|
|
|
| Weighted voting model for advanced intelligent perception of tunnel faults based on TBM rock-machine information
|
| LI Long1，LIU Zaobao1，ZHOU Hongyuan1，QI Wenbiao2，ZHA Wenhua3 |
| (1. Key Laboratory of Ministry of Education on Safe Mining of Deep Metal Mines，College of Resources and Civil Engineering，Northeastern University，Shenyang，Liaoning 110819，China；2. Jilin Province Water Resource and Hydropower Consultative Company，Changchun，Jilin 130021，China；3. School of Civil and Architectural Engineering，East China University of Technology，Nanchang，Jiangxi 330013，China) |
|
|
|
|
Abstract In order to realize the advance perception of unfavorable geology of TBM tunnel，this paper is devoted to tunnel fault intelligent perception using machine learning methods based on 4.08 billion records of TBM tunnel construction data of Jilin Yinsong project. The variation laws of TBM rock-machine interaction parameters (cutter-head torque and top shield pressure，etc.) are analyzed，and the excavation segments near the fault were divided into stable section，early warning section and fault section. Seven key parameters were selected as the features for fault perception by Pearson correlation analysis. A weighted integrated voting model was constructed for the intelligent perception of tunnel faults using the random forest and support vector machine method as the base classifiers. The effective tunneling cycle data of 771 groups of TBM construction near the fault was selected，and the model was trained (539 groups) and tested (232 groups) with accuracy as the target. The accuracy，recall rate and F1-score were used to evaluate the performance of the weighted integrated voting model for fault advance perception. Partial dependence plot was used to determine the sensitivity of key interaction parameters in different prediction categories. The results show that the key interaction parameters such as cutter-head torque and top shield pressure exhibit different degrees of responses prone to the fault. The weighted integrated voting model can effectively predict the stable，early warning and fault section with an overall accuracy of 89.22%. The work provides supports for early warning analysis and pre-control measures near the TBM tunnel fault.
|
|
|
|
|
|
|
|
|
2020, Vol. 39 
