An improved generalized predictive control algorithm is presented in this paper by incorporating offlineidentification into online identification. Unlike the existing generalized predictive control algorithms, the proposedapproach divides parameters of a predictive model into the time invariant and time-varying ones, which are treatedrespectively by offiine and online identification algorithms. Therefore, both the reliability and accuracy of thepredictive model are improved. Two simulation examples of control of a fixed bed reactor show that this newalgorithm is not only reliable and stable in the case of uncertainties and abnormal disturbances, but also adaptableto slow time varying processes.