Research on Section Coal Pillar Deformation Prediction Based on Fiber Optic Sensing Monitoring and Machine Learning Algorithms

The mining face under the close coal seam group is affected by the superposition of the concentrated stress of the overlying residual diagonally intersecting coal pillar and the mining stress, which can easily cause the instability and damage of the section coal pillars during the process of mining back to the downward face. Additionally, the traditional methods of monitoring such as numerical simulation, drilling peeping, and acoustic emission fail to realize the real-time and accurate deformation monitoring of the internal deformation of the section coal pillars. The introduction of the drill-hole-implanted fiber-optic grating monitoring method can realize real-time deformation monitoring for the whole area inside the coal pillar, which solves the short board problem of coal pillar deformation monitoring. However, fiber-optic monitoring is easily disturbed by the external environment, which is especially sensitive to the background noise of the complex underground mining environment. Therefore, taking the live chicken and rabbit well of Shaanxi Daliuta Coal Mine as the engineering background, the ensemble empirical modal decomposition (EEMD) is introduced for primary noise reduction and signal reconstruction by the threshold determination (DE) algorithm, and then the singular matrix decomposition (SVD) is introduced for secondary noise reduction. Finally, a machine learning algorithm is combined with the noise reduction algorithm for the prediction of the fiber grating strain signals of coal pillar in a zone, and DBO-LSTM-BP is constructed as the prediction model. The experimental results demonstrate that compared with the other two noise reduction prediction models, the SNR of the EEMD-DE-SVD-DBO-LSTM-BP model is improved by 0.8–2.3 dB on average, and the prediction accuracy is in the range of 88–99%, which realizes the over-advanced prediction of the deformation state of the coal column in the section.