The paper presents a combination modeling procedure and the implementation of a nonlinear predictive control scheme for the optimization of industrial chemical processes. The model structure is first based on a simple step response method. This provides a way to use prior knowledge about the dynamics, which has a general validity, while additional information about the process behavior is derived from measured plant data-model error. This data error driven model framework is applicable for a wide range of chemical operating units under a certain control policy. The same idea is also used to solve the online optimization problem in the predictive controller. The efficiency and effectiveness of the modeling training algorithm and the nonlinear predictive control approach are demonstrated through a coke furnace case study. A good model fitting for the nonlinear plant is obtained by using the new method. A comparison with traditional approaches shows that the new algorithm can considerably reduce modeling error and improve control accuracy.