Modeling the neutral state of the “Human-operator” system

Abstract

Artificial intelligence is increasingly being applied in railway control to support real time dispatching, train trajectory optimization, automatic train operation, and safety supervision. This review examines how AI is used across the main control layers of railway systems, including network traffic management, train level control, safety assurance, predictive support, and virtual coupling. The literature shows that reinforcement learning and other data driven methods can improve adaptability and execution speed, especially under disturbance and uncertainty. However, the review also shows that AI does not replace the classical foundations of railway control. Feasibility still depends on infrastructure constraints, signalling logic, braking laws, operational rules, and explicit safety supervision. The strongest recent approaches are therefore hybrid, combining learning based decision making with optimization, expert knowledge, shielding, formal safety methods, and supervisory control. Virtual coupling further highlights this trend by linking traffic management, train following, and safety into one tightly constrained control problem. Overall, the field is moving toward layered intelligent railway architectures in which prediction, optimization, learning, and safety assurance operate together rather than separately.