Research On Forecasting Of Track Quality Index Based On Real-time Track Detection

Track quality has profound impact on the safety of railway transportation,the comfort of passengers and the maintenance cost of the department of railway maintenance.In recent years,the passenger and freight transport in China has shown a trend of high speed and heavy load after six times railway speed increasing,resulting in the worsening of the track quality,which also affects the railway safety.To slow down the further deterioration of the track quality has become an important basis for the normal railway transportation.Traditional track maintenance methods are mainly divided into "fault repair" and "periodic repair",which have been unable to meet the demand of railway transportation.The forecasting of track quality refers to the training of the collected track quality data through mathematical methods to get its development trend and to forecast the future development,thus providing some guidance for the department of railway maintenance when arranging track maintenance schedule.In this thesis,the research achievements made by many scholars at home and abroad in the field of track quality forecasting are analyzed,and the simulation experiments on the forecasting of track quality have been carried out with MATLAB.Track quality index is taken as the main research object.Grey model and neural network model are also systematically analyzed.Due to the abnormal value of the original track data collected by track inspection cars,the abnormal removal of the data and the mileage correction must be carried out first.Then it is found that track quality data present a general exponential growth rule after the mathematical analysis.The grey model can be used for preliminary modeling to obtain the general trend of the development of TQI.On this basis,the thesis focuses on the application of intelligent heuristic searching optimization algorithm and neural network forecasting model on data forecasting.And the residual error of preliminary forecasting results is corrected so as to get more accurate forecasting results.It not only analyzes the development trend of TQI data,but also takes its randomness into account.Through the above analysis,this thesis first aims at the instability of the traditional BP neural network in processing small sample data.In the fourth chapter,the thesis uses the Mind Evolutionary Algorithm(MEA)to optimize the BP neural network.Because MEA can efficiently calculate the optimal solution,the forecasting result obtained by BP network optimized by MEA is better than that of the traditional BP network.However,because the training algorithm of BP network is one of the local searching algorithms,it is easy to converge to the local minimum and lead to the failure of training,and the convergence rate of the BP network becomes slower.When the number of iterations increases,the efficiency of the network training is greatly reduced.This thesis proposes a forecasting method that combines unequal-interval Grey Model and Elman Neural Network(GM-GA-Elman)to overcome the inherent defects of BP network.The Grey Model GM(1,1)is previously exploited to obtain an approximate forecast of original TQI series and its residual error.Then genetic algorithm is used to find the optimal initial weight and threshold of the Elman neural network.Finally,the optimized Elman network is used to correct the TQI residual error sequence,and a more accurate TQI forecasting sequence is finally obtained.The proposed method is demonstrated with practically four section measured data of Shanghai-Kunming Railway Line.The forecasting results show that,comparing to other forecasting methods,the method which uses Elman Network to correct residual error correction reaches higher forecasting accuracy in many statistical index.