基于深度学习的超大直径盾构姿态预测研究OA北大核心CSTPCD
Research on attitude prediction of super large diameter shield based on deep learning
传统的盾构姿态纠偏措施多是在盾构机实际轴线已经偏离设计轴线之后采取的被动控制措施,具有一定的滞后性,而盾构姿态纠偏不及时会给施工过程和完成后的隧道本身带来严重的危害.为了准确预测盾构姿态偏差,为提前纠偏提供决策支持,本文依托江阴—靖江长江隧道超大直径盾构施工项目,提出一种基于CNN-EMD-LSTM的深度学习模型,该模型既能捕捉时间序列的维度特征和时变特征,又能提高盾构姿态数据分解重构方法的预测精度;通过消融实验对CNN-EMD-LSTM模型中每个部分的重要性进行探讨,并对CNN-EMD-LSTM模型在不同窗口长度、不同滑动步长下的预测效果进行对比.研究结果表明:CNN-EMD-LSTM模型对超大直径盾构姿态的预测效果较好;可以通过调节不同推进区间的压力进行盾构姿态纠偏;CNN-EMD-LSTM模型中各个部分按重要性从大到小排序依次为EMD、CNN、LSTM;窗口长度过大或过小都会增大模型预测误差,而滑动步长越小模型的预测效果越好.
The traditional attitude correction measure of the shield machine is mostly a passive control measure taken after the actual axis of the shield machine has deviated from the design axis,which has a certain lag.The lagged attitude correction of the shield will bring serious harm to the construction process and the tunnel itself after completion.To accurately predict the attitude deviation of the shield and provide decision support for correction in advance,a CNN-EMD-LSTM based deep learning model was proposed in this paper based on the Jiangyin—Jingjiang Yangtze River Tunnel super large diameter shield construction project,which can not only capture the dimensional and time-varying features of the time series,but also improve the prediction accuracy by decomposing and reconstructing the attitude data of the shield.The importance of each part of the CNN-EMD-LSTM model was discussed through ablation experiment.Furthermore,the prediction effect of CNN-EMD-LSTM model under different window lengths and different sliding steps was compared.The results show that the CNN-EMD-LSTM model has a good effect on predicting the attitude of the super large diameter shield.The attitude deviation of the shield can be corrected by adjusting the pressure in different advancing regions.The importance of each part of the CNN-EMD-LSTM model is EMD,CNN and LSTM in descending order.Too large or too small window length will increase the prediction error of the model,and the smaller the sliding step length,the better the prediction effect of the model.
丰土根;胡锦健;张箭
河海大学岩土力学与堤坝工程教育部重点实验室,江苏南京,210098
交通运输
超大直径盾构姿态预测姿态纠偏消融实验CNN-EMD-LSTM
super large diameter shieldattitude predictionattitude correctionablation experimentCNN-EMD-LSTM
《中南大学学报(自然科学版)》 2024 (004)
1477-1491 / 15
国家自然科学基金资助项目(52178386,52378336);中央高校基本科研业务费专项资金资助项目(B220202016)(Projects(52178386,52378336)supported by the National Natural Science Foundation of China;Project(B220202016)supported by the Fundamental Research Funds for the Central Universities)
评论