Abstract: The automatic identification of microseismic signals is an urgent issue to be addressed in microseismic monitoring technology, and its performance determines the accuracy and timeliness of early warning. Although machine learning methods have been extensively utilized in the identification of microseismic signals, they exhibit notable limitations in identifying original signals with low signal - to - noise ratios. Both feature sets and algorithmic models jointly determine signal recognition accuracy in machine learning approaches. However, the construction of feature sets is yet to be standardized uniformly. To address this issue, the automatic identification of microseismic signals in a metal mine was investigated based on the improved multi - criterion fusion feature selection algorithm. Firstly, a dynamic signal feature library was established, which categorizes validated recognition features into three domains. Subsequently, an improved multi - criterion fusion feature selection algorithm was applied to quantitatively score the features, through which an optimal feature subset was selected. Finally, this subset was utilized as input parameters for the particle swarm optimization-support vector machine (PSO - SVM) recognition algorithm to identify signals automatically. The results indicate that the optimal subset constructed using the improved multi - criterion fusion feature selection algorithm includes 32 features. Compared to the feature sets constructed by using traditional methods, the optimal subset contains a richer variety of feature categories. When used as signal input with few training samples, the recognition accuracy of the training and test sets is 100% and 99.23%, respectively. Feature contribution analysis revealed that time - frequency domain features have a better performance compared to time - domain and frequency - domain features. This study establishes a new paradigm for the identification of microseismic signals. It holds substantial significance in propelling the extensive utilization of microseismic monitoring techniques in engineering applications.

Key words: microseismic signals identification, feature selection, PSO-SVM algorithm, feature library, original signal