Abstract: The rapid development of China's economy drives the rapid growth of coal consumption, but also lead to serious environmental pollution.Therefore, it is of great significance to clarify the future coal demand under environmental constraints for China to adjust the energy structure and coordinate the balance between economic growth and environmental pollution.This paper compares and demonstrates a variety of coal demand forecasting methods, and finally chooses BP neural network method based on Matlab to forecast the long-term coal demand.Based on the first level air quality standard in the Outline of The 13th Five Year Plan for National Environmental Protection and the annual reference removal rate of pollutants in the reference concentration of 100 μg/m³ in the environmental atmosphere of mainland China, the annual emission limits of SO₂ and smoke (dust) in China are calculated to be 3.23 Mt and 6.46 Mt respectively.The environmental pressure parameters are substituted into the model, supplemented by economic and energy angle parameters for model training, and then the coal demand in 2020-2050 under the preset environmental pressure, high, medium and low economic growth scenarios is predicted.The results show that China's coal demand peaked around 2025, about 3 Btce of standard coal, and China's coal demand is about 2 Btce in 2050, which is in line with the characteristics of energy demand structure in the later development stage of developed countries.