Similarity search and performance prediction of shield tunnels in operation through time series data mining

Hehua Zhu, Xin Wang, Xueqin Chen, Lianyang Zhang

Research output: Contribution to journalArticlepeer-review

21 Scopus citations


Performance analysis and prediction of shield tunnels in operation have become increasingly important during the maintenance strategy planning process. An existing assessment indicator, Tunnel Serviceability Index (TSI), was employed to evaluate the condition of shield tunnels in soft soils in this study. Data mining methods including Long Short-Term Memory (LSTM) and clustering analysis based on Dynamic Time Wrapping (DTW) were utilized to identify the different degradation patterns and predict the performance of shield tunnels. The data mining methods were applied to 4 tunnel intervals of Shanghai Metro Line 1. Four degradation patterns were determined objectively through the clustering approach, each of which showed similar trends and characteristics. Based on the distribution of clusters, interval 2 was determined to have the worst overall condition. The LSTM network was used for performance prediction in each cluster. Compared with the traditional multilayer neural network, the LSTM also exhibited good prediction effectiveness. The predicted data are well consistent with the observed data with correlation coefficient R2 equaling to 88.4%. Finally, a case study using the data from Shanghai Metro Line 2 is conducted for further validation of the above data mining methods.

Original languageEnglish (US)
Article number103178
JournalAutomation in Construction
StatePublished - Jun 2020


  • Clustering analysis
  • Deep learning network
  • Dynamic time wrapping
  • Long Short-Term Memory
  • Tunnel Serviceability Index (TSI)

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Civil and Structural Engineering
  • Building and Construction


Dive into the research topics of 'Similarity search and performance prediction of shield tunnels in operation through time series data mining'. Together they form a unique fingerprint.

Cite this