基于Bi-LSTM-ARIMA的超大直径泥水平衡盾构油脂消耗量预测模型  

Prediction Model for Grease Consumption of Super-Large Diameter Slurry Shield Based on Bidirectional Long Short-Term Memory and Autoregressive Integrated Moving Average Model

作　　者：刘四进 王宇博 方勇[2] 熊英健 马浴阳 LIU Sijin;WANG Yubo;FANG Yong;XIONG Yingjian;MA Yuyang(China Railway 14th Bureau Group Co.,Ltd.,Jinan 250101,Shandong,China;Key Laboratory of Transportation Tunnel Engineering,Ministry of Education,Southwest Jiaotong University,Chengdu 610031,Sichuan,China)

机构地区：[1]中铁十四局集团有限公司,山东济南250101 [2]西南交通大学交通隧道工程教育部重点实验室,四川成都610031

出　　处：《隧道建设（中英文）》2023年第3期408-416,共9页Tunnel Construction

基　　金：中国铁建股份有限公司科研计划课题(2018-B06);中铁十四局集团科技研发计划课题(913700001630559891202215);第六届中国科协青托项目(2020-2022QNRC001)。

摘　　要：为了对盾构盾尾油脂消耗控制提供指导,以盾尾密封油脂消耗量预测精度为目标,采用双向LSTM与ARIMA模型相结合的方法,构建Bi-LSTM-ARIMA盾尾密封油脂消耗时间序列预测模型,在综合考虑盾构掘进参数与相关工程地质及水文地质参数的基础上,建立盾尾密封油脂消耗BP神经网络预测模型。以济南黄河隧道为依托,基于区间隧道既有盾尾油脂消耗数据对盾构东线区间盾尾密封油脂消耗量进行训练和预测。研究结果表明:Bi-LSTM-ARIMA模型对盾尾密封油脂消耗预测的均方根误差为13.47,平均相对误差仅为3.13%,相较于ARIMA时间序列模型和BP神经网络模型具有更高的预测精度,具有较好的实用性与可靠性。Grease is a major consumable material in the shield tunneling process,and the accurate prediction of grease consumption is significant in controlling the project cost.In this study,a time series prediction model of shield tail sealing grease consumption based on bidirectional long short-term memory and autoregressive integrated moving average model(Bi-LSTM-ARIMA)is constructed to manage and predict the shield tail sealing grease consumption accurately.Moreover,a back-propagation(BP)neural network prediction model of shield tail sealing grease consumption is established,considering shield tunneling parameters and related engineering,geological,and hydrogeological parameters.The prediction model was applied to the Yellow river tunnel in Jinan,China,and the shield tail sealing grease consumption in the eastern section of the tunnel was trained and predicted.The results show that the root mean square error of the applied Bi-LSTM-ARIMA model is 13.47 and the mean absolute percentage error is only 3.13%,indicating better practicability and reliability than the ARIMA time series and the BP neural network models.

关 键 词：超大直径泥水平衡盾构 油脂消耗 时间序列 神经网络 预测模型 

分 类 号：U45[建筑科学—桥梁与隧道工程]