基于TCN-LSTM的盾构刀盘扭矩实时预测研究  

Research on Real-time Prediction of Shield Cutterhead Torque Based on TCN-LSTM

作　　者：冯通 胡锦健 李研 张箭 梁禹 丰土根[2] FENG Tong;HU Jinjian;LI Yan;ZHANG Jian;LIANG Yu;FENG Tugen(The Third Engineering Co.,Ltd of China Railway Seventh Group,Xi′an 710000;Key Laboratory for Geotechnical Engineering of Ministry of Water Resource,Hohai University,Nanjing 210098;School of Civil Engineering,Sun Yat-sen University,Zhuhai 519082;Key Laboratory for Tunnel Engineering,Guangzhou 510275)

机构地区：[1]中铁七局集团第三工程有限公司,西安710000 [2]河海大学岩土力学与堤坝工程教育部重点实验室,南京210098 [3]中山大学土木工程学院,珠海519082 [4]隧道工程灾变防控与智能建养全国重点实验室,广州510275

出　　处：《现代隧道技术》2024年第5期120-128,166,共10页Modern Tunnelling Technology

基　　金：国家自然科学基金项目(52378336,52178386,52378427)。

摘　　要：盾构刀盘扭矩反映了刀盘与地层相互作用的力学特性,实时准确预测刀盘扭矩变化情况,可为掘进参数提前调整、机器平稳运行并减少刀具磨损提供保障。基于此,提出一种基于时间卷积网络(TCN)-长短时记忆网络(LSTM)的深度学习模型对刀盘扭矩进行实时预测研究。研究结果表明:TCN-LSTM模型能够捕捉输入参数的局部特征并建立长期依赖关系,相比于其他模型具有最高的预测精度;TCN-LSTM模型在多步预测中表现稳定,可以实现更长时间内的刀盘扭矩超前预测,按照4∶1∶1的比例划分数据集可以获得性能最优的预测模型。The shield cutterhead torque reflects the mechanical interaction characteristics between the cutterhead and the stratum.Accurately predicting torque changes in real-time can help adjust tunnelling parameters in advance,ensure smooth machine operation,and reduce cutting tool wear.Therefore,this paper proposes a deep learning model based on Temporal Convolutional Network(TCN)and Long Short-Term Memory(LSTM)for real-time prediction of cutterhead torque.The results indicate that the TCN-LSTM model can capture the local features of the input parameters and establish long-term dependencies,achieving the highest prediction accuracy compared to other models.The model performs stably in multi-step predictions,enabling longer lead-time predictions of cutterhead torque.A 4∶1∶1 data set split ratio yields the optimal performance for the prediction model.

关 键 词：刀盘扭矩 时间卷积网络 长短时记忆网络 多步预测 划分比例 

分 类 号：U455.43[建筑科学—桥梁与隧道工程]