It is difficult to establish a dynamic model of quality variables in polymerization, because of the high complexity of the nonlinear process behavior and time-consuming laboratory quality measurements. A decomposition scheme is presented in this paper to model the dynamic nonlinear behaviors of the gas-phase polyethylene process; by decomposing the overall process based on its structure into three relatively simpler problems. However, difficulties still come from nonlinearity and nonuniform sampling. Especially, the samples of the product quality in the transitional process are too deficient for traditional data-driven methods. In order to further overcome the difficulties incurred by nonlinearity, the adaptive hinging hyperplanes (AHH) model is adopted. Meanwhile, an impulse response template (IRT) is introduced to model the accumulation process dynamics to deal with the scarcity of sampled data. The proposed model is validated to be effective in comparison with black-box models on the melt index data of 10 transitional processes.