This paper presents a data-driven modeling technique used to build a multi-sampling-rate RBF-ARX (MSR-RBF-ARX) model for capturing and quantifying global nonlinear characteristics of the head and tail stage drying process of a cylinder-type cut tobacco drier. In order to take full account of influence of the input variables to outlet cut tobacco moisture content in whole drying process, and meanwhile, to avoid orders of the model too large, this paper designs a special MSR-RBF-ARX model structure that incorporates the advantages of parametric model and nonparametric model in nonlinear dynamics description for the process. Considering this industrial process identification problem, a hybrid optimization algorithm is proposed to identify the MSR-RBF-ARX model using the multi-segment historical data set in different seasons and working conditions. To obtain better long-term forecasting performance of the model, a long-term forecasting performance index is introduced in the algorithm. To accelerate the computational convergence, in the hybrid algorithm, one-step predictive errors of the model are minimized first to get a set of the model parameters that are just used as the model initial parameters, and then, the model parameters are further optimized by minimizing long-term forecasting errors of the model. Based on the estimated model, a set of optimal setting curves of the input variables are obtained by optimizing parameters of the designed input variable models. The effectiveness of the proposed modeling and setting control strategy for the process are demonstrated by simulation studies.