As mining activities continue to intensify, underground roadways encounter unprecedented stability challenges due to increasingly complex stress environments, dynamic geological variations, and anthropogenic disturbances. Accurate prediction of roadway deformation is essential for ensuring mining safety and optimizing operational layouts. However, conventional single neural network models often struggle to effectively capture both abrupt local features and long-term evolutionary trends in nonlinear displacement time series. To address these limitations, this study introduces an innovative hybrid model combining a Temporal Convolutional Network and Long Short-Term Memory(TCN-LSTM-AddAttn) architecture, enhanced with an additive attention mechanism, to achieve high-precision predictions of roadway deformation. The proposed architecture employs a parallel framework to leverage the strengths of both TCN and LSTM. An additive attention mechanism is incorporated to dynamically prioritize critical patterns from both networks. The model employs learnable parameters to calculate feature similarity and utilizes the Softmax function to generate normalized weights, facilitating the adaptive fusion of multi-scale representations. Validation of the model is conducted using displacement data from four monitoring points(W1430-10, W1430-11, W1480-7, W1530-11) across various roadways in the Yunnan Zizou iron mine. Data preprocessing involves the removal of outliers using a Hampel filter, interpolation of missing values via cubic spline, and min-max normalization to standardize input scales. The processed data are divided into training, validation, and test sets in an 8∶1∶1 ratio. Hyperparameters, including TCN channels (32), LSTM hidden dimensions (64), batch size (32), and learning rate (0.001), are optimized through grid search to ensuring generalization across diverse mining scenarios. Experimental results indicate that the TCN-LSTM-AddAttn model outperforms standalone TCN, LSTM, and the TCN-LSTM hybrid models. In the case of W1430-10, the TCN-LSTM-AddAttn model demonstrates a Mean Absolute Error (MAE) of 0.0292 mm, representing a 47.95% reduction compared to the TCN-LSTM model. Additionally, it achieves a Root Mean Square Error (RMSE) of 0.0396 mm, marking a 37.34% reduction, and a Symmetric Mean Absolute Percentage Error (SMAPE) of 0.0337, indicating a 47.91% reduction. The Adjusted R² (Radj) value of 0.9861 suggests near-perfect prediction accuracy. For W1430-11, the model records an MAE of 0.0282 mm (28.79% reduction), an RMSE of 0.0366 mm (27.52% reduction), a SMAPE of 0.0738 (28.70% reduction), and an Radj of 0.9799. Comparable improvements are noted for W1480-7 and W1530-11, with prediction errors consistently remaining below 0.1 mm. By incorporating multi-scale feature decoupling and dynamic weighting, the proposed model offers robust technical support for assessing mine roadway stability, identifying risk zones, and providing early safety warnings.
近年来,随着计算机和人工智能技术的发展,众多学者使用机器学习模型对井巷位移进行预测。其中,一维卷积神经网络(1DCNN)、门控循环单元(GRU)、长短期记忆网络(LSTM)及其改进模型在井巷和隧道位移预测中得到广泛应用(赵楠等,2021;He et al,2023;崔靖奇等,2024;刘辉等,2025)。此外,结合模拟退火算法的自适应神经模糊网络(SA-ANFIS)、基于鲸鱼优化算法的BP神经网络(WOA-BP)和采用粒子群优化算法的长短期记忆网络(PSO-LSTM)等方法,通过对神经网络模型的参数优化,在提升井巷位移的预测精度中也展现出优势(Xie et al,2021;Du et al,2022;吴泽鑫等,2024)。上述研究证实了使用神经网络模型对井巷位移进行预测能够取得较好的预测效果。然而,现有研究采用单一神经网络模型的方法存在一定局限性。具体表现为BP神经网络易陷入局部极值,卷积神经网络在全局特征提取方面存在局限,循环神经网络易受梯度消失和爆炸问题困扰(郭风景等,2023;翟小伟等,2025)。此外,仅采用单一模型存在难以协同解析井巷位移中短期扰动响应与长期演化趋势的多时间尺度耦合特征的缺点。因此,构建具备多尺度特征解耦能力的预测模型,对于进一步提升井巷预测精度和增强模型泛化性能具有重要意义。
加性注意力是一种重要的注意力机制。在对时序数据进行分析时,引入加性注意力机制能够显著增强对时间序列内关键信息的捕捉能力,提高信息的利用率(Wu et al,2021;胡倩伟等,2024)。在加性注意力运行过程中,将输入数据送入线性变换层通过权重矩阵转换为全局查询向量,该过程能够高效地从输入中提取后续注意力计算需要的关键特征表示;再将查询向量和键向量 分别通过各自线性变换矩阵进行线性变换得到中间向量,随后通过可学习的参数向量,将中间特征映射为标量相似度评分,随后使用Softmax函数对评分进行归一化,生成注意力权重分布,最终依据权重对值向量 进行加权求和得到输出。
AkossouA Y J, PalmR,2013.Impact of data structure on the estimators R-square and adjusted R-square in linear regression[J].International Journal of Mathematics and Computation,20:84-93.
[2]
ChiccoD, WarrensM J, JurmanG,2021.The coefficient of determination R-squared is more informative than SMAPE,MAE,MAPE,MSE and RMSE in regression analysis evaluation[J].PeerJ Computer Science,7:e623.
[3]
DuB, YiH H, YangF,2022.Deformation prediction and analysis of soft rock roadway with high altitude and large buried depth based on particle swarm optimization LSTM model[J].Mobile Information Systems,(1):5907051.
[4]
HeY C, ChenQ N,2023.Construction and application of LSTM-based prediction model for tunnel surrounding rock deformation[J].Sustainability,15(8):6877.
[5]
RibeiroM H D M, da SilvaR G, RibeiroG T,et al,2023.Cooperative ensemble learning model improves electric short-term load forecasting[J].Chaos,Solitons and Fractals,166:112982.
[6]
SongH, YuanR, LüY,et al,2023.Cubic spline interpolation-based refined composite multiscale dispersion entropy and its application to bearing fault identification[J].Structural Health Monitoring,22(4):2651-2673.
[7]
WuC H, WuF Z, QiT,et al,2021.Fastformer:additive attention can be all you need[J].Arxiv Preprint: 2108.09084.
[8]
XieC Y, NguyenH, BuiX N,et al,2021.Predicting roof displacement of roadways in underground coal mines using adaptive neuro-fuzzy inference system optimized by various physics-based optimization algorithms[J].Journal of Rock Mechanics and Geotechnical Engineering,13(6):1452-1465.
CuiJingqi, WuShunchuan, ChengHaiyong,et al,2024.Time series prediction of the surrounding rock displacement of a soft rock tunnel in the Central Yunnan Water Diversion Project[J].Journal of Tsinghua University(Science and Technology),64(7):1215-1225.
GuoFengjing, WangBin, JiaPengtao,et al,2023.Online prediction of gas concentration based on gradient boosting decision tree[J].China Coal,49(10):61-67.
HuQianwei, WangXiuqing, AnYang,et al,2024.A study on the application of A2 former model in time series forecasting[J].Artificial Intelligence Science and Engineering,49(1):41-50.
KangHongpu, FanMingjian, GaoFuqiang,et al,2015.Deformation and support of rock roadway at depth more than 1 000 meters[J].Chinese Journal of Rock Mechanics and Engineering,34(11):2227-2241.
LiuHui, LiGuoqiang, ZhuXiaojun,et al,2025.Exploration and application of deep learning based wellbore deformation forecasting model[J].Journal of China Coal Society,50(2):732-747.
LuNannan, ChenYongheng, MaZhanguo,et al,2019.Research on informatization technology for roadway surrounding rock structure monitoring in deep underground[J].Industry and Mine Automation,45(6):1-9.
SunChaoyi, ChenCongxin, XiaKaizong,et al,2024.Mechanism of surrounding rock failure and strata movement in complex goaf of metal mines[J].Chinese Journal of Rock Mechanics and Engineering,43(3):670-682.
WangWenqiang, YanBo, ZhangJunru,2024.Prediction model for deformation of shallow buried tunnel portal collapse treatment based on Hampel-DeepAR model[J].Automation and Instrumentation,(4):40-46.
WangXufeng, ChenXuyang, WangJiyao,et al,2023.Creep failure mechanism and control of the deep soft rock roadway in Pingdingshan mining area[J].Journal of Mining and Safety Engineering,40(6):1139-1150.
WuZexin, ZhangChengliang, ZhangHuachao,et al,2024.Study on optimization of neural network based on whale algorithm for prediction of vault subsidence in slope[J].Nonferrous Metals Engineering,14(4):150-160,174.
XieHeping, GaoFeng, JuYang,2015.Research and development of rock mechanics in deep ground engineering[J].Chinese Journal of Rock Mechanics and Engineering,34(11):2161-2178.
ZhaiXiaowei, WangBo, JiangXueming,et al,2025.Prediction method and application of coal temperature in goaf based on DPO-BP neural network[J].Safety in Coal Mines,56(4):75-81.
ZhaoNan, LiJie,2021.Displacement prediction of tunnel surrounding rock based on LSTM-SVM[J].Highway,66(6):404-407.
[37]
周小燕,李双平,冉鲁光,等,2025.基于深度学习的大坝边坡深部变形时空预测模型研究[J].中国农村水利水电,(7):182-187,195. Zhou Xiaoyan,Li Shuangping,Ran Luguang,et al,2025.Research on spatiotemporal prediction model for deep deformation of dam slope based on deep learning[J].China Rural Water and Hydropower,(7):182-187,195.