Research On Short-term Passenger Flow Forecasting Of Urban Rail Transit Based On Multi-feature Fusion

Passenger flow forecasting plays an important role in the operation and management of urban rail transit systems.It is an important prerequisite for the rational deployment of transportation resources,the optimization of train operation plans,the guidance of daily transportation organization and the evaluation of economic benefits.With the rapid development of information collection technology,computing power,and artificial intelligence,the accuracy of short-term passenger flow derivation is especially critical under the conditions of network formation.Based on the data of the AFC system,this paper focuses on the spatial and temporal characteristics and prediction models of the inbound passenger flow of urban rail transit stations.The main work and results are as follows:(1)Temporal and spatial characteristics of inbound passenger flowAn analysis of the spatial-temporal characteristics of the inbound passenger flow of urban rail transit is made in study.First,the time characteristics of passenger flow are analyzed from all-day time sharing,daily passenger flow within the week,and periodicity;second,the spatial correlation of passenger flow is analyzed using Spearman correlation coefficient correlation coefficients.The results show that the passenger flow at different space stations has different wave patterns,mainly including single-peak type,double-peak type,full-peak type,peak-free type and sudden-peak type,and they have obvious periodicity,nonlinearity and spatial difference.Characteristics.(2)Construction of multi-feature fusion site passenger flow prediction modelFor the selection of predictors,first we use the Pearson and Spearman correlation coefficient to determine the time dependence and the spatial correlation of the passenger flow sequences of different stations to determine the input factors of the time and spatial dimensions;then we introduce MAE error to judge the impact of different types of weather on passenger flow,and further determine the reasonable weather category characteristics according to the size;finally we use One-hot coding technology to quantify the qualitative indicators.It presents an "end-to-end" short-term inbound passenger flow prediction framework with multi-features and LSTM neural network as the core is proposed.First,we make a preliminary prediction based on the time-dependent features of the LSTM neural network mining passenger flow sequences,use One-hot to encode weather and time period features,and embed external factor sparse matrix through the Embedding layer.Then we use the fully connected layer to fuse different sites and external factors to obtain the prediction results of passenger flow time series.(3)Case analysisTaking Chengdu South Railway Station as an example,we selected the ARIMA model and LSTM neural network as the reference model.After multiple experiments,the combined prediction model with multi-features has the best prediction performance,and the MAE errors of the training set and the verification set are 11.49 and 10.85.We perform generalization performance test on the model on the test set,the results show that the model has better prediction accuracy and robustness.