CEEMDAN-LSTM framework for distinguishing natural and mining-induced seismicity

Accurate identification of mine-induced seismic events and natural earthquakes is of significant importance. Using a purely natural earthquake catalog for advanced seismological studies such as active fault delineation, strong earthquake prediction, and stress field calculation is important; on the other hand, establishing a comprehensive mine-induced seismic event catalog aids regulatory oversight for mining activities. Focuses on the spectral characteristics of mine-induced seismic events and natural earthquakes, by identifying short-period surface waves in the data and utilizing an improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) method, combined with Long Short-Term Memory (LSTM) networks, we conducted a discriminative analysis of seismic events in Liaoning, China, and Japan. Initially, waveform data of both mine-induced seismic events and natural earthquakes underwent preprocessing steps such as baseline correction and P-wave arrival time determination; CEEMDAN was then used to decompose the data into different Intrinsic Mode Functions (IMFs), and the variance contribution rate of the IMFs was extracted as a feature to identify the short-period surface wave components of mine-induced seismic events; These features were subsequently input into an LSTM model for classification training, resulting in an accurate classification model. The results demonstrate that the CEEMDAN-LSTM model effectively addresses the issue of noise and short-period surface wave aliasing within the modes, significantly enhancing classification accuracy. The classification success rate improved markedly. Additionally, explored the advantages and limitations of various classification features and models, providing effective technical support and new perspectives for the automatic identification of seismic events in the future. This not only enhances the understanding of the characteristics of mine-induced seismic events and natural earthquakes, but it also provides a scientific basis for earthquake early warning and disaster prevention, and suggests that future research can further optimize the model in terms of speed and apply the model to classify more non-natural seismic events.