Abstract: To address the issues of lag in coal mine safety performance evaluation and static mismatch of indicators, a bottom-up dynamic evaluation framework for coal mine safety performance is constructed based on artificial intelligence algorithms. To accommodate short-cycle rapid evaluation, a cross-fusion method of prior indicators and posterior indicators is proposed during the indicator construction stage. On the basis of establishing a safety performance evaluation indicator system for underground coal mines and constructing its analytic network process (ANP) hierarchy, the grey wolf optimizer (GWO) is introduced to dynamically optimize the weights of safety performance indicators by incorporating historical prior indicators. Subsequently, the random forest (RF) and term frequency-inverse document frequency (TF-IDF) methods are jointly employed to mine hidden risk features from historical coal mine safety performance data. The developed RF-TF-IDF model achieves an average area under the curve (AUC) value of 0.98. On this basis, real-time perception and trend prediction of short-term safety performance status are realized using the long short-term memory (LSTM) algorithm. Experimental results show that the accuracy of the RF-TF-IDF model on the test set is 82.5%, and the root mean square error (RMSE) of the LSTM prediction model is 0.16, demonstrating its effectiveness in supporting performance evaluation and decision-making. The research outcomes can further improve the response speed and prediction accuracy of the coal mine safety performance appraisal system, providing a new intelligent tool system for dynamic evaluation, timely warning, and decision support in coal mine safety performance.