Low performance control loops imply in higher production costs, leading to off-specification production loss and unnecessary wear of the final control elements. Depending on the controlled process, the deficient loop can also lead to non-expected plant stops and even on environment damage. In this sense, model predictive control (MPC) is one of the few algorithms proved capable to stabilize processes in the presence of nonlinearities and constraints. To meet its control objectives, the classic MPC algorithm is based on an optimization problem which relies in the system's linear model. Although, the removal of the linear model from its designed condition deteriorates the model's and controller's effectiveness. Hence, the general objective of the presented work relies in the non-linear description of the system for improving the control performance. The main objectives were the study and analysis of a generalized predictive controller (GPC) operating in parallel with a linear local model network, identified by the LOLIMOT algorithm, able to adequate the controller's prediction model for the process operation range. For the quality assessment of the proposed controller, tests were evaluated in an experimental apparatus for pH control. The control strategy was implemented in an open source system for monitoring and control. Therefore, considering the static nonlinear characteristics of the process, comparative studies were applied between the traditional GPC (based on an single global model) and the proposed approach. The results showed that the dynamic network was able to effectively represent the system output, resulting in a simple structure, given the fact that the local models are indeed local ARX models. It was also shown that the GPC using the linear local model network performed satisfactorily and even better than the single model GPC. It was observed that the output calculated by the proposed controller has been considerably less aggressive than the traditional controller, leading to a considerable reduction in the system's control effort. The results showed that there was a saving up to 45% in the control effort. It was also observed that the developed monitoring and control system is suitable for real applications, since the predictive control strategy, implemented in Scilab, succeeded in solving the control problem within the sampling time even with the embedded constrained QP (quadratic programming) problem. Considering that the structure of these systems allows them to be used in similar applications to linear models, it was also proved the viability and applicability of using the linear local models network directly into advanced control algorithms already available to industry as in the GPC controllers.