The aim of this work is to present a technique, which is capable of dealing with discontinuous time varying systems based on the identification of local linear models at different operating conditions. The ideal process trajectory is predicted using a Model Predictive Control Scheme (MPC) in cascade with conventional PID controllers which are responsible for driving the system along at the optimal conditions. An on-line application to a laboratory scale fermenter for the production of gluconic acid is discussed to assess the reliability and performance of this strategy.