Research On Short-term Traffic Flow Prediction Model Based On Real-time Data

Urban rail transit is considered the best solution for alleviating traffic congestion due to its high passenger capacity,speed,punctuality,safety,and environmental friendliness.Short-term passenger flow forecasting in rail transit provides a basis for timely adjustment of operational strategies by transportation departments and serves as a key indicator for measuring the service level of rail transit.This paper reviews and summarizes domestic and international shortterm passenger flow forecasting methods and models,and analyzes the characteristics,advantages,and disadvantages of each forecasting model.Short-term passenger flow forecasting is introduced into statistical forecasting models,machine learning forecasting methods,neural network forecasting methods,and deep learning forecasting models to improve the accuracy and precision of short-term passenger flow forecasting through model combinations.Using the 2019 Suzhou subway AFC entry and exit card swiping data as an example,the feasibility and practicality of the four types of forecasting models in short-term passenger flow forecasting are verified.By comparing and analyzing various forecasting models and combining their characteristics,advantages,and disadvantages,it is concluded that deep learning models,relying on powerful network structures and parameter quantities,outperform traditional machine learning models in short-term rail transit passenger flow forecasting.The Transformer model achieves the best prediction accuracy in short-term rail transit passenger flow forecasting,while the linear regression model obtains the best prediction speed.The random forest model performs well in both accuracy and speed,making it suitable for application scenarios that require both accuracy and speed and can be used as the first-choice solution.