Abstract: This study addresses the challenges of complex and high-risk underground coal mine environments, difficulties in personnel health monitoring, and the limitations of single monitoring methods. It proposes a multi-dimensional status monitoring and real-time risk prediction system based on smart wristbands, portable monitors, and smart information-capable miner's lamps. The system collects real-time physiological parameters (e.g., heart rate, body temperature, blood oxygen saturation) and environmental parameters (e.g., methane, carbon monoxide concentrations) from underground personnel. Data is transmitted to a surface server via the mine IoT (Internet of Things) system, establishing a unified monitoring platform.For data processing, the study innovatively establishes a fatigue assessment model based on a Backpropagation Neural Network (BPNN). This model achieves precise estimation of miners' fatigue levels by analyzing the correlation between physiological parameters and perspiration pH value. Furthermore, by Conditional Score-based Diffusion Models (CSDI), the system fuses multi-dimensional data (physiological, environmental, and personnel location) to enable dynamic risk prediction and proactive health assessment.Experimental results demonstrate that the system achieves a fatigue assessment error not exceeding 0.05 and a prediction accuracy rate of 95% or higher, significantly enhancing operational safety and health management for mine personnel. This research provides an intelligent solution for coal mine safety production, holding substantial theoretical value and practical significance.