@article{LIN2026, 
author = {Peng LIN and Suping PENG and Na XU and Yulin LIU},
title = {Design of an integrated seismic structural intelligent interpretation experiment based on machine learning},
year = {2026},
journal = {Experimental Technology and Management},
volume = {43},
number = {9},
pages = {208-217},
keywords = {geophysical simulation teaching, seismic data interpretation, fault identification, machine learning},
url = {https://www.sciopen.com/article/10.16791/j.cnki.sjg.2026.09.026},
doi = {10.16791/j.cnki.sjg.2026.09.026},
abstract = {ObjectiveWith the implementation of the “Transparent Mine” and “Intelligent Mining” strategies, seismic geological interpretation in coal mines is shifting toward intelligent and data-driven approaches. However, traditional geophysics course experiments on fault identification still rely heavily on manual interpretation and seismic attribute analysis and lack practical training in intelligent modeling. To address this gap, this study designs an integrated experimental teaching scheme for the intelligent interpretation of seismic structural data based on a machine learning model. The aim is to cultivate students’ capabilities in data-driven geological analysis and intelligent fault recognition.MethodsThe proposed experiment is built upon the PCA-ISSA-SVM model, which integrates principal component analysis (PCA) for feature dimensionality reduction, an improved sparrow search algorithm (ISSA) for hyperparameter optimization, and a support vector machine (SVM) for fault classification. The experimental process comprises six modules: data construction, feature extraction and dimensionality reduction, model optimization, model training, multi-model comparison, and engineering validation using real seismic data. A synthetic fault model was first constructed to generate seismic data. Six seismic attributes (variance, instantaneous phase, frequency, reflection intensity, chaos, and root-mean-square amplitude) were extracted from the generated data. PCA was then applied to reduce data redundancy and select three principal components that retained 86.23% of the original information. The ISSA was used to optimize the SVM’s penalty factor c and kernel parameter g, with a population size of 50 and a maximum of 100 iterations. The optimized model was compared with the GA-SVM and PSO-SVM models. Finally, the trained model was validated using real 3D seismic data from a mining area in eastern Yunnan, China, with a dataset of 325 labeled samples.ResultsExperimental results on synthetic data showed that the PCA-ISSA-SVM model achieved a classification accuracy of 99.63%, a recall of 100%, and an F1-score of 0.986, notably outperforming the ISSA-SVM model without PCA, which achieved only 95.19% accuracy, 91.67% recall, and an F1-score of 0.835. The confusion matrix indicated that only one non-fault point was misclassified in the PCA-ISSA-SVM model, demonstrating the effectiveness of feature dimensionality reduction in improving model performance. Compared with PCA-GA-SVM (accuracy 97.04%, recall 91.67%, F1-score 0.880) and PCA-PSO-SVM (accuracy 96.30%, recall 88.89%, F1-score 0.843), the PCA-ISSA-SVM model shows higher convergence speed and better classification stability, with the lowest initial classification error and the fastest stabilization during iteration. On real seismic data, the model maintained a high recognition performance, with an accuracy of 93.52%, a recall of 96.30%, and an F1-score of 0.937, despite the increased noise and complexity of real-world conditions.ConclusionsThe integrated experimental design successfully bridges the gap between traditional seismic interpretation teaching and modern intelligent modeling techniques. It provides students with a complete workflow from data construction, feature engineering, and model optimization to engineering validation. Through comparative experiments, students gain a deep understanding of how feature processing and parameter optimization affect model performance. The use of both synthetic and real data enhances students’ ability to handle complex geological conditions and fosters data-driven geophysical thinking. This teaching reform not only improves students’ understanding of machine learning algorithms but also equips them with practical skills for intelligent coal mine construction, offering a valuable reference for modernizing geophysics education.}
}