It is the prerequisite for artificial pancreas (AP) to predict the excursion of glucose and insulin concentrations in the plasma, while only the subcutaneous glucose levels can be measured in real-time by Continuous Glucose Monitoring System (CGMS). The purpose of this study is to predict glucose and insulin concentrations excursion in plasma properly, only using CGMS data and insulin infusion rate data. Firstly, Ruan model, a glucose-insulin model is introduced to build a state space to describe the dynamics of insulin transportation, glucose-insulin interaction,and glucose transportation, taking the carbohydrate intake into consideration. Then, considering the inter- and intra-individual variability, we incorporate parameter uncertainty to describe the high variability, and extend the observable parameters to extended state variable, which can be updatated according to the real-time collected data. Next, three types of Bayesian filter observers, Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), and Particle Filter (PF) are designed to track all the states dynamically. Finally, based on the updated states, the glucose and insulin concentrations in plasma are estimated and predicted. We evaluate this methodology with an in-silico study with 30 Type-1 diabetes patients during one week. It is proven that our method can track the excursion of glucose and insulin concentrations in the plasma effectively. Moreover, the PF outperforms other two filters in predicting the glucose and insulin concentrations, with MSE of 1.56 ± 0.69 mmol2/L2 and 11.703.14 mU2/L2 respectively.