摘要
针对铁路路基沉降探测波长预测精度不足、适应性差的问题,提出了基于深度学习的铁路路基沉降探测波长预测模型,旨在通过挖掘光纤传感器波长数据的时空特征,提升路基沉降风险的实时监测与预警能力。该模型通过位置编码模块捕捉时序关系,利用多头自注意力机制捕获全局空间依赖,结合时间卷积网络(TCN,Temporal Convolutional Network)的膨胀卷积提取多尺度时间模式,并引入双向长短期记忆(BLSTM,Bi-directional Long Short-Term Memory)网络增强对序列的记忆能力和上下文理解能力,实现对波长数据的时空特征提取与预测。基于潍烟(潍坊—烟台)高速铁路实测数据进行实验,实验结果表明,该模型在测试集上的预测误差较低,可准确识别异常数据,各项指标均优于通用模型,具有工程应用价值,为铁路路基沉降探测波长的高精度监测提供了技术支撑。
This paper proposed a deep learning based wavelength prediction model for railway subgrade settlement detection to address the issues of insufficient accuracy and poor adaptability.The aim was to improve the real-time monitoring and early warning capabilities of subgrade settlement risks by mining the spatiotemporal characteristics of wavelength data from fiber optic sensors.This model captured temporal relationships through a position encoding module,captured global spatial dependencies using a multi head self-attention mechanism,extracts multi-scale temporal patterns using dilated convolutions of a Temporal Convolutional Network(TCN),and introduced a Bi-directional Long Short Term Memory(BLSTM)network to enhance the capacity of sequence memory and contextual understanding,implement spatiotemporal feature extraction and prediction of wavelength data.Based on the actual measurement data of the Weifang-Yantai high-speed railway,the experimental results show that the model has low prediction error on the test set,can accurately identify abnormal data,and all indicators are better than the general model.It has engineering application value and provides technical support for high-precision monitoring of railway subgrade settlement detection wavelength.
作者
朱剑锋
尧靖
尹浩东
姚向明
ZHU Jianfeng;YAO Jing;YIN Haodong;YAO Xiangming(Beijing Dalu Instruments Technology Co.Ltd.,Beijing 100044,China;School of Systems Science,Beijing Jiaotong University,Beijing 100044,China;School of Traffic and Transportation,Beijing Jiaotong University,Beijing 100044,China)
出处
《铁路计算机应用》
2025年第6期8-16,共9页
Railway Computer Application
基金
国家自然科学基金(52372299)。
