基于振动响应与卷积长短期记忆模型的路基沉降智能识别
Intelligent Identification of Subgrade Settlement Based on Vibration Responses and a Convolutional Long Short-Term Memory Model

路基不均匀沉降是诱发无砟轨道结构变形、影响高速列车运行安全与平稳性的核心病害。针对传统路基沉降监测方法效率偏低、成本较高、难以实现全天候实时精准评估的技术短板，本文提出一种基于CNN-LSTM混合神经网络的无砟轨道路基沉降智能识别方法。以CRTS II型双块式无砟轨道为研究载体，搭建车辆–轨道–路基垂向耦合动力学仿真模型，系统分析不同沉降病害工况下高速列车车辆系统的动力振动响应规律。通过多工况动力响应数据对比分析，筛选出车体垂向加速度、车体点头角速度、钢轨垂向位移等对路基沉降高度敏感的核心特征参数。融合CNN特征提取优势与LSTM时序数据拟合能力，构建路基沉降状态智能识别模型，完成不同沉降工况的精准识别与分类，并通过仿真试验验证模型的有效性与稳定性。试验结果表明，所构建的CNN-LSTM识别模型整体准确率可达99.26%，能够精准、高效地识别无砟轨道路基沉降病害状态，可为高速铁路路基沉降智能化监测、病害预警及运维整治提供全新的技术支撑。
Uneven subgrade settlement is a critical form of deterioration that induces structural deformation of ballastless tracks and affects the operational safety and ride comfort of high-speed trains. To address the limitations of conventional subgrade settlement monitoring methods, such as low efficiency, high cost, and the difficulty of achieving real-time, accurate, and all-weather assessment, this study proposes an intelligent identification method for ballastless track subgrade settlement based on a CNN-LSTM hybrid neural network. Taking the CRTS II double-block ballastless track as the research object, a vertical coupled vehicle-track-subgrade dynamic model is established to systematically investigate the dynamic vibration response characteristics of the high-speed train vehicle system under different subgrade settlement conditions. Through comparative analysis of dynamic response data under multiple operating conditions, key feature parameters that are highly sensitive to subgrade settlement, including car-body vertical acceleration, car-body pitching angular velocity, and rail vertical displacement, are identified. By integrating the feature extraction capability of CNNs with the time-series data fitting capability of LSTMs, an intelligent subgrade settlement state identification model is developed to accurately identify and classify different settlement conditions. The effectiveness and stability of the proposed model are further validated through numerical simulation experiments. The results demonstrate that the proposed CNN-LSTM identification model achieves an overall accuracy of 99.26%, enabling accurate and efficient identification of ballastless track subgrade settlement conditions. The proposed method can provide new technical support for intelligent monitoring, early warning, and maintenance of subgrade settlement in high-speed railways.