摘要
为了实现工作面大规模矿压数据的智能分析与准确预测,基于深度学习理论,提出了一种结合时序卷积网络、门控循环单元网络与注意力机制的工作面矿压预测模型(TCN GRU DAB)。对工作面矿压历史数据中的异常值、缺失值进行处理并归一化;采用时序卷积网络(TCN)提取处理后数据的时序特征,利用TCN的扩张卷积快速并行地捕获数据局部和长时间的依赖关系;在模型中引入门控循环单元网络(GRU),利用GRU能够捕捉更复杂和深层次信息的能力,对TCN提取的时序特征进行进一步的建模,增强模型对矿压数据动态变化的理解和预测能力;引入可变形注意力机制(DAB)提取空间特征,并使模型在计算资源有限的情况下聚焦于最重要的特征,进一步提升模型的预测能力;选择Nlinear、DLinear、RNN、LSTM、GRU、BiGRU、BiLSTM、Informer等预测模型作为对比模型,在实际数据集上验证TCN GRU DAB模型。工程应用结果表明,相较于对比模型,TCN GRU DAB模型在矿压数据上的预测精度具有明显优势,能够显著提高工作面矿压预测的泛化能力和预测效率。
To achieve intelligent analysis and accurate prediction of large-scale mine pressure data at the working face,a prediction model(TCN GRU DAB)was proposed based on deep learning theory.This model combines the time series convolutional network(TCN),gated recurrent unit(GRU),and deformable attention mechanism(DAB).The historical mine pressure data was preprocessed by handling outliers and missing values,followed by normalization.Next,the TCN was employed to extract time-series features,leveraging its dilated convolution to efficiently capture both local and long-term dependencies in parallel;the GRU network was incorporated to model these extracted features,utilizing its capability to capture complex and deep temporal patterns,thereby enhancing the model’s ability to understand and predict dynamic changes in mine pressure;the deformable attention mechanism(DAB)was introduced to extract spatial features and enabled the model to focus on the most critical features under the condition of limited computational resources,further improving the prediction ability of the model;several models-such as NLinear,DLinear,RNN,LSTM,GRU,BiGRU,BiLSTM,and Informer-were selected for comparison to validate the effectiveness of the TCN GRU DAB model on real-world datasets.The results of engineering applications demonstrated that,compared to the baseline models,the TCN GRU DAB model could significantly improve prediction accuracy,generalization ability and inference efficiency.
作者
问永忠
贾澎涛
杨鸿宇
张龙刚
WEN Yongzhong;JIA Pengtao;YANG Hongyu;ZHANG Longgang(Shaanxi Shanmei Pubai Mining Co.,Ltd,Weinan,Shaanxi 715517,China;College of Computer Science and Technology,Xi’an University of Science and Technology,Xi'an,Shaanxi 710054,China)
出处
《中国煤炭》
北大核心
2025年第2期102-112,共11页
China Coal
关键词
矿山压力预测
预测模型
深度学习
智能分析
预测效率
mine pressure prediction
prediction model
deep learning
intelligent analysis
prediction efficiency
