基于卷积神经网络的列车位置指纹定位算法研究

Study of Train Location Fingerprint Positioning Algorithm Based on Convolutional Neural Network

针对高速铁路隧道环境下采用位置指纹定位时定位精度低的问题,提出将深度卷积神经网络应用于列车位置指纹的定位中。首先采用2σ准则、模糊C均值聚类FCM及类数据加权,对采集到的下一代铁路通信系统LTE-R中的信号强度值进行预处理,降低异常值的影响,提高指纹数据的有效性;然后引入定时提前量,增强指纹特征值;接着将处理后的指纹数据量转换为灰度图片指纹条,基于图像样本建立FCM-CNN指纹定位模型;最后以现场实测数据为基础对定位模型进行测试验证。结果表明,相较于采用未经处理的数据作为样本的CNN模型及传统的位置指纹定位方法,基于FCM-CNN的列车位置指纹定位方法,提高了数据质量,在离线阶段具有较大的指纹采集间距,大幅减少了指纹采集工作量,模型训练时间较短,定位精度小于10 m的概率可达100%,满足列车在中密度线路对定位精度的要求。

In response to the problem of low positioning accuracy when using location fingerprint under high-speed railway tunnel environment,deep convolutional neural network was applied to the positioning of train location fingerprint.Firstly,2σcriteria,FCM and class data weighting were used to preprocess the signal strength value in the next generation railway communication system LTE-R collected to reduce the influence of abnormal values and to improve the effectiveness of fingerprint data.Secondly,the timing advance was used to enhance the characteristic values of the fingerprints.Then the processed fingerprint data were converted into grayscale picture fingerprint strips,and an FCM-CNN fingerprint positioning model based on image samples was established.The test results based on the field data show that the FCM-CNN train position fingerprint model improves data quality,reduces the workload of fingerprint collection due to its larger fingerprint collection spacing in the offline stage,and shortens the training hours in comparison with the CNN model based on unprocessed data and some other conventional location fingerprint positioning methods.The probability of less than 10 m positioning accuracy can reach 100%,meeting the requirements for train positioning accuracy on medium density railways.