Fatigue Detection of Air Traffic Controllers Based on Radiotelephony Communications and Self-Adaption Quantum Genetic Algorithm Optimization Ensemble Learning

Air traffic controller (ATC) fatigue has become a major cause of air traffic accidents. Speech-based fatigue-state detection is proposed in this paper. The speech signal is preprocessed to further extract the Mel frequency cepstrum coefficient (MFCC) from speech discourse. The machine learning method is used in fatigue detection. However, single machine learning fatigue detection methods often have low detection accuracy. To solve this problem, an ensemble learning method based on self-adaption quantum genetic algorithm (SQGA) heterogeneous learning methods is proposed. Pattern-level and feature-level resampling are used to increase the differences in the base learner's training dataset. To enlarge the diversity of single learners, k-nearest neighbor (KNN), Bayesian network (BN), back propagation neural network (BPNN) and support vector machine (SVM) are adopted for the heterogeneous ensemble. On this basis, finally, the detection result is obtained by weighted summation. The weight of each base learner was determined by SQGA. The SQGA method combines the quantum genetic algorithm with the adaptive strategy. The adaptive strategy includes adaptive adjustment of the quantum rotation gate, adaptive generation of crossover probability and adaptive generation of mutation probability. The experiments on real civil aviation radio land-air communication show that the proposed method can obtain 98.5% detection accuracy, with a 1.2% false and 3.0% missing report rate, whereas the SVM only obtains 94.0% detection accuracy, with a 5.4% false and 9.0% missing report rate.