A Model Predictive Control (MPC) is a system which allows us to control a production plant. Thanks to this type of system is possible to make a production that comes close to “zero defects”. In order to achieve its main goal, this kind of systems consists of several phases. One of the most important is the phase that predicts the situation in which the plant is going to be in a given time. Currently, the majority of the research in this field are related to linear MPC, although the process, which the model tries to represent, may not be. Thus, this paper presents several experiments that proof that the forecast phase, usually represented by a single mathematical function, can be represented by machine-learning models.