In view of the abrupt and phased features of natural gas consumption, this paper attempts to predict natural gas consumption in China with a refined forecasting approach. First, we establish a Markov switching (MS) model to identify the phase characteristics after eliminating change points in the natural gas consumption sequence, using the product partition model (PPM). The results show that there are"rapid growth"and"slow growth"regimes in the development process of natural gas consumption in China. Second, the Bayesian model average (BMA) method is employed to determine the core determinants of natural gas consumption under sub-regimes, and it is determined that there are significant differences in the influencing factors under different regimes and periods. Third, this paper establishes the BMA model of the"rapid growth"regime after predicting the state of future natural gas consumption in China. We find that, compared to some other models, the BMA model that fully recognizes the regime without considering change points has the best predictive performance. Finally, the results of static and dynamic scenario analyses show that natural gas consumption continues to rise in 2019 and has obvious seasonal charac-teristics, while possible ultra-rapid growth of consumption in the future provides a new requirement for the supply of natural gas.