The Prediction And Analysis Of Short-term Passenger Flow Of Urban Rail Transit Based On The Improved Elman Neural Network And Spark

Rail transit has become the first choice for people's travel and the main ways to alleviate the traffic congestion for its characteristics of large passenger traffic,less pollution and high punctuality rate.Besides,the rail transit passenger flow forecast is quite important for the operation department to make the train operation schedules and train operation plans.It can greatly help the operation department in optimizing train departure interval,realizing train dynamic dispatch and improving passenger travel experience.Due to the large difference in passenger traffic at different times of a day,compared with the long-term passenger flow forecast,short-term passenger flow prediction is more difficult and the effects of nonlinearity and randomness are greater.Inspired by the aforementioned situations,the main contributions of this thesis are listed as follows:Firstly,analysis and comparisons are carried out for several common passenger flow prediction methods.And based on the analysis results,the neural network is selected as the main model of passenger flow prediction for its self-learning and self-adaptive ability.A DHOH_1stOH_2ndF?Double Hidden Output-First Hidden Output-Second Hidden Feedback?network structure is constructed based on the common network and the improved OIF-Elman?Output Input Feedback-Elman?and OHF-Elman?Output Hidden Feedback-Elman?models;Secondly,considering that the training time of network model grows rapidly with the increasing sample size,a method of training is proposed based on the Spark distributed parallel computing framework.Furthermore,a combination of the Spark and artifical netural network model is designed to reduce the time cost of network training;Thirdly,a hierarchical clustering algorithm is proposed based on the principle of cluster center distance minimization to divide each day of a week.When it comes to the selection of the passenger flow forecast factors,a spatial-temporal correlation analysis algorithm is designed by taking two different levels of time and space dimensions into the candidate range;Finally,sites and sections short-term passenger flow predictions are carried out to test and verify the stability and applicability of the proposed model via the actual passenger flow data of Guangzhou rail transit.The experimental results show that the prediction accuracy of the proposed model meets the requirement of 80%of the project's medium-term index.