Prediction of jacking force using PSO-BPNN and PSO-SVR algorithm in curved pipe roof

Abstract

Jacking force prediction is essential for pipe jacking construction, especially when facing randomness, uncertainty, and nonlinear conditions, which is a challenge for most analysis methods. To address this issue, machine learning (ML), a data analysis method that has been successfully applied in other construction fields such as TBM thrust force forecast, was introduced in this study. Specifically, the PSO-BPNN and PSO-SVR models were developed for jacking force prediction, involving raw data preprocessing, sensitivity analysis, hyper-parameter selection, and prediction accuracy evaluation. To validate and compare the two ML-based models, a curved pipe roof project was employed, which included three geometric, five geological, and eight operational parameters, totaling 4398 data points. The results show that both models exhibit good performance in predicting jacking force, and the PSO-SVR model outperforms the PSO-BPNN model; on the other hand, both models demonstrate better prediction accuracy in the frontal resistance, grouting pressure, and rotational speed of the cutterhead. Additionally, the generalization ability of the jacking force prediction models was examined using a new dataset, and the PSO-SVR model exhibited superior performance over the PSO-BPNN model. The study highlights the unique features of the two models in their application, and the advantages of the ML models provide valuable insights for jacking force prediction in pipe jacking construction. Thus, this study serves as a reference for future research in this area.