Ruipeng Liu , Jieyan Zhang, Pengfei Chen, Yunxun Liu, Wanlin Quan, Junliang Su
{"title":"VDTformer: A transformer-based framework for accurate deformation risk detection in cable tunnels","authors":"Ruipeng Liu , Jieyan Zhang, Pengfei Chen, Yunxun Liu, Wanlin Quan, Junliang Su","doi":"10.1016/j.array.2025.100422","DOIUrl":null,"url":null,"abstract":"<div><div>Cable tunnels are critical components of urban power systems, ensuring reliable transmission of high-voltage electricity. However, their structural integrity is threatened by deformation risks, such as crown settlement and wall displacement, due to geological and environmental factors. Current monitoring methods using distributed optical fiber sensors face significant challenges because the acquired vibration signals are highly noisy and non-stationary, which hampers accurate risk detection. In this work, we propose VDTformer, a Transformer-based framework that integrates a Filter Bank Convolution (FBC) module for denoising and feature extraction with a Wavelet Transform-based Attention (WTA) mechanism for capturing non-stationary characteristics. Extensive experiments on real-world data demonstrate that our approach significantly improves detection accuracy and robustness over conventional methods, achieving an accuracy of 95.8% and a Macro-F1 score of 95.5%.</div></div>","PeriodicalId":8417,"journal":{"name":"Array","volume":"27 ","pages":"Article 100422"},"PeriodicalIF":4.5000,"publicationDate":"2025-06-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Array","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S2590005625000499","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"COMPUTER SCIENCE, THEORY & METHODS","Score":null,"Total":0}
引用次数: 0
Abstract
Cable tunnels are critical components of urban power systems, ensuring reliable transmission of high-voltage electricity. However, their structural integrity is threatened by deformation risks, such as crown settlement and wall displacement, due to geological and environmental factors. Current monitoring methods using distributed optical fiber sensors face significant challenges because the acquired vibration signals are highly noisy and non-stationary, which hampers accurate risk detection. In this work, we propose VDTformer, a Transformer-based framework that integrates a Filter Bank Convolution (FBC) module for denoising and feature extraction with a Wavelet Transform-based Attention (WTA) mechanism for capturing non-stationary characteristics. Extensive experiments on real-world data demonstrate that our approach significantly improves detection accuracy and robustness over conventional methods, achieving an accuracy of 95.8% and a Macro-F1 score of 95.5%.