Blood glucose prediction algorithm via PINN under Bergman’s Minimal Model Constraints

The blood glucose regulation of patients with type 1 diabetes is complex, and there is an urgent need for accurate prediction. Data - driven models have problems such as neglecting physiological mechanisms and poor generalization ability in sparse data scenarios. The traditional Bergman Minimal Model has insufficient adaptability. This paper proposes an innovative blood glucose prediction algorithm based on a Physics - Informed Neural Network constrained by the Bergman Minimal Model. This algorithm constructs a dual - channel neural network architecture, integrating the flexibility of data - driven models and the mechanistic constraints of physiological models. Experiments on the Shanghai T1DM patient dataset show that this model outperforms traditional models in terms of prediction accuracy and physiological plausibility, providing a new and effective method for blood glucose prediction and diabetes management.