results demonstrate that the hybrid approach improves assessment realism, robustness to uncertainty, and the scientific validity of decision-making in the protection of critical facilities.