Passenger Flow Data Prediction Of Rail Transit Stations Based On Deep Learning

As of the end of 2022,China’s rail transit passenger volume reached 19.4 billion people.At the same time,with various large-scale activities held in the city,the subway passenger flow shows a short-term surge,bringing huge safety hazards to residents’ travel.Therefore,how to efficiently predict traffic passenger flow is crucial for urban traffic safety.This paper first studies the back propagation(BP)prediction model of deep learning,and constructs a new weight update formula to alleviate its Vanishing gradient problem.Afterwards,based on the Long Short Term Memory(LSTM)model,inheritance gates were added and output gates were deleted to design an inherit-LSTM model suitable for short-term passenger flow prediction.On this basis,a RI-LSTM model is constructed based on the distribution characteristics of spatial stations to achieve high-precision short-term multi-station passenger flow prediction.The specific content is as follows:(a)The in-depth learning BP neural network model is used to predict passenger flow growth and alleviate the problem of Vanishing gradient problem.By estimating the gradient decline amplitude,a new weight update formula is constructed with the number of hidden layers as the variable.The experimental results show that compared with the traditional update formula,the deep BP neural network model can effectively alleviate the Vanishing gradient problem and improve the model training speed.(b)Design the inherit LSTM model for short-term rail transit passenger flow prediction.The inherit-LSTM model adds an inheritance gate on the basis of LSTM,filtering redundant historical information one step ahead of the forgetting gate,while deleting the output gate to avoid secondary filtering of input information,reducing the complexity of the model while preserving the current input information to the greatest extent possible,making the inherit-LSTM model suitable for short-term passenger flow prediction when passenger flow surges.Comparative experiments were conducted between the attention LSTM,read first LSTM,and the inherit LSTM model,and the results showed that the inherit LSTM model predicted more accurately.(c)Construct an RI-LSTM model to predict rail transit passenger flow based on spatiotemporal characteristics.Firstly,multiple sites with strong spatial correlation were selected and their distribution characteristics were analyzed to construct the Rela-LSTM model.Secondly,an RI-LSTM model was constructed by combining the inherit LSTM model,which is good at short-term prediction,to solve the problem of short-term multi-site traffic passenger flow prediction.Finally,the RI LSTM model was compared with Rela LSTM,inherit LSTM,attention LSTM,and read first LSTM models.The results showed that the RI LSTM model had the highest prediction accuracy,with an improvement of about 9%compared to inherit LSTM,about 11%compared to Rela LSTM,about 24.1%compared to attention LSTM,and about 24.5%compared to read first LSTM.