Abstract: Crowd sensing technology can realize the sensing and computation of people, machines, and environment in smart industrial IoT coal mines, and provide solutions for safety monitoring through distributed intelligent optimization, however, the traditional long short-term memory (LSTM) method does not consider the relationship between neighboring machines, which leads to the low accuracy of the human body's position prediction and the prediction of the pressure value, therefore, we propose the crowd sensing coal mine based on smart industrial IoT safety monitoring system. The results show that the PE algorithm used has the lowest average cumulative prediction error, the trajectory fitting rate is improved by 24.1%, 13.9% and 8.7% over Kalman (Kalman) filtering, Elman neural network and Kalman + Elman algorithms, respectively, and the RMSE values of ADI-LSTM are compared with the single-input ARIMA, RNN, LSTM and GRU, respectively reduced by 36.6%, 52%, 32% and 13.7%; and MAPE values reduced by 0.0003%, 0.9482%, 1.1844% and 0.3620%, respectively.