To accurately predict the time series having limited data with stochastic disturbances and nonlinearity, this paper proposed a composite forecasting model by adaptive data preprocessing and optimized nonlinear grey Bernoulli model. Specifically, improvements in the proposed model lie in the following aspects: firstly, the new-information-based buffer operator is utilized to eliminate stochastic disturbance effects and thereby enhance the smoothness of the system series, which can effectively extract the potential patterns of recent development. Secondly, based on the preprocessed data, the nonlinear grey Bernoulli model provided with the newly-designed initial condition that abides by the principle of new information priority without data lapses enables the forecasting precision and applicability enhancement. Thirdly, due to the nonlinear relationships between the prediction errors and the parameters introduced by the above optimization paths, the Particle Swarm Optimization algorithm is employed to ascertain the optimal parameters simultaneously, whose efficacy and robustness have been validated by experimental comparison and sensitivity analysis. On the foundation of the above functional improvements, the novel composite forecasting framework successfully overcomes the limitations of conventional grey models to obtain more accurate and robust forecasts, which are substantiated by the applications for predicting the sales of new energy vehicles in China and Norway. Experimental results and discussions demonstrate that the new-information-based buffer operator can be regarded as an excellent alternative tool for data preprocessing. Besides, the grey models with buffered preprocessing can deliver much more accurate forecasts than their corresponding versions without buffered preprocessing. Furthermore, the proposed model with buffered preprocessing outperforms any of the other competitors and can be considered as the most promising technique for prognosticating NEVs sales.