Short-term Prediction Of Passenger Flow Of Urban Rail Transit Under External Transportation Hub Based On LGB-LSTM-DRS Model

As an important measure to relieve urban congestion,urban rail transit has gradually become an irreplaceable public transportation method in recent years.At the same time,the city's external passenger transportation hub,as the main place to realize the transfer connection between external traffic and urban internal traffic,has continuously attracted a large number of people to gather in a shorter time and smaller space,resulting in the connected urban track Traffic stations frequently have large passenger flows that affect passenger safety and station operation.The accurate short-term prediction of the passenger flow of the urban rail transit in and out of the external hub can provide the relevant management departments with the necessary information to guide decision-making,which is of great significance to the safe operation and fine management of the external passenger hub and urban rail transit.This paper studies the short-term prediction of rail transit passenger flow in external transportation hubs.The main work is as follows:(1)Introduce the relevant concepts of external passenger transport hub and the influencing factors of rail transit passenger flow.First this paper defines the concept,characteristics and functions of the external transportation hub.then explains the definition and description indicators of urban rail transit passenger flow under the external transportation hub.Finally,it qualitatively analyzes several types of influencing factors of passenger flow in and out of the external hub.(2)Given the complete feature engineering process for solving the short-term passenger flow forecasting problem under the machine learning framework,firstly,the conventional features are constructed based on the existing research and experience;then,using feature derivation,the hub connection track for frequent fluctuations is innovatively proposed The method of constructing statistical characteristics based on traffic flow.Through the Isomap algorithm of manifold learning,feature visualization and dimensionality reduction are performed,and the feasibility of the established conventional and statistical features is visually verified.(3)The idea of model establishment is clarified for the two difficulties in fitting passenger flow situation and passenger peak values.1)Accurate fitting of the passenger flow situation requires the model to have good real-time performance.To this end,a long-short-term memory network LSTM model that can process long-term dependent time-series data is established.Classify the features according to time series,improve the network structure of the LSTM long-short-term memory network;realize multi-source feature input;use the RAdam optimizer to optimize the weight matrix,and simultaneously use the cosine annealing algorithm to set the learning rate to iterate,which improves the model iteration process Stability.2)The accurate fitting of the passenger peak values requires the model to have good generalization ability.For this reason,the light-weight implementation algorithm Light GBM of the Gradient Boosting Decision Tree is introduced,and the short-term prediction model is built synchronously.By using the Histogram and the Leaf-wise growth strategy,Light GBM handles a large number of passenger peak values,reduces the computational cost and improves the model accuracy.3)Based on the K-nearest neighbor algorithm,the DRS is selected by using the Dynamic Regression Selection that can find the local optimal fusion method,and the local optimal fusion prediction model of LGB-LSTM-DRS is established.(4)Take the AFC passenger flow data of Chengdu East Station as an example to verify the feature engineering and model construction method in this paper.In terms of feature validity,the method of controlling variables is used to verify the validity of the statistical features in this paper;The comparison analysis of the fitting effect and the residual error is carried out,and compared with other commonly used passenger flow prediction models such as LR,RF,GBDT,etc.,which proves that the proposed LGB-LSTM-DRS has the best prediction performance.