Prediction of high-embankment settlement combining joint denoising technique and enhanced GWO-ν-SVR method

Abstract

Reliable long-term settlement prediction of a high embankment relates to mountain infrastructure safety. This study developed a novel hybrid model (NHM) that combines a joint denoising technique with an enhanced gray wolf optimizer (EGWO)-ν-support vector regression (ν-SVR) method. High-embankment field measurements were preprocessed using the joint denoising technique, which includes complete ensemble empirical mode decomposition, singular value decomposition, and wavelet packet transform. Furthermore, high-embankment settlements were predicted using the EGWO-ν-SVR method. In this method, the standard gray wolf optimizer (GWO) was improved to obtain the EGWO to better tune the ν-SVR model hyperparameters. The proposed NHM was then tested in two case studies. Finally, the influences of the data division ratio and kernel function on the EGWO-ν-SVR forecasting performance and prediction efficiency were investigated. The results indicate that the NHM suppresses noise and restores details in high-embankment field measurements. Simultaneously, the NHM outperforms other alternative prediction methods in prediction accuracy and robustness. This demonstrates that the proposed NHM is effective in predicting high-embankment settlements with noisy field measurements. Moreover, the appropriate data division ratio and kernel function for EGWO-ν-SVR are 7:3 and radial basis function, respectively.