New insight into identification of rock Kaiser effect: A chaotic deep learning model and its application in deep mining

The acoustic emission (AE) Kaiser effect method is widely used for in-situ stress measurement because of its nondestructive nature, operational efficiency, and low cost. A key step in this method is the identification of the Kaiser point; however, traditional manual approaches still require further improvement, indicating the importance of developing intelligent identification methods. In this study, an intelligent Kaiser point identification method based on a dual-branch gated recurrent unit (GRU) deep learning framework and phase space reconstruction (PSR) is proposed. In the proposed framework, AE waveform data are processed via PSR and principal component analysis (PCA) to generate chaotic feature representations, which are then fused with the original waveform data through a dual-branch GRU architecture for classification. The classification results are subsequently used for Kaiser point identification. The proposed model achieves an accuracy of 90.7% and an area under the curve (AUC) of 0.9357 on the test set, indicating good discrimination ability between the Felicity and Kaiser areas. Compared with representative deep learning baseline models, including long short-term memory (LSTM), bidirectional LSTM (BiLSTM), temporal convolutional network (TCN), and Transformer, the proposed model exhibits superior overall performance. The ablation results confirm the positive contributions of both the original waveform branch and the chaotic feature branch, while the interpretability analysis reveals that the model relies mainly on a limited number of salient waveform segments and chaotic feature groups, both of which are physically relevant. Moreover, compared with the traditional manual method, the proposed method identifies the Kaiser point more stably and accurately. Furthermore, the proposed method was applied to a porphyry copper mine in western China, and the results demonstrate its practicability in engineering applications.