Fatigue discrimination of workers in low oxygen environment: a multi-feature fusion method

Abstract

Many plateau areas have abundant fossil energy resources, but the low oxygen environment in the plateau seriously limits the development of exploration work. Monitoring the fatigue status of exploration workers on the plateau is a necessary measure to ensure their safety. In this study, we designed a task-based experiment to explore the behavior of fatigue in exploration operations, with environmental oxygen partial pressure as the variable. Physiological signals including electrocardiogram (ECG), respiration, and oxygen saturation were collected, along with subjective fatigue scale scores. A total of 96 features were extracted from the time domain, frequency domain, and nonlinear domain, and a dataset of fatigue feature for exploration tasks in plateau was constructed. Based on different dimensionality reduction methods and classifiers, a model for discriminating the fatigue state of exploration tasks in plateau was constructed and optimized. The results show that 80 % of the workers had a severe or extreme fatigue state before the end of the task in a 17 % oxygen concentration (Oxygen partial pressure 17.23 kPa) environment. Correlation analysis and XGBoost algorithm show that all signals provide effective contributions to discriminating the fatigue state. Among them, the ECG signal plays a leading role. Finally, through the model building and optimization route of Principal Component Analysis-Support Vector Machines-Differential Evolution Algorithm (PCA-SVM-DE), we constructed a multi-feature information fusion geological exploration operation fatigue state discrimination model. The model achieves an accuracy of 99.01 % in the four-level fatigue classification. The fatigue state discrimination method can be effectively applied in low oxygen working environment to reduce operational risks.