Prediction Of Short-time Passenger Flow Of Metro Based On Combined Features

After the reform and opening up,China's rapid economic development and increasing population have promoted the vigorous development of the transportation industry,especially the promotion and application of subways,which greatly facilitated citizens' travel.However,as the structure of subway lines becomes more complex and the population in cities is increasing,the number of passengers carried by subways has also increased.The overload of subway brings a lot of problems to the operation scheduling and management of the relevant department of metro,so the prediction of the number of short-time passengers of subway is particularly important.Based on the above background,this paper takes the passenger flow of the subway in Hangzhou as the research object,and puts forward a set of methods that can accurately predict the number of short-time inflow and outflow of passengers of subway,which provides decision-making basis for the operation scheduling and management of the relevant department of metro.The main research work of this paper is as follows: put forward an effective method of selecting the preliminary features,extract the relevant spatiotemporal features,and predict the number of short-time inflow and outflow of passengers of subway.(1)The Light Gradient Boosting Machine(LightGBM)model is used to select preliminary features for predicting the number of short-time passengers of subway.The LightGBM model is used to calculate the importance ranking of features,and the same set of parameters is used to verify the prediction errors under different number of features.According to the experimental results,the top 20 features of importance ranking are selected as the preliminary features of predicting the number of passengers entering the subway,and the top 16 features of importance ranking are selected as the preliminary features of predicting the number of passengers leaving the subway.(2)The Convolutional Long Short Term Memory Network(ConvLSTM)model is used to extract the spatiotemporal features related to the number of short-time passengers of subway.The preliminary features selected by the LightGBM model are rearranged according to the time period,station and the dimensions of the preliminary features,and the size of the preliminary features is changed into: days ? the number of time periods per day ? the number of stations ? dimensions of preliminary features.After inputting the data with changed dimension into the ConvLSTM model to obtain the spatiotemporal features,the dimension of data becomes: days ? the number of time periods per day ? the number of stations ? dimensions of spatiotemporal features.Then the spatiotemporal features are rearranged,and the size of the spatiotemporal features becomes: the number of samples ? the dimensions of the spatiotemporal features.(3)The Long Short Term Memory Network(LSTM)model is used to predict the number of short-time inflow and outflow of passengers of subway.First,the preliminary features are combined with the spatiotemporal features,and then the LSTM model is used to predict the number of passengers of subway.In order to analyze the influence of different parameters on the prediction results,based on the number of hidden layers of the LSTM model,the learning rate,the number of iterations and the number of neurons in each hidden layer,comparative experiments of five sets of parameters is performed in this paper to obtain the predicted values of the number of passengers entering and leaving the subway under the optimal parameters.At the same time,according to the results of the comparative experiments,the feasibility of the model proposed in this paper in predicting the number of short-time passengers of subway is verified.In conclusion,the prediction method proposed in this paper can effectively predict the number of short-time inflow and outflow of passengers of subway,so it can provide reliable decision-making basis for the operation scheduling and management of the relevant department of metro.