Research On High Speed Railway Track Geometric Anomaly Detection Method Based On Multi-source Data

The vigorous development of high-speed railway has not only greatly improved the state of transportation in China,but also promotes the national economic development and produces great social benefits.In the process of ensuring the normal operation of high-speed railways,the regular inspections of railway track inspection vehicles provide important support for ensuring the safety,stability and comfort of trains and guiding track maintenance.At the same time,numerous track dynamic inspection data have been accumulated,and the research on its deep mining is still in its infancy.One of the key challenges is how to detect the railway track geometry anomalies caused by equipment failure or potential track diseases.In response to this challenge,from the perspective of the intersection of rail transit and deep learning,this paper conducts research on the detection method of high-speed railway track geometry abnormal state based on track dynamic inspection data.The specific research contents are as follows:(1)Aiming at the problems of noise,unlabeled data and unbalanced sample scale in track dynamic inspection data,a track geometry anomaly detection algorithm based on attention autoencoder is proposed.First,the track geometry anomaly detection model is constructed based on sparse denoising autoencoder.The model is divided into two parts:an encoder and a decoder.The encoder maps the data to the feature space,and the decoder restores the code to the original data.In this process,by adding mileage attention module and attribute attention module in the encoder,the long-term trend of data and the correlation information between attributes are extracted.After that,the model is trained and the data to be detected is fed into the trained model for data reconstruction.Finally,the reconstruction error of the data is used as the anomaly score to detect the abnormal states.The experimental results show that the proposed algorithm is superior to other anomaly detection algorithms in the task of track geometry abnormal state detection.(2)Aiming at the problem of small-scale track geometry abnormal state data and difficulty in extracting sequence features,a semi-supervised track geometry abnormal state classification algorithm is proposed.First,the mileage attention module and the attribute attention module are used to extract the relevant features of the track geometry abnormal state.After that,a multi-layer fully connected network is used to compress and merge the extracted features.Finally,input the compressed features into the classifier to classify the track geometry abnormal state.In addition,this paper introduces the consistency regularization loss of the unlabeled data,and trains the model in a semi-supervised learning manner to prevent the model from overfitting.The results of comparative experiments show that,compared with the current existing time series classification models,the accuracy of the model proposed in this paper is improved by more than 7%on the task of track geometry abnormal state classification.