Cognitive load assessment method for multifactorial flight conflict detection

Real-time monitoring and evaluation of air traffic controllers’ cognitive load is of great significance to the operational safety of the air traffic control system. Cognitive load variation in control command, which is reflected by the physiological parameters, facilitates timely discovery of the undesirable working condition that affecting control effectiveness, in order to realize forward movement of the risk control gate. The radar control simulation experimental platform is used to create conflict detection scenarios under various airspace complexity conditions. A repeated within-subjects measurement experimental scheme is designed with three factors: minimum distance (4 km, 12 km, 16 km), convergence angle (45°, 90°, 135°), and speed characteristics (fast speed priority, same speed, slow speed priority). The effect of different complexity factors on the cognitive load and the changing laws of physiological indexes are studied based on multifactorial analysis of variance through collecting the subjects' eye movement and electrocardiogram data, so as to select feature physiological indexes that can effectively express airspace complexity factors. Based on this, the individuals' cognitive load is assessed using three machine learning algorithms: random forest (RF), support vector machine (SVM), and long short-term memory network (LSTM). The results show that among different levels of the same type factor, the cognitive load is highest when the minimum distance (MD) is closed to the warning interval and the convergence angle (CA) is an acute angle, as well as lowest when the speed characteristics (SC) is the same speed respectively. Nine physiological indexes can be used as feature indexes to effectively express cognitive load at different levels of MD, including number of fixations (NrF), number of saccades (NrS), mean saccade duration (SD), average saccade amplitude (SA), average saccade peak velocity (SPV), blinking frequency (BF), pupil diameter (PD), mean respiratory rate (RR) and power ratio of low-frequency to high-frequency in heart rate variability (LF/HF). The assessment accuracy of the SVM model based on a single-modal eye movement signal is 94.69%, which is higher than the single-modal assessment model with electrocardiographic signal and the dual-modal assessment model with eye movement and electrocardiographic signals. Removing strongly correlated feature indexes will affect model performance to some extent.