A multi-source physiological data-driven method for assessing the hazard perception level of operators.

With the continuous advancement of coal mine intelligence, unmanned operations at the working face have become an inevitable trend, and the safety monitoring center plays a central role in ensuring mine safety. Its operation relies not only on system monitoring but also on operators’ timely perception and judgment of hazard information. Therefore, the hazard perception level of operators is critical for accident prevention and safety management. In this context, an experiment is conducted to assess the hazard perception levels of safety monitoring center operators. Physiological signals, including electroencephalogram (EEG), electrodermal activity (EDA), and heart rate variability (HRV), are collected during hazard perception tasks. Using these physiological signals in conjunction with machine learning and deep learning methods, a high-accuracy hazard perception assessment model is developed, and its performance is evaluated using reliable metrics. The results show that certain features derived from all three physiological signals exhibit distinct responses corresponding to different hazard perception levels. The LightGBM model achieves the highest performance, with an accuracy of 99.89%. Furthermore, the study demonstrates the feasibility of high-accuracy hazard perception assessment based on the combined use of EEG, EDA, and HRV signals. These findings provide a practical pathway for future coal mine safety management, enhance the capacity for accident prevention in intelligent coal mines, and offer strong support for safe production.