Microseismic monitoring technologies, renowned for their high sensitivity and resolution, serve as an essential tool for monitoring sudden geological hazards. However, in practical applications, microseismic monitoring signals (MMSs) often exhibit complex and diverse patterns due to the combined influence of construction interference and genuine geological events. Traditional methods, relying on simplistic microseismic signal features (MSFs) and a single model, face significant challenges in accurately identifying multiple microseismic signal patterns (MSPs). To address this problem, this study proposes an intelligent recognition framework based on multidimensional feature optimization and a collaborative hybrid model. The model was developed and tested using onsite MMS data collected from a steep rock slope in Guiyang, Guizhou, China. The methodology involved the extraction of multidirectional MSFs, division of potential MSPs via self-organizing map (SOM)–K-means clustering, calibration of typical MSPs, performance comparison of multiple models (namely, support vector machines, backpropagation neural networks, convolutional neural networks, and random forests), and SHapley Additive exPlanations (SHAP) analysis of the optimal recognition model. The effectiveness of the model was ultimately validated via four consecutive days of continuous data monitoring. The results demonstrated that this intelligent recognition method achieves high precision and stability in identifying multiple MSPs, including excavator operations, drilling operations, rockfalls, downtimes, and rock mass fractures, under complex conditions. This approach offers novel insights for intelligent microseismic monitoring and early warning of sudden geological hazards.
An et al. (2026) studied this question.