Research on the DAS Recognition Method for Blockages in Long‐Distance Downhole Backfilling Pipelines of Goaf Coal Gangue Slurry

Against the background of the continuous expansion of deep coal mining scale, goaf stability control and gangue disposal have become key issues restricting the safe and efficient advancement of engineering projects. The underground goaf gangue slurry backfilling technology, as a core method for gangue disposal in deep mining, not only fundamentally avoids pollution problems such as dust and land occupation caused by hoisting gangue to the ground but also significantly reduces economic costs by simplifying the gangue transportation process, thereby providing an effective approach to balance the green development and economic efficiency of deep mining. However, due to the characteristics of high solid‐phase content and uneven particle gradation of coal gangue slurry, failures such as blockage and corrosion are prone to occur during pipeline transportation. Among these issues, accurate localization and early warning of blockage points directly affect the continuous operation of the backfilling system and have become a core technical bottleneck in the engineering practice and technical research of deep goaf backfilling. Existing monitoring technologies are limited by factors such as slurry flow field disturbance and weak blockage signal characteristics, making it difficult to accurately capture the evolution process of blockages in pipelines and unable to meet the engineering requirements for timeliness and accuracy of fault early warning in deep backfilling projects. To address this problem, this study proposes a collaborative monitoring scheme integrating distributed acoustic sensing (DAS) technology and the patch‐based time series transformer (PatchTST) time series prediction network for the scenario of slurry transportation in deep goaf backfilling: The DAS technology is used to collect acoustic signals in real time during the blockage evolution process in pipelines, so as to accurately capture the changes in signal characteristics under different blockage degrees; meanwhile, the PatchTST network is employed to perform in‐depth feature extraction and time series prediction on blockage data under multiple working conditions, thereby constructing an early warning model for blockage risks and achieving a technological breakthrough from “passive inspection” to “active early warning.” To verify the applicability of the proposed scheme in the deep goaf backfilling scenario, a 15.14‐m looped pipeline test system was built to simulate the working conditions of deep backfilling pipelines, and tests with four different blockage degrees (20%, 40%, 60%, and 80%) were carried out. The results show that the PatchTST network achieves prediction accuracies of 99.48%, 99.33%, 98.98%, and 98.95% for the four blockage degrees, respectively, demonstrating its ability to accurately identify the blockage degree; under the complex working condition of dynamic changes in blockage degree, the continuous prediction accuracy for a single blockage point remains 90.88%, which can effectively track the evolution trend of blockages. In conclusion, this collaborative scheme provides a reliable technical approach for the accurate monitoring and early warning of blockages in long‐distance backfilling pipelines of goaf gangue slurry, which can ensure the stable operation of the deep backfilling system and possesses significant engineering application value and promotion potential.