The three-level air tank system is widely used in the industrial manufacture as gas storage device. Because of its features like time-delay and great inertia, usually it is very difficult for the traditional PID controller to achieve ideal control performance. Aiming at solving practical problems, in this paper, we have proposed a constrained generalized predictive control (GPC) algorithm that the input amplitude, input-change-rate and output are constrained. In this algorithm, the system parameters are identified by recursive least squares (RLS) algorithm, and the prediction model is obtained by solving the Diophantine equations. The active set method is firstly applied to solve the quadratic programming, and then this rolling optimization strategy is used to calculate and obtain the optimal control law. Finally, compared with the other methods in physical experiments, our approach is more effective and can remarkably improve the control performance.