The reliability of artificial pancreas is crucial for the safety and security of type 1 diabetes. In this paper, new modeling and analysis of the closed-loop glucose control system are proposed to verify and evaluate the performance of control algorithms at the system level. Priced timed automata are used to model the physiological processes, the control algorithm, and the adversary. Two control algorithms are evaluated in normal condition and under replay attack using simulation and model checking. The results show that one of the control algorithms outperforms the other in terms of safety properties. The results also demonstrate the impact of the target attack on the system behavior. The proposed approach provides a new way to evaluate sophisticated control algorithms, security attacks, and attack mitigation techniques.