摘要
在泥水平衡盾构掘进过程中,掘进速度对刀具磨损、同步注浆及盾构机姿态有着重大影响,合理的掘进速度对于提高施工效率、降低施工风险具有重要意义。利用PLC系统采集到的历史数据构建智能预测模型,对掘进速度进行实时预测,是未来实现盾构机无人驾驶和智能控制的重要基础。基于珠江三角洲水资源配置工程中采集到的掘进参数时序数据,选取掘进速度作为预测目标,采用皮尔逊相关分析方法提取重要特征参数,利用均值滤波法对特征参数时序数据进行去噪,计算去噪后序列的均值和方差构成特征向量,基于XGBoost算法构建相应的预测模型。讨论了采用均值滤波法去噪前后的数据集及XGBoost算法中不同超参数对模型预测性能的影响。结果表明,在5-折交叉验证下,利用均值滤波进行降噪处理后的数据能够构建一个更为准确的盾构机掘进速度预测模型。采用XGBoost算法,在去噪后的数据集上盾构机掘进速度的预测准确率达到了99.97%,在未去噪的数据集上的预测准确率也达到了99.94%,优于主流随机森林算法、支持向量机回归算法和梯度提升决策树算法。试验结果验证了均值滤波法对时序数据的降噪效果和利用XGBoost算法对掘进速度进行预测的可行性。
In the process of slurry balance shield tunneling, the tunneling speed has a significant impact on cutter wear, synchronous grouting and shield attitude. A reasonable tunneling speed is of great significance for improving construction efficiency and reducing construction risks. The historical data collected by the PLC system is used to build an intelligent prediction model to predict the tunneling speed in real time, which is an important basis for the realization of unmanned shield and intelligent control in the future. Based on the time series data of tunneling parameters collected in the Pearl River Delta Water Resources Allocation Project, the tunneling speed is selected as the prediction target, the Pearson correlation analysis method is used to extract important characteristic parameters,the mean filtering method is used to denoise the time series data of characteristic parameters, to calculate the mean and the variance of the denoised series to form the feature vector, and to build the corresponding prediction model based on the XGBoost algorithm. The influence of data sets before and after denoising by mean filtering and different hyper-parameters in the XGBoost algorithm on the prediction performance of the model is discussed. The results show that under the 5-fold cross validation, a more accurate prediction model of the tunneling speed of the shield machine can be established by using the data after denoising processing with mean filtering. With the XGBoost algorithm, the prediction accuracy of the tunneling speed of the shield machine based on the denoised data set has reached 99.97%, and the prediction accuracy based on the non-denoised data set has also reached 99.94%, which is superior to the mainstream random forest algorithm, the SVR algorithm and the GBDT algorithm. The experimental results prove the noise reduction effect of the mean filtering method on the time series data and the feasibility of using XGBoost algorithm to predict the tunneling speed.
作者
杜庆峰
张双俐
张晨曦
李旭辉
肖永生
李晓军
赵思成
付艳斌
DU Qingfeng;ZHANG Shuangli;ZHANG Chenxi;LI Xuhui;XIAO Yongsheng;LI Xiaojun;ZHAO Sicheng;FU Yanbin(Software College,Tongji University,Shanghai 201804;Guangdong GDH Pearl River Delta Water Supply Co.,Ltd,Guangzhou 511455;College of Civil Engineering,Tongji University,Shanghai 200092;School of Civil and Traffic Engineering,Shenzhen University,Shenzhen 518060)
出处
《现代隧道技术》
CSCD
北大核心
2022年第6期14-23,共10页
Modern Tunnelling Technology
基金
国家自然科学基金(52078304,51678363)
珠江三角洲水资源配置工程项目(CD88-GC02-2020-0038).
