Hebei State Key Laboratory of Mine Disaster Prevention, North China Institute of Science and Technology
College of Construction Engineering, Jilin University
School of Energy and Mining Engineering, China University of Mining and Technology (Beijing)

矿井工作面涌水量预测对确保矿山安全、优化资源配置、提高工作效率等都具有重要作用。为提高预测结果的准确性和稳定性，基于钻孔水位和微震能量数据与涌水量的强关联性，选择其作为多因素特征变量，提出SSA-CG-Attention多因素矿井工作面涌水量预测模型。该模型在门控循环单元(Gated Recurrent Unit，GRU)提取时序特征的基础上，与卷积神经网络(Convolutional Neural Network，CNN)融合形成新的网络结构提取数据的有效非线性局部特征，并且加入注意力机制(Attention)，在预测过程中将注意力集中在输入元素上，提高模型的准确性。最后通过麻雀搜索算法(Sparrow Search Algorithm，SSA)优化模型参数，避免局部最优解的问题。将提出的模型分别与传统的BP神经网络、LSTM、GRU单因素涌水量预测模型以及MLP、SLP、SVR、LSTM、GRU、SSA-LSTM、SSA-GRU多因素涌水量预测模型的预测结果进行对比分析，结果表明：SSA算法以最少迭代次数快速寻优，避免了局部最优解的缺陷；SSA-CG-Attention多因素涌水量预测模型整体预测指标绝对误差(EMA)、均方根误差(ERMS)以及平均绝对百分比误差(EMAP)分别为5.24 m3/h、7.25 m3/h、6%，指标方差和为8.90。相较于其他预测模型预测精度更高，相较于单因素涌水量预测模型，多因素涌水量预测模型预测结果更加稳定。研究结果为矿井工作面涌水量预测提供了新的思路与方法，对矿井工作面涌水量预测及防控有着借鉴与指导作用，具有一定的理论价值和现实意义。

Predicting mine water inflow plays an important role in ensuring mine safety, optimizing resource allocation, and improving work efficiency. This study aims to improve the accuracy and stability of the predicted mine water inflow. Given their strong correlations with water inflow, borehole water level and microseismic energy were chosen as multifactor characteristic variables. Using these variables, this study developed the SSA-CG-Attention multifactor prediction model for mine water inflow along mining faces. The new model extracted effective nonlinear local features of data utilizing a new network structure, which was formed by integrating a convolutional neural network (CNN) based on the time sequence features extracted with a gated recurrent unit (GRU). Furthermore, this model introduced the attention mechanism to focus on input elements during prediction, thus improving the prediction accuracy. Finally, the sparrow search algorithm (SSA) was employed to optimize the model parameters and avoid the occurrence of locally optimal solutions. The new model was compared with traditional single-factor prediction models, including BP neural network, LSTM, and GRU, and multifactor prediction models, consisting of MLP, SLP, SVR, LSTM, GRU, SSA-LSTM, and SSA-GRU. The results indicate that the SSA algorithm allowed for quick optimization within the fewest iterations, thus ruling out the possibility of locally optimal solutions. The new model yielded an absolute error (EMA), a root mean square error (ERMS), and a mean absolute percentage error (EMAP) of 5.24 m3/h, 7.25 m3/h, and 6%, respectively, with a variance sum of 8.9. Furthermore, this model exhibited higher prediction accuracy than other prediction models, and the multifactor prediction models yielded more stable predicted results compared to the single-factor ones. The results of this study provide a new philosophy and methodology for the prediction of mine water inflow along mining faces and serve as a reference and guide for its prediction, prevention, and control, holding theoretical and practical significance.

国家自然科学基金项目(51974126，51774136)；教育部“创新团队发展计划”滚动支持项目(IRT_17R37)；中央高校基金项目(3142022003)

丁莹莹，尹尚先，连会青，等. 基于SSA-CG-Attention模型的多因素采煤工作面涌水量预测[J]. 煤田地质与勘探，2024，52(4)：1−9.

DING Yingying，YIN Shangxian，LIAN Huiqing，et al. Prediction of mine water inflow along mining faces using the SSA-CG-Attention multifactor model[J]. Coal Geology & Exploration，2024，52(4)：1−9.