Since the behavior of a complex dynamic system for a large oil field in Iraq is significantly influenced by many nonlinearities, its dependent parameters exhibit non-stationary with a very high delay time. Developing white-box modelling approaches for such dynamic oil well production cannot handle these large data sets with all dependent dimensions and their non-linear effects. Therefore, this study adopts the hybrid model that combines white-box and black-box to address such problems because the model outputs require various variable types to achieve optimal fitness to measured values. The hybrid model structure needs to evolve with changes in the physical parameters (white-box part) and Neural Networks' Weights (black-box part). The model structure of the proposed hybrid network relied on converting fuzzy rules in a Takagi–Sugeno–Kang Fuzzy System (TSK-FS) into a multilayer perceptron network (MLP). The hybrid parameters are formulated concerning six-dimensional dependent variables to describe them in matrix form or layer and by which can quantify total model outputs. After mapping categorical variables to tuples of MLP, the Gauss-Newton regression (GNR) provides an optimal update of the hybrid parameters to get the best fitting of the model outputs with the target of the dataset. The clustering technique and GNR promote predictive performance due to reducing uncertainties in the hybrid parameters. Due to time being the most effective of the independent variables for predicting oil production, datasets are classified into different clusters based on time. The actual field dataset for training and validation is collected from Zubair Oil Field (9 oil wells), which is implemented to build the proposed model. The results of the hybrid model indicate that the development of the proposed structure has achieved the high capability to represent such big data which is the most imperative feature of the proposed model. Furthermore, obtained results show its accuracy far outpacing competitors and achieving a significant improvement in predictive performance.