Research And Implementation Of Railway Track Slab Prediction Model Based On Deep Learning

With the continuous improvement of China’s comprehensive national strength,the development of China’s high-speed railway is also booming,which brings great convenience to people’s lives.With the continuous progress and development of science and technology,ballast track has been widely used in the construction of high-speed railways nationwide,and has become an important mode of high-speed railway track operation in China.However,due to various environmental or human factors,after the ballast track is put into use,there will be rail slab cracks,subgrade subsidence,geometric deformation of the rail slab and other diseases,which seriously threaten the safety of the train.In order to effectively avoid the risks caused by rail diseases to the smoothness and stability of the railway,it is necessary to analyze the causes of the diseases and locate the disease points in combination with advanced non-destructive exploration methods.However,the existing detection methods are mainly to improve the efficiency and accuracy of disease exploration,and it is difficult to ensure the timely and effective detection and treatment of possible diseases.Therefore,it is necessary to establish an algorithm that can effectively predict railway diseases and timely warn of ballast tracks that may cause disease problems,so as to deal with railway diseases more timely and effectively,ensure the safety of railway operations,and promote the long-term development of high-speed railway ballast tracks.This paper proposes the Raynet network model to detect railway diseases.The model first improved the traditional recurrent network model RNN,and proposed the BICT-SRNN model to accelerate RNN calculation through sequence segmentation calculation and information breakpoint fusion,etc.,to alleviate the problems of long calculation time and difficult training when inputting long sequences.Afterwards,an attention mechanism is added to enable the model to focus on the key information in the input data.Adding position coding alleviates the position insensitivity of the attention mechanism and improves the predictive ability of the model.The prediction task in this paper belongs to the regression task,which is to predict the high and low value of the railway.The evaluation criteria used the RMSE(root mean square error)commonly used in regression tasks.In order to prove that the final algorithm and the basic RNN model and the classic machine learning method are more accurate than the prediction,this article did on the same test set.In contrast to the experiment,the prediction algorithm proposed in this paper is reduced from 0.492 to 0.313 compared to the Root Mean Square Error of LSTM,and reduced to 0.313 from 0.522 by the Random Forest.The final prediction ability of the algorithm is improved by about two percentage points.