A Diffusion Model Integrating Physical and Physiological Priors: Sudden Fatigue Event Detection Based on Conformal Risk Control

Fatigue is a critical factor affecting athletes' health and work safety. Accurately predicting and timely detecting sudden fatigue events is of great significance. Existing fatigue monitoring studies primarily focus on single physiological indicators or rely on subjective questionnaires, which struggle to capture complex physical-physiological coupling characteristics and exhibit insufficient sensitivity to sudden events. This paper proposes a diffusion model that integrates physical and physiological priors. It employs Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) to process multimodal signals, incorporates physical and physiological priors based on energy conservation and heart rate-load coupling, and constrains the risk of sudden event detection through a conformal risk control framework. The model achieves precise modeling of fatigue features across different frequency bands using cross-modal bilinear attention fusion and an adaptive diffusion sampling strategy. The proposed risk control mechanism dynamically adjusts confidence intervals during the inference stage to ensure detection reliability. Experimental results demonstrate that the proposed method achieves significant performance improvements on our constructed 12-week multimodal fatigue monitoring dataset, reducing RMSE, MAE, and MAPE by 20--35\\% compared to the best baseline. The recall rate for high-fatigue events reaches 96.5\\%, validating the effectiveness and superiority of the diffusion model integrating priors and risk control for sudden fatigue event detection.