Abstract: To obtain effective precursor information for disasters, classification of microseismic events is a necessary prerequisite. In response to the low automatic classification accuracy between rock failure signals and blasting signals,a feature extraction and classification method for mine microseismic signals based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) is proposed. First,the CEEMDAN is employed to obtain the multi-order IMF components of the microseismic signal. Subsequently,the main components are selected through correlation coefficient analysis. The variance contribution rate and energy spectrum coefficients of each main component are then calculated,serving as feature vectors for classification learning. Finally,the rock failure signals and blasting signals are classified using a Convolutional Long Short-Term Memory Neural Network (CNN-LSTM) optimized by the Whale Optimization Algorithm (WOA). Results show that the main components of CEEMDAN are from PC1 to PC8. As the decomposition layers increase,the average variance contribution rate and energy spectrum coefficients of the rock failure signals first increase and then decrease,while those of the blasting signals show a declining trend. Compared to the correlation coefficient and variance contribution rate, the feature vector energy spectrum coefficient,when used as an input for WOA-CNN-LSTM,Support Vector Machine (SVM),and BP neural network,has the highest classification accuracy. The classification performance of WOA-CNN-LSTM is significantly better than the Bayes method,SVM,and BP neural network. Moreover,the classification accuracy of WOA-CNN-LSTM based on the main component energy spectrum coefficient reached 91. 50%.