Prediction of turnout health status based on 1D-CNN and Transformer models
CHEN Junzhu
CHEN Guangwu
SHI Jianqiang
XING Dongfeng
Abstract：To address the high failure rate,low maintenance efficiency,and challenges in predicting the health status of railway turnouts,this study proposes a predictive method based on the integration of a One-dimensional Convolutional Neural Network(1D-CNN)and a Transformer model,using the S700K turnout machine as the research object.First,1D-CNN is employed to extract features from the raw data,generating 10 feature sets after training.Then,through feature evaluation,the five most representative feature sets for assessing turnout health are selected.These features,along with the health label values derived from the turnout power curve,are used to train the Transformer model,yielding the predicted health index.Finally,to evaluate the health status of the turnout system,Fisher's optimal segmentation algorithm is used to classify health stages,determining the optimal num-ber of health levels as three.Guidance is provided for maintenance work at different health stages.The research results indicate that the combined 1D-CNN and Transformer model exhibits superior predic-tive performance and generalization ability.Compared to commonly used models such as Gated Recur-rent Unit(GRU)and Long Short-Term Memory(LSTM),the Transformer model achieves better performance in processing long time-series data.The proposed hybrid model significantly improves the accuracy of turnout health status prediction,reducing Mean Absolute Error(MAE)and Root Mean Square Error(RMSE)by 31.2％and 30.5％,respectively,compared to the 1D-CNN and LSTM model combination.
Keywords：railway turnoutsone-dimensional convolutional neural network(1D-CNN)Trans-former modelhealth status predictionFisher's optimal segmentation
Publication Date：2025-06-30
Online Publishing Date：2025-08-21(First online date of this platform, not the publication date of the document)
Pages：11( 33-43 )