Abstract: The unsafe psychological state of miners is a significant precursor to accidents in coal mines. To identify this precursor, a study on the early warning of miners' unsafe psychological states was conducted based on grounded theory, 47 initial documents were coded to establish a warning indicator system comprising 3 core categories and 8 main categories related to miners' unsafe psychological states. Data were collected through a questionnaire survey, and the CRITIC-Entropy method was employed to determine the weights of the indicators subsequently, a three-layer BP neural network early warning model, structured as "8-20-4" was developed for miners' unsafe psychological states. Experimental results indicate that the unsafe psychological state of miners is influenced by individual, operational, and organizational factors, with workload, physical condition, work environment, and interpersonal relationships identified as key triggering factors. The proposed early warning model achieved an accuracy rate of 93.7%, effectively identifying the levels of miners' unsafe psychological states. Moreover, a "warning-countermeasure" mechanism was proposed based on the psychological warning results for miners.