Abstract: This article first studied the thermal decomposition behavior of four types of coal under solvent extraction and non-solvent extraction conditions using a thermogravimetric analyzer under non-isothermal conditions. The kinetic parameters and activation energy of coal were determined using four methods: Arrhenius, Coats Redfern, Doyle Ozawa, and Friedman. The experimental results showed that solvent extraction of coal reduced the activation energy by about 50%. Pearson correlation analysis was conducted on the chemical composition and ignition performance of 16 typical coals and 48 mixed coals. The results showed that the moisture, volatile matter, fixed carbon, calorific value, oxygen, and carbon content of coal were the most relevant factors affecting ignition temperature and activation energy. Secondly, this article proposes a three-layer BP neural network model and trains it on 100 coal samples. The trained model is then used to predict the ignition characteristics of 15 mixed coals. The experimental results show that the model has high prediction accuracy, with relative average errors of 0.42% and 4.49% for predicting ignition temperature and activation energy, and mean square deviations of 9.89 and 103.82, respectively. This has certain guiding significance for further blending of mixed coals to improve the ignition characteristics of raw coal in the future.