HYBRID MODEL FOR INTEGRAL PERFORMANCE ASSESSMENT OF AIR TRAFFIC CONTROLLERS IN SIMULATOR TRAINING USING ARTIFICIAL INTELLIGENCE TECHNOLOGIES

Abstract

This research is aimed at developing a hybrid performance assessment model for air traffic controllers (ATCs) during the execution of emergency scenarios, integrating the ICAO Competency-Based Training and Assessment (CBTA) framework with Artificial Intelligence (AI) technologies. A logical-algorithmic architecture is employed to analyze ATC simulator telemetry alongside Natural Language Processing (NLP) technologies. To aggregate objective machinegenerated data and subjective instructor evaluations, a Sugeno-type Adaptive Neuro-Fuzzy Inference System (ANFIS) is utilized. Machine Learning (ML) algorithms are applied to dynamically adjust the penalty indices within the defined error taxonomy. The application of ANFIS enabled the mathematical mitigation of the human factor's epistemic uncertainty during the scoring process. The implementation of explainable AI methodology based on SHAP analysis ensures absolute transparency of the automatically generated scores, providing visually structured reports for post-simulation debriefings. The proposed hybrid assessment model algorithmizes a specialized ATC error taxonomy through dynamic ML-driven weighting. Unlike static models, the applied approach allows the assessment system to continuously adapt to the spatiotemporal complexity of the emergency situation itself, the capabilities of the controller working position equipment, active airspace restrictions and limitations, and the evolution of general air traffic and meteorological conditions within the ATC's sector of responsibility and adjacent sectors. The proposed integral performance assessment model for ATC simulator training minimizes the subjectivity of instructor oversight regarding the professional proficiency level of ATCs and facilitates the generation of individualized training trajectories for the targeted remediation of skill gaps. The synergy of the instructor's assessment component and the simulator's digital data through neuro-fuzzy modeling establishes a robust mechanism that modernizes simulator evaluation processes in compliance with current ICAO and EASA standards to enhance flight safety.