A Research On Disturbance Control Methods In A Dispatching Area Based On Process Mining Approach

On entering the 21 st century,the informatization construction of Chinese railway has greatly promoted the modernization of transportation capabilities and services.All kinds of facilities in the railway network are interconnected and interoperable,forming the basic conditions of data acquisition and aggregation.On the background of increasing demand on railway operation,research on the application of data mining method in train dispatching system would provide new thoughts for promoting operation quality and improving train operation methods.The paper designs disturbance control methods in a dispatching area,which establish disturbance cases through delay prediction and disturbance identification methods,forming a timetable rescheduling problem.Rescheduling problem is solved with relevant algorithm later and the goal of disturbance control is then attained.Delay prediction and disturbance identification methods are based on the process mining model of train dispatching data.Through researches on delay prediction,disturbance identification and timetable rescheduling,disturbance control methods provide new thoughts for the intellectualization of train dispatching in a practical way.Details of this paper include three aspects:delay prediction,disturbance identification and timetable rescheduling.Delay prediction model is based on delay cause analysis method,which is the causal analysis method of delay propagation.Delay propagation result can be expressed in the form of delay propagation chain.All train events in delay propagation chain can be categorized as primary delay and consecutive delay.By the means of process mining algorithm,adjacent relationships between train events are established.By introducing disturbance propagation parameters,the paper constructs a cause analysis network,where source of disturbance is obtained and delay propagation chain is found.Based on the relationship between delay propagation and delay,the paper designs a delay prediction model.Based on the practical situation of China's train dispatching system,prediction model is formulated as a classification model where delays are categorized in different levels.The prediction model is constructed by BP neural network and verified by plan and real-time data from Beijing railway administration.Resulting accuracy is 85.5% within a permissible error of 5 minutes.The main research object of disturbance identification method is tight tracking.Tight tracking is caused by small tracking interval between two trains.Traditional method for section disturbance identification can not identify tight tracking disturbance.In this paper,a process mining algorithm is established to collect logical train number records in train dispatching system in order to obtain occupied time length of train block sections in high accuracy,assisting an improved disturbance identification method.Furthermore,the paper adopts blocking time model to describe occupancy time of train block sections,and discovers statistical rule of the occupancy time through machine learning algorithm and distribution fitting test.On the basis of statistical rules,the occupancy time of different types of trains in the blocked section is estimated and reconstructed into blocking time model.Finally,by using disturbing data from real-time dispatching records,train line is reconstructed and expressed in blocking time model,and section disturbance caused by tight tracking of two trains is successfully determined.The improved method has accomplished tight tracking identification which traditional method can not achieved,and provides some thoughts for solving disturbance identification problems in train dispatching system.Integer programming is the traditional algorithm for timetable rescheduling.Due to large-scale combinations of decision variables in the solving process,the algorithm results in low calculation efficiency.The paper reconstructs timetable rescheduling problem as a muti-variable combination constraint satisfaction problem without changing the constraints about timetable intervals,station capacity and dispatching methods adopted by the traditional algorithm.Reconstructed problem is solved by constrained programming algorithm,which triggers condition checks and rescheduling rules by constraint propagation and trial-and-error mechanism.Number of variables is also controlled in the process to avoid low calculation efficiency.To verify the computational efficiency and solution quality of the constraint programming algorithm,the paper also designs an integer programming model with the same constraints and a PERT simulation method which uses no rescheduling rules.The algorithm is verified by plan data records from Beijing-Tianjin intercity dispatching area.Total delay and computation efficiency of 3 algorithms are checked and compared.Results of the experiment has verified the computational efficiency and solution quality of the constraint programming algorithm.Through the application of process mining models and constraint programming method,disturbance control research in the dispatching area has achieved good performance in the aspects of delay prediction,disturbance identification and timetable rescheduling.The method has realized the whole control process of disturbance control and reached high calculation efficiency,the feasibility of process mining approach and relevant algorithms are also verified.