Empirical models tend to suffer from unreliable extrapolation behavior, and this presents an issue when they are applied in model-based controller strategies such as nonlinear model predictive control (NMPC). This paper presents the development and implementation of the parallel OBF-NN model in the NMPC framework. The aim is to evaluate the applicability and the potential extrapolation benefits of the model in a closed-loop environment. For this purpose, closed-loop performance comparison is analyzed between the parallel OBF-NN and the conventional neural networks (NN) models. Results on two nonlinear case studies show that the NMPC based on the parallel OBF-NN model notably improved the closed-loop performance in the extrapolated regions of operation when compared to NMPC based on the conventional NN model without the need for re-training or any adaptive scheme.