Carbon emission from the thermal power generation sector is one of the main contributors to global warming. It is critical to objectively evaluate the influence of relevant factors on carbon emissions, which may be beneficial for the implementation of policies and the formulation of development plans. However, the current decomposition approach based on production-theoretical decomposition analysis (PDA) and weak disposability may cause deviation on potential emission reduction. This paper proposes a modified PDA model under the assumption of semi-disposability and decomposes the carbon emission changes of China's thermal power generation sector accordingly. The results show that provinces, which have the highest non-disposal degrees during the study period, are mostly located along the east coast of China. The growth in installed capacity is identified as the largest contributor to the increase in national carbon emissions, while the decrease in capacity utilization rate significantly benefits the carbon emission reduction. A cluster analysis is also conducted to give policy implications for provinces with similar characteristics. Compared with traditional methods under strong and weak disposability assumption, the proposed model shows great advantages in identifying the potential emission reduction and reflecting the real production process, which makes it more suitable for decomposing the changes of carbon emissions and other environmental pollutants.