Artificial-Intelligence-Based Model for Early Strong Wind Warnings for High-Speed Railway System DOI Open Access
Wei Gu, Hongyan Xing, Guoyuan Yang

et al.

Electronics, Journal Year: 2024, Volume and Issue: 13(23), P. 4582 - 4582

Published: Nov. 21, 2024

Wind speed prediction (WSP) provides future wind information and is crucial for ensuring the safety of high-speed railway systems (HSRs). However, accurate (WS) remains a challenge due to nonstationary nonlinearity patterns. To address this issue, novel artificial-intelligence-based WSP model (EE-VMD-TCGRU) proposed in paper. EE-VMD-TCGRU combines energy-entropy-guided variational mode decomposition (EE-VMD) with customized hybrid network, TCGRU, that incorporates loss function: Gaussian kernel mean square error (GMSE). Initially, raw WS sequence decomposed into various frequency-band components using EE-VMD. TCGRU then applied each component capture both long-term trends short-term fluctuations. Furthermore, function, GMSE, introduced training analyze WS’s nonlinear patterns improve accuracy. Experiments conducted on real-world data from Beijing–Baotou demonstrate outperforms benchmark models, achieving absolute (MAE) 0.4986, (MSE) 0.4962, root (RMSE) 0.7044, coefficient determination (R2) 94.58%. These results prove efficacy train operation under strong environments.

Language: Английский

RTINet: A Lightweight and High-Performance Railway Turnout Identification Network Based on Semantic Segmentation DOI Creative Commons
Dehua Wei, Wenjun Zhang,

Haijun Li

et al.

Entropy, Journal Year: 2024, Volume and Issue: 26(10), P. 878 - 878

Published: Oct. 19, 2024

To lighten the workload of train drivers and enhance railway transportation safety, a novel intelligent method for turnout identification is investigated based on semantic segmentation. More specifically, scene perception (RTSP) dataset constructed annotated manually in this paper, wherein innovative concept side rails introduced as part labeling process. After that, work Deeplabv3+, combined with lightweight design an attention mechanism, network (RTINet) proposed. Firstly, consideration need rapid response deployment model high-speed trains, paper selects MobileNetV2 network, renowned its suitability deployment, backbone RTINet model. Secondly, to reduce computational load while ensuring accuracy, depth-separable convolutions are employed replace standard within architecture. Thirdly, bottleneck module (BAM) integrated into position feature information perception, bolster robustness quality segmentation masks generated, ensure that outcomes characterized by precision reliability. Finally, address issue foreground background imbalance recognition, Dice loss function incorporated training procedure. Both quantitative qualitative experimental results demonstrate proposed feasible identification, it outperformed compared baseline models. In particular, was able achieve remarkable mIoU 85.94%, coupled inference speed 78 fps customized dataset. Furthermore, effectiveness each optimized component verified additional ablation study.

Language: Английский

Citations

0
Artificial-Intelligence-Based Model for Early Strong Wind Warnings for High-Speed Railway System DOI Open Access
Wei Gu, Hongyan Xing, Guoyuan Yang

et al.

Electronics, Journal Year: 2024, Volume and Issue: 13(23), P. 4582 - 4582

Published: Nov. 21, 2024

Wind speed prediction (WSP) provides future wind information and is crucial for ensuring the safety of high-speed railway systems (HSRs). However, accurate (WS) remains a challenge due to nonstationary nonlinearity patterns. To address this issue, novel artificial-intelligence-based WSP model (EE-VMD-TCGRU) proposed in paper. EE-VMD-TCGRU combines energy-entropy-guided variational mode decomposition (EE-VMD) with customized hybrid network, TCGRU, that incorporates loss function: Gaussian kernel mean square error (GMSE). Initially, raw WS sequence decomposed into various frequency-band components using EE-VMD. TCGRU then applied each component capture both long-term trends short-term fluctuations. Furthermore, function, GMSE, introduced training analyze WS’s nonlinear patterns improve accuracy. Experiments conducted on real-world data from Beijing–Baotou demonstrate outperforms benchmark models, achieving absolute (MAE) 0.4986, (MSE) 0.4962, root (RMSE) 0.7044, coefficient determination (R2) 94.58%. These results prove efficacy train operation under strong environments.

Language: Английский

Citations

0