In this work the problem of regulating glycemia in type I diabetic patients is studied by means of an impulsive zone model predictive control (impulsive ZMPC) based on a novel long-term glucose-insulin model. Taking advantage of the model - which features real life properties of diabetes patients that some other popular models do not - the proposed control ensures the stability under moderate-to-severe disturbances. A long-term scenario - including meals - are simulated, and the results appear to be satisfactory as long as every hyperglycemia and hypoglycemia episodes are suitably controlled.