基于盾构机运行参数的局部切空间排列与Xgboost融合的地质类型识别

Geological-type identification with LTSA and Xgboost algorithm based on EPB shield operating data

针对土压平衡盾构机掘进过程难以实时感知掌子面地质类型的问题,提出了局部切空间排列(LTSA)与极限梯度提升(Xgboost)相结合的盾构机掌子面地质类型实时识别方法。首先,通过分析众多盾构机运行参数与掌子面地质性质的相关性,选取177个盾构机运行参数作为模型输入;其次,利用LTSA算法从高维盾构机运行参数中提取内蕴低维特征作为分类模型输入参数,基于Xgboost的识别模型实现掌子面地质类型识别;最后,采用新加坡某地铁施工数据验证算法的有效性和优越性。研究结果表明:所提算法对该工程沿线5种地质类型的识别准确率达到98.48%;采用本文方法所得的识别准确率相比于将运行参数直接作为模型输入的识别准确率提升20.96%,相比于采用总推进力、推进速度、刀盘总扭矩和刀盘转速4维特征作为输入,本文所提出方法的识别准确率提升50.16%。LTSA算法能够减少所选盾构运行参数中的冗余信息并保留其中的地质特征,解决了输入参数维度过高造成的识别模型准确率下降和训练效率降低的问题。

Aiming at the difficulty in real-time perception of the geological type of palm surface in earth pressure balance shield tunneling process,a real-time recognition algorithm for the geological type of palm surface was proposed,which combined the local tangent space arrangement(LTSA)and the extreme gradient boosting(Xgboost).Firstly,based on the correlation between the operating parameters of shield machine and the rock and soil properties of tunnel face,177 operating parameters of the shield machine were selected as the model input.Secondly,LTSA algorithm was used to extract the intrinsic low-dimensional features from the high-dimensional operating parameters of shield machine as input parameters of the identification model,and an identification model based on Xgboost was established to realize the identification of face geological types.Finally,the validity and superiority of the method were verified by a subway construction data in Singapore.The results show that the proposed method achieves 98.48%accuracy in identifying the five geological types of this engineering section.The identification accuracy of the proposed method increases by 20.96%compared with that of taking all the operating parameters as the input of the Xgboost model.The identification accuracy of the proposed method increases by 50.16%compared with that of using the total propulsion force,propulsion speed,total torque of cutterhead and rotational speed of the cutterhead as the input of the Xgboost model.The LTSA algorithm can reduce the redundant information in the selected shield operating parameters and retain the geological features,which solves the problems of the lower recognition accuracy and the lower training efficiency caused by the high dimension of the input parameters.