Research On Short-term Metro Passenger Flow Forecasting Method Based On Machine Learning

With the rapid growth of the urban population,the pressure of urban rail transit in China is also increasing,and with the vigorous development of the city’s economy,people’s need for transportation is getting higher and higher.The metro is the main way for passengers to travel,and the passenger volume of subway lines and stations is constantly rising and showing large differences.A large amount of subway card swiping data has subsequently been generated,and these data play a vital role in constructing efficient,safe,and convenient transportation.Algorithms in the field of machine learning can predict the distribution characteristics of passenger flow from these historical data,and quickly and accurately predict the short-term subway passenger flow.We not only need to understand the distribution of subway passenger flow in lines and stations,but also need to further accurately predict the passenger flow of each station of the line.In order to solve the above problems,based on the research of many scholars,this paper further studies the algorithm model in the short-term subway entrance and exit passenger flow prediction.Focusing on the problem of short-term passenger flow prediction in Hangzhou Metro,based on machine learning related technologies,research and design a mechanism for feature selection and fusion to construct effective features.On this basis,an algorithm model for short-term passenger flow prediction is proposed to achieve accurate short-term passenger flow prediction.And perform application analysis on the prediction results The core tasks include two aspects: one is the construction of feature engineering,and the other is the construction of the prediction model.The main research results are as follows:(1)A short-term subway passenger flow prediction model based on Random forest and LSTM is studied and implemented.From most features in the data set,select features that have a greater impact on the prediction results to further build the model.In this hybrid model,the random forest calculates the importance of features and ranks them to fully capture important features.Then,build a long and short-term memory network model that is superior in time series problems.LSTM has a good memory function for the time correlation of historical data.Through continuous testing and optimization,the model is finally established to accurately predict the passenger flow of short-term subway in and out stations.The experiment uses real historical data of the Hangzhou subway for training.The experimental results show that the important features of random forest screening are better than the artificial feature extraction in the prediction results.The prediction result of the proposed combined prediction model fits the actual passenger flow more than the single model,and the accuracy is greatly improved.(2)A short-term subway passenger flow prediction model based on CNN and Light GBM is studied and implemented.In this combined model,CNN uses the feature mapping process and downsampling process to iterate several times toautomatically extract the effective features in the time series data and prepare the features for the next model prediction.The original feature and the features that convolutional neural network automatically extract are merged into a new multi-dimensional feature set.The new feature set is used as the input of Light GBM for supervised training.Through continuous parameter optimization,CNN-Light GBM short-term passenger flow prediction model is finally obtained.Comparative experiments show that the combined model of convolutional neural network and Light GBM shows better results in short-term subway passenger flow.Based on the above work,this paper makes an in-depth analysis of the application of subway passenger flow prediction results,and puts forward the visualization application of prediction results,as well as the application of subway operation scheduling,capacity coordination,and passenger travel.