Train Timetable Rescheduling Optimization Method Considering Delay Prediction And Complex Disrupted Scenarios Decoupling For High-speed Railways

The complex train operating environment and failure of equipment and infrastructure result in train delay,temporary speed limit,or track blockage.To guarantee safe and high-quality train services,it is one of the effective ways to reschedule train timetables in real-time when disruptions occur.When disruptions occur,trains are disordered and train delay propagates.Therefore,how to predict train delay based on the delay which has occurred,and generate high-quality rescheduled timetable intelligently is one of the hot topics.On one hand,how to develop a train delay prediction method with high accuracy based on historical operation data is the difficulty of the train delay prediction.On the other hand,how to formualte the stochastic model based on predicted delay,and construct the novel model in case of disruptions is the challenge of train timetable rescheduling problem.Based on this analysis of problem,the main research work is summarized as follows.(1)Based on the big operation data,the spatial and temporal characteristics of key parameters of the high-speed train movement are analyzed.Based on the historical perturbation response log data,the cause of the primary delay is excavated by means of the mathematical-statistical analysis.Based on the historical operation data,the distribution characteristic with positive skewness and long-tail is analyzed.The departure delay distribution at each station is formulated by means of the non-linear regression fitting method.The running time and dwell time and the correlation between different temporal-spatial train delays are used as the input of other parts.(2)A novel model-data-driven method for train delay propagation analysis and prediction is developed.The method includes a train event delay explanatory variable generation model based on a delay propagation network and train event delay prediction model based on a deep neural network considering dropout strategy.Considering the influence of online monitoring information and dispatching strategies on the structure of the train delay propagation network,the explanatory variables of train arrival and departure delay are generated in the procedure of train delay forward and knock-on propagation.Moreover,the train event delay prediction models based on the deep neural network are developed.The hyperparameters of the deep neural network are obtained by the genetic algorithm considering the trade-off between training time and testing accuracy.A real case study shows the proposed method has a small prediction error and robust performance on different data sets.This method effectively uses the actual historical operation data,overcomes the difficulty of describing the interdependence between train movement,online monitoring information,and dispatching strategies on the evolution of the temporal-spatial train delays,and enhances the reasonableness of the proposed models.(3)The stochastic model and algorithm for the train timetable rescheduling problem considering predicted train delay are developed.Considering the uncertainty of the predicted delay,the total train delay expectation for all predicted delay scenarios is defined as the train delay risk.The train timetable rescheduling stochastic model minimizing the train delay risk is formulated considering multiple dispatching strategies such as retiming,reordering,and changing stop patterns.In order to obtain rescheduled timetable online,the rule of feasible train order is constructed.Based on the rule,a two-stage iterative algorithm under the framework of the branch and cut algorithm is developed.The results of a real case study demonstrate that the high-quality rescheduled timetable can be obtained by the proposed model and algorithm in a short computational time.(4)A novel model for the train timetable rescheduling problem in case of complex multi-disruptions is developed.The event activity network which describes train movement and interdependences is formulated.Based on the event activity network,the constraints in case of serious primary delay,temporary speed limit,and track blockage are proposed.Auxiliary decision variables and the big M method is used to decouple the non-linear and strong-coupling constraints.Moreover,the constraints of station capacity and rolling stock connection are also considered.And then,the train timetable rescheduling model minimizing the deviation between the rescheduled and the planned timetable is developed.Combining the rolling horizon algorithm and branch and cut algorithm,an algorithm for solving large-scale problems is designed.Results of a real case study demonstrate that the proposed model and algorithm can effectively solve the train timetable rescheduling problem in case of complex multi-disruptions,and obtain a high-quality rescheduled timetable.(5)A prototype decision support system for delay prediction and train dispatching is developed.Based on the requirements of the high-speed railway management department for the statistical analysis and mining of train operation data and the intelligent decision support system for train timetable rescheduling under disrupted scenarios,the main functional modules such as analysis of the train operation data,train delay prediction,and online dispatching strategy generation are designed.The system architecture including the UI layer,the model layer,and the data layer is proposed.The main functional interfaces of the decision support system are shown.This thesis includes 95 figures,39 tables,and 151 references.