Short-term Passenger Flow Prediction Of Urban Rail Transit Based On Spatio-temporal Features

As an important part of intelligent transportation system,urban rail transit(URT)is favored by the public because of its convenient,comfortable and green travel characteristics.Passenger flow prediction can provide basis for intelligent planning and scheduling of operation departments and guide passengers to choose travel modes and routes.Short-term passenger flow prediction has greater randomness and volatility than medium and long term prediction,and the accuracy of prediction is directly related to the efficiency and rationality of urban rail transit organization operation.Most of the existing studies are based on the station ticket clearance system,and the passenger flow data after data pretreatment belongs to the time series,which has periodicity,tendency and random fluctuation in time.Meanwhile,urban rail transit has the characteristics of wire network,passenger flow direction is fixed as up or down,and the nature of land use around the station also affects the change of passenger flow,so there is a certain spatial connection.Therefore,this thesis mainly considers the influencing factors of passenger flow from two aspects of time and space,classifies the stations on the two lines,constructs the short-term passenger flow grouping prediction model based on the associated spatio-temporal characteristics of multiple stations,and establishes the common station category prediction model and special station category prediction model respectively considering different categories.The main work includes:Firstly,the influencing factors of passenger flow change are analyzed from the perspective of time and space,and the similar distribution of passenger flow is clustered and classified to extract features,and the stations are grouped considering the spatial relevance of the whole line.Since few studies have considered the connection and similarity of passenger flow between multiple stations on the whole subway line,hierarchical clustering algorithm and correlation analysis method are used to analyze the influencing factors of time and space on passenger flow,and the inner connection of the spatio-temporal characteristics of passenger flow is explored.From the angle of time,similar passenger flow distribution within the day and week was clustered and different category features were extracted;daily passenger flow with high correlation coefficient was extracted from week to week as prediction features;stations were grouped according to passenger flow feature clustering in space.It is proved that time and space are the important factors affecting passenger flow.Secondly,the nonlinear combined model is used to predict the short-term passenger flow of urban rail transit.Based on the structure of the long short term memory(LSTM)neural network,the stacked long short term memory(SLSTM)neural network was introduced to increase the network depth,and the optimized stacked long and short duration memory neural network model was constructed by optimizing the network structure with genetic algorithm.The validity of the model was verified by comparing several models on ordinary and special transfer sites.At the same time,for the special large passenger flow stations that are independently classified,the Empirical Mode Decomposition(EMD)is used to decompose the passenger flow sequence with large random fluctuation into time series with higher stationarity and periodicity,and then different prediction models are used to predict.The combined prediction model is constructed by comparing the characteristics of traditional linear model and nonlinear model.Finally,taking Hangzhou Metro Line 1 and Line 9 as an example,the short-term passenger flow grouping prediction model associated with temporal and spatial characteristics is constructed.Experiment compared the single site prediction methods,only basic features extraction and increase the time features of two kinds of combination forecast,and the method in this thesis,the prediction results of each site were analyzed,and the experimental results show that the combination of the spatio-temporal features associated multi-site group can improve the short-term passenger flow forecasting accuracy,especially for the peak passenger flow prediction accuracy to improve the effect is remarkable.It is further proved that temporal and spatial characteristics are very important for short-term passenger flow prediction.