基于贝叶斯优化RF-BiLSTM的盾构机掘进速度预测的研究  

Research for the speed prediction of tunnel boring machine based on Bayesian optimized RF-BiLSTM

作　　者：王振东 WANG Zhendong(China Railway 16th Bureau Group Rail Company,Beijing,101100)

机构地区：[1]中铁十六局集团轨道公司,北京101100

出　　处：《中国建材科技》2023年第6期142-146,共5页China Building Materials Science & Technology

摘　　要：盾构机掘进速度的预测是保障盾构施工的重要参考指标。为了实现盾构机掘进速度的预测,本文提出了基于贝叶斯优化RF-BiLSTM的盾构机掘进速度预测方法,即TPE-RF-BiLSTM。首先通过随机森林实现盾构机运行数据的筛选,接着利用BiLSTM实现对盾构机掘进速度的预测。此外,为了提高超参数的搜索效率,贝叶斯优化被用于掘进速度预测模型的超参数搜索,以自动化的构建掘进速度的预测模型。最后,通过郑州某地铁施工段的真实数据验证所提方法,实验结果表明,所提的方法能够有效实现掘进速度的预测。即R2=0.9650,RMSE=1.684,表现优于XGBoost,LSTM等广泛应用的成熟机器学习算法。The speed of tunnel boring machine(TBM)is an important reference indicator to ensure the construction of shield.In order to predict the speed of tunnel boring machine,this paper proposes a method based on Bayesian optimized RF-BiLSTM,i.e.TPE-RF-BiLSTM.Firstly,random forest was used to filter the tunnel boring machine operation data.Then,BiLSTM was used to achieve the prediction of tunnel boring machine(TBM)tunneling speed.Besides,in order to improve the efficiency of hyperparameter searching,Bayesian optimization was used to realize hyperparameter searching for tunneling speed prediction model so that automated prediction model could be constructed.Finally,the proposed method was verified by real data of a subway construction section in Zhenzhou,and the experimental results show that TPE-RF-BiLSTM could achieve the advance rate prediction.,i.e.,R2=0.9650,RMSE=1.684,which outperforms the widely used mature machine learning algorithms such as XGBoost,LSTM,etc.

关 键 词：贝叶斯优化 随机森林 双向长短记忆网络 盾构机 掘进速度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]