基于CNN-GRU网络的无砟轨道路基沉降智能识别

Intelligent Identification of Subgrade Settlement in Ballastless Track Based on CNN-GRU Network

摘要: 随着高速铁路的快速发展，对轨道稳定性和列车运行安全性的要求日益提高，而路基沉降问题已成为影响其性能的关键因素之一。目前，轨道路基沉降监测技术普遍存在效率低下、成本高昂的问题，且难以实现实时、精准的沉降状态评估。针对这一挑战，本研究选取CRTS II型双块式无砟轨道路基沉降病害作为研究对象，建立了车辆–轨道–路基垂向耦合动力学模型，深入探讨了车辆系统在沉降条件下的振动特性及其变化规律。通过相关性分析，筛选出对沉降响应敏感的关键指标，并结合卷积神经网络(CNN)与门控循环单元(GRU)算法，提出了一种高效的无砟轨道路基沉降病害识别方法。实验结果表明，所构建的CNN-GRU模型在路基沉降识别中表现出色，识别准确率达到95.56%。本研究验证了CNN-GRU算法在无砟轨道路基沉降识别中的有效性，为高速铁路路基沉降监测提供了一种新的技术手段，具有重要的理论和实际应用价值。

Abstract: With the rapid development of high-speed railways, increasing demands have been placed on track stability and train operation safety, among which subgrade settlement has emerged as a critical factor affecting system performance. However, existing subgrade settlement monitoring technologies generally suffer from low efficiency, high costs, and limited capability for real-time and accurate assessment. To address these challenges, this study investigates the subgrade settlement defects of CRTS II double-block ballastless tracks. A vertical vehicle-track-subgrade coupled dynamics model is established to comprehensively analyze the vibration characteristics of the vehicle system under settlement conditions and their evolution patterns. Based on correlation analysis, key indicators that are highly sensitive to settlement responses are identified. By integrating a Convolutional Neural Network (CNN) with a Gated Recurrent Unit (GRU), an efficient method for identifying subgrade settlement defects in ballastless tracks is proposed. Experimental results show that the proposed CNN-GRU model achieves excellent performance in identifying subgrade settlement, with an accuracy of 95.56%. This study demonstrates the effectiveness of the CNN-GRU algorithm in ballastless track subgrade settlement identification and provides a novel technical approach for subgrade condition monitoring in high-speed railways, offering significant theoretical and practical value.