Metro Passenger Flow Forecasting Systems Based On Deep Neural Networks

In recent years,with the rapid economic development and the rapid growth of urban population,the traffic congestion problem in large cities has become more and more serious.The urban rail transit system has effectively alleviated this problem.However,with the expansion of metro lines and the complexity of operation methods,the safety and efficiency of metro operations are facing more severe challenges.Accurate forecasting of passenger flow can help optimize headway and make a reasonable operation plan to alleviate traffic congestion and improve passenger comfort effectively.Big data and deep learning technology have played an essential role in many fields.Research shows that deep neural networks have excellent performance in regression prediction problems under sufficient data conditions.With the development of urban rail transit informatization,the data collection methods are gradually improved.It is of great practical significance to study the passenger flow big data processing of the urban rail transit and use the deep neural network to accurately predict the urban rail transit passenger flow.Firstly,the thesis preprocesses the passenger flow data of Beijing Metro Automatic Fair Collection System(AFC).Then the thesis builds a long-short-term memory(LSTM)neural network and a convolutional neural network(CNN)model,and the model is used to predict the passenger flow of the metro accurately.Finally,combined with the Teradata big data environment,a system which predicts the real-time passenger flow of metro is realized based on B/S architecture.The main research work of the thesis is as follows:(1)Based on the original AFC passenger flow data,the thesis makes a statistical analysis of the metro passenger flow data from the time dimension,and the thesis studies the distribution difference and periodic variation of the workday and holiday passenger flow.According to the difference in passenger flow distribution in different stations,the thesis makes statistical analysis of passenger flow from two dimensions of time and space.(2)To complete the passenger flow prediction in the time dimension,firstly,the thesis performs data preprocessing on the original AFC passenger flow data,and converts the data into time series one-dimensional data.Then an LSTM deep neural network model is built,and the optimal hyperparameter combination of the LSTM model is found based on the grid search algorithm.Finally,the LSTM model is used to predict the passenger flow data of the Beijing Metro.The simulation results show that compared with the traditional machine learning model,the prediction effect of LSTM model is greatly improved.(3)To complete the space and time two-dimensional passenger flow prediction,firstly,the thesis converts the original AFC passenger flow data into two-dimensional data with time as the abscissa and station as the ordinate.Then the CNN deep neural network model is established and the optimal hyperparameter combination of the CNN model is found in the thesis.Finally,the CNN model is used to predict the passenger flow data of the Beijing Metro.Compared with the traditional machine learning model,the prediction results are also greatly improved.Different from the LSTM model,the CNN model fully considers the spatial characteristics between stations.(4)To apply the deep learning model of the thesis to the real system,the thesis realizes the visual prediction software based on B/S architecture.The software can obtain the passenger flow data from the database in real-time,and predict the passenger flow by using the trained deep neural network.The advantage of the B/S architecture software is installation-free,and it can use a Web browser to display the predicted results and trends of the passenger flow.There are 69 pictures,13 tables,and 88 references.