Study On The Train Delay Prediction Model By Using The Real Operation Data Of The Dutch Railway Network

The train delay prediction is a process of estimating the delay probability of a train at a subsequent recorded point based on train running history data,usually measured by arrival(departure)delay.On the one hand,train delay will seriously affect the railway transportation organization and reduce the quality of railway transportation services;on the other hand,it will increase travel time and seriously affect passenger travel plans.Therefore,understanding the delay propagation mechanism of trains and conducting delay prediction research on trains is of great significance to dispatchers in daily work scheduling,and provides a theoretical basis for improving the punctuality rate of railway passenger trainsIn the daily work of railway dispatchers,it is necessary to accurately analyze and judge the influence of delay.Firstly,the temporal and spatial distribution law of train delay should be understood,and on this basis,the train delay should be predicted,which can be used as the basis for the formulation of dispatching adjustment strategy and the optimization of transportation organization.Therefore,this paper first selects 12 stations with large traffic volume and busy traffic from more than 400 stations of Dutch railway to analyze the temporal and spatial law of train delay.Specific work includes:(1)Analyzing the law of train arrival delay.The day is divided into 10 time periods,and the arrival delay rate,the average arrival delay time,the departure delay rate,and the average departure delay time of each station are counted,and the time and space law of train delay is analyzed.Then,the train arrival delay time and departure delay time are distributed and fitted in periods.And finally,the K-S test is used to obtain the optimal fit distribution model as a log normal distribution model.(2)Analyzing the train delay time increase and recovery laws.Similarly,the day is divided into 10 time periods,and the increase of the delay rate,the average increase delay time,the recovery rate,and the average recovery time of the station trains in each time period are respectively counted,and the regularity of the train increase delay time and recovery delay time is analyzed.Then the distribution fitting of train delay increase time and recovery time is carried out.Finally,the most suitable fitting distribution model of increase delay time in most periods is the log-normal distribution model,and the fitting model of recovery time is the gamma distribution modelBased on the analysis of train delay law,the delay prediction of Dutch railway train is carried out.Specific work includes:(1)At the macro level,prediction are made for the Dutch railway network.We extracted the main factors affecting the train delay,and predicted the delay situation of the entire railway network in the Netherlands after 20 minutes.The alternative prediction models are random forest model,artificial neural network model and XGBoost model.Through the adjustment of parameters,the random forest model has the best prediction effect and the best prediction performance.(2)At the micro level,trains are predicted delay for the busy section of the Dutch railway.By extracting the feature vector,we mainly use the GRU model to predict the train delay.The comparison models are LSTM model,random forest model,artificial neural network model and XGBoost model.After comparison,the GRU model has higher prediction accuracy and prediction performance.