Machine Learning Based Identification of Microseismic Signals Using Characteristic Parameters.

Area of Science:

• Rock mechanics and mining engineering
• Computational intelligence and machine learning applications

Background:

• Accurate microseismic signal identification is crucial for rock engineering stability analysis in deep mines.
• Current manual classification of microseismic events and blasts relies heavily on experience, leading to potential inaccuracies and delays.

Purpose of the Study:

• To develop an intelligent and accurate microseismic signal identification system using machine learning.
• To discriminate between microseismic events and blasts for enhanced ground stress monitoring in deep mines.

Main Methods:

• A machine learning framework was established for classifying microseismic events and blasts.
• Ten algorithms (Decision Tree, Random Forest, Logistic Regression, SVM, KNN, GBDT, Naive Bayes, Bagging, AdaBoost, MLP) were evaluated.
• Data was split into 50% training and 50% test sets, with performance assessed using ACC, PPV, SEN, NPV, SPE, FAR, and ROC curves via cross-validation.

Main Results:

• Logistic Regression demonstrated the best performance, achieving cross-validation accuracy exceeding 0.95.
• Random Forest, Decision Tree, and Naive Bayes also showed strong performance.
• Expanded databases are recommended to improve accuracy and avoid data distribution issues from small training sets.

Conclusions:

• Machine learning algorithms, particularly Logistic Regression, significantly improve the accuracy and timeliness of microseismic event and blast identification.
• This AI-driven approach resolves ambiguities in manual classification based on waveform characteristics.
• The system facilitates easier acquisition of source parameters, ensuring reliable microseismic monitoring for mine safety.