We found a match
Your institution may have access to this item. Find your institution then sign in to continue.
- Title
CHIM-Net: A Combined Hierarchical Information Model for Predicting Time, Space and Intensity of Mining Microseismic Events: CHIM-Net: A Combined Hierarchical Information Model for Predicting Time....: H. Luo et al.
- Authors
Luo, Hao; Zhang, Huan; Pan, Yishan; Dai, Lianpeng; Kong, Chao; Bai, Mingyu
- Abstract
During the coal-mining process, high-energy mining microseismic events constrain the safety production of coal mines. To address the issues of low accuracy in predicting the time, space, and intensity (energy) of mining microseismic events, as well as insufficient feature extraction of mining microseismic monitoring data, the Combined Hierarchical Information Modeling (CHIM-Net) is proposed, integrating deep learning theory and technology. This model consists of a data decomposition module, a data partitioning module, and two prediction branch modules. The original mining microseismic monitoring data is decomposed and segmented, and then input into different prediction modules for training to obtain the trained model, ultimately obtaining the prediction results. To evaluate the model, mining microseismic monitoring data collected from coal mines in different provinces were used to train and test the model. Compared with several models such as Autoformer and Informer, the effectiveness of this model in the field of mining microseismic prediction is verified. The ablation experiment demonstrated the effectiveness of the data decomposition module in improving prediction performance. The results show that when at a prediction length of 48, the model achieves an average reduction of 18.92% in mean squared error and 10.57% in mean absolute error compared to other models, while at a prediction length of 96, it reduces by an average of 8.46% and 10.23%, respectively. Experimental results demonstrate that the proposed model performs well in predicting high-energy mining microseismic events, providing valuable insights for mining microseismic prediction and early warning. Highlights: To reduce the volatility of the original data, a decomposition method is employed to partition the non-stationary mining microseismic events sequence data into multiple stable sub-sequences of different scales. These sub-sequences, compared to the original sequence, are easier to predict, which helps improve prediction accuracy. Currently, research on predicting mining microseismic events is limited, especially in terms of the application of deep learning techniques. Most existing methods rely on building mathematical or physical models for prediction. This paper adopts a method of decomposing and merging to predict mining microseismic events. Considering that the decomposed sub-sequences have different characteristics, this paper analyzes the Hurst exponent of each sub-signal, dividing them into fluctuation components and stable components. These components are input into different models for prediction, effectively utilizing the characteristics of each component to improve prediction accuracy. The model proposed in this paper does not require specific geological conditions or coal mining situations. It only needs access to mining microseismic monitoring data to be utilized. According to experiments, compared to other time series forecasting models, the model proposed in this paper exhibits higher prediction accuracy.
- Subjects
COAL mining; COAL mining safety; INFORMATION modeling; DECOMPOSITION method; PREDICTION models
- Publication
Rock Mechanics & Rock Engineering, 2025, Vol 58, Issue 1, p447
- ISSN
0723-2632
- Publication type
Academic Journal
- DOI
10.1007/s00603-024-04179-9