A study on the classification and prediction of firefighter's operational fatigue level

Firefighting operations in high-rise building fires require firefighters to navigate complex environments while undertaking physically demanding, heavy-load tasks, which often lead to severe fatigue, impairing their operational efficiency and decision-making. This study aims to develop a robust fatigue classification and prediction model to assess and forecast firefighters' fatigue levels. Key metrics, including electrocardiogram (ECG) signals, subjective fatigue ratings, and reaction time data, were utilized. Experiments involving six healthy adult male participants simulated firefighting scenarios, during which subjective fatigue levels (6-20 Borg's RPE scale) and reaction times were recorded. A five-level fatigue classification was established using K-means clustering, and entropy weight analysis was applied to define a comprehensive fatigue index (F), enabling a three-tier fatigue classification: light, moderate, and severe fatigue. A BP neural network was employed for dynamic fatigue prediction, with 10 features derived from heart rate and heart rate variability (HRV) metrics serving as inputs and the comprehensive fatigue index (F) as the output. The BP neural network model achieved a high prediction accuracy with an R² value of 93.24%, demonstrating its capability to accurately predict firefighters' fatigue states. This approach provides a scientific basis for optimizing firefighter training protocols and enhancing operational effectiveness during fire rescue missions. The findings highlight the significant potential of this method for advancing firefighter fatigue monitoring and management.