Research On Theory And Method For Multi-state Short-term Prediction Of Passenger Flow In Urban Rail Transit Using Multi-source Data

With the networking of urban rail transit,the scale and complexity of urban rail transit network increase continuously and its scale effect is becoming more and more obvious,which brings about more challenges for networking operation and management in urban rail transit.For example,the intensified contradiction between transport capacity and traffic volume in some operation scenarios such as regular days,holidays,and predictable large passenger flow and the sudden large passenger flow at the station pose challenges for managerial stuffs on how to improve transport efficiency,operation and management level and promote service quality.The multi-state short-term prediction for passenger flow is the basis for ensuring the safety of daily operations and it is of great significance for scientific and efficient management as well.In order to solve the abovementioned problems,this paper introduces multi-source data as the basis to conduct research on the theory and method for multi-state short-term prediction of passenger flow.The specific research work is as follows:(1)Analyses on basic theories and related problems of multi-state short-term prediction for passenger flow.First,this article analyzes the spatial and temporal distributional characteristics of passenger flow under the networking operation of urban rail transit.Secondly,the predictive error is relatively large based on single AFC data since there exists delay in the uploading process of AFC data.This article conducts the quality analysis on multi-source data such as AFC data and mobile phone signaling data in a southwestern city of China to provide data support for passenger flow prediction.Finally,some problems in passenger flow prediction and the main content that this article needs to solve are elaborated.(2)The construction of multi-state short-term prediction model and algorithm for passenger flow based on computational graph architecture.First,the symbol system and multi-source data-driven computational graph architecture of this paper are illustrated.Secondly,a multi-level traffic flow network with the minimum sum of prediction errors at various stages as the goal is used to construct a nonlinear error comprehensive optimization model.In addition,this paper achieves short-term prediction of multi-state passenger flow indicators such as OD distribution,inbound and outbound quantity and sectional passenger flow at different periods of the day by adopting the computational graph architecture to fuse the multi-source data and using the forward and back propagation error layered optimization algorithm.The specific method is to integrate multi-source data by using the multi-level traffic flow network constituted by the passenger flow generation layer,OD distribution layer,route distribution layer and indicators output layer.At the same time,the FR conjugate gradient method is used to perform hierarchical iterative optimization on the error of predicted variables in the back propagation process.Finally,this paper constructs a small-scale case and uses the MATLAB R2019 b to verify the performance of error optimization of different methods in the back propagation process.(3)Design and implementation of multi-state short-term prediction system for passenger flow.First,this paper designs the function and architecture of the system.Secondly,this paper develops the multi-state short-term prediction system using multisource data based on the model and algorithm constructed in Chapter 3,Visual Studio 2017 software combined with C# programming language,WPF framework and Postgre SQL database.Finally,the functions of system including multi-state short-term prediction for passenger flow and statistics of multi-dimensional passenger flow indicators are stated.(4)Case analysis.The article takes the urban rail transit of a certain city as an example,for two short-term prediction scenarios using single data source and multisource data separately,the passenger flow indicators of different dimensions including network,line,section and station are respectively predicted and obtained based on the system.Second,the accuracy of predicted indicators is analyzed and error of predicted indicators of network layer using multi-source data is generally within 5%.The analysis and verification for the predicted indicators indicate the research content of this paper is of great significance for assisting operation,management and decision-making and improving the service level of passenger transportation.