A method for predicting passenger flow in a subway network

Abstract

The present invention discloses a method for predicting the passenger flow of a subway network, which includes the following steps: Step 1, passenger flow data collection; Step 2, data preprocessing: analyze the characteristics of long-term trend changes, seasonal changes, weekly cyclic changes, and abnormal changes in the prediction input data, and preprocess the prediction input data; Step 3, passenger flow prediction on normal days: (1) According to the preprocessed time series data, establish an ARIMA passenger flow prediction model, perform parameter calibration and model optimization tests, output the passenger flow prediction results and conduct index evaluation; (2) According to the preprocessed time series data, establish an LSTM artificial neural network passenger flow prediction model, conduct batch training, set the learning rate lr and the number of training times epoch, output the fitting results and denormalize them, conduct multiple rounds of training, calculate the evaluation indexes and optimize the model parameters, conduct passenger flow prediction and output the prediction results; Step 4, passenger flow prediction on holidays: The present invention takes a short time and the passenger flow prediction level is stable.

Description

Technical Field

[0001] The present invention relates to the technical field of subway network passenger flow prediction, and particularly to a subway network passenger flow prediction method. Background Art

[0002] Accurate passenger flow prediction is the key to accurate capacity allocation in the urban rail industry. The passenger flow prediction results provide scientific support for capacity allocation of the network during holidays and large-scale sports events, and provide decision-making support for personnel guarantee, train operation organization, and passenger flow organization of urban rail transit operation units, playing an important guiding role in dispatching command and on-site operation. Common passenger flow prediction methods have low prediction accuracy, overly simple prediction models, and lack of theoretical depth. Manual prediction takes a long time and the passenger flow prediction level is unstable.

[0003] The existing double exponential smoothing method has two main defects: one is that it requires the passenger flow prediction input data to show a linear growth trend and a relatively long smoothing chain. Since the actual historical passenger flow data of the network does not fully meet the requirements of the model, the prediction results will have large deviations; the other is that different weights are given to the passenger flow prediction input data in chronological order using the exponential smoothing coefficient. In actual prediction, different exponential smoothing coefficients are adopted due to different judgments of the passenger flow law by the predictors, resulting in non-unique prediction results and it is difficult to determine a more appropriate predicted value. Summary of the Invention

[0004] In order to overcome the above deficiencies of the prior art, the purpose of the present invention is to provide a subway network passenger flow prediction method, which can quickly respond to passenger flow fluctuations, formulate corresponding capacity allocation plans, take effective passenger transport organization measures, ensure the balance of resource allocation and the safety of the operation process, take less time, and have a stable passenger flow prediction level.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0006] A subway network passenger flow prediction method, comprising the following steps;

[0007] Step 1, passenger flow data collection:

[0008] The passenger flow data is obtained from the network data warehouse, and the data source is the clearing data given by the ACC system. According to the prediction index, spatial granularity, time granularity, and line or specific station, the corresponding historical passenger flow data is selected;

[0009] Step 2, data preprocessing:

[0010] Analyze the characteristics of long-term trend changes, seasonal changes, weekly cycle changes, and abnormal changes of the prediction input data. On this basis, preprocess the prediction input data;

[0011] Step 3. Usual-day passenger flow prediction:

[0012] (1) Based on the time series data preprocessed in Step 2, establish an ARIMA passenger flow prediction model, conduct parameter calibration and model optimization tests, output the passenger flow prediction results and conduct index evaluations;

[0013] (2) Based on the preprocessed time series data, establish an LSTM artificial neural network passenger flow prediction model, conduct batch training, set the learning rate lr and the number of training epochs, output the fitting results and denormalize them, conduct multiple rounds of training, calculate evaluation indicators and optimize model parameters, conduct passenger flow prediction and output the prediction results;

[0014] Step 4. Holiday passenger flow prediction:

[0015] (1) Divide the urban rail transit calendar into (legal) holidays, large-scale events, weekdays, Saturdays and Sundays;

[0016] (2) Based on the same holidays and similar holidays, and using weekdays as a reference, adopt the proportional coefficient method to predict holiday passenger flow;

[0017] (3) First, use the short-term passenger flow prediction model to predict the normalized passenger flow, and then adjust it through the holiday fluctuation coefficient to predict the holiday passenger flow.

[0018] The preprocessing in Step 2 includes: data cleaning, data stationarization, and data normalization:

[0019] Data cleaning specifically includes: eliminating the historical passenger flow data with highly predictable large passenger flows due to holidays, large-scale events, etc., or historical data that causes sudden changes in passenger flow patterns. During the period of normalized epidemic prevention and control, the passenger flow data can be used as prediction input data;

[0020] Data stationarization specifically includes: conducting a stationarity test on the predicted input time series data, observing the trend changes of the data, calculating the moving average and standard deviation of the data, conducting a unit root test. If it is stationary, conduct a pure randomness test. After passing the test, establish a model; if the data is not stationary, conduct differencing processing, and the maximum differencing order does not exceed 2. If the differenced data is stationary, conduct a pure randomness test. After passing the test, establish a model; if the data still does not meet the stationarity condition after differencing processing, reselect the predicted input data, expand the selection range, and repeat the above operations.

[0021] Data normalization: For historical passenger flow data with a large amount and a wide time range, normalization processing is required to improve the efficiency of model data processing and training.

[0022] In step 3, the index evaluation specifically includes: the parameter calibration uses the autocorrelation coefficient and the partial autocorrelation coefficient; the goodness of fit is measured by AIC and BIC; the model test uses the standard residual, sum density, normal distribution, and autocorrelation function; the evaluation indexes include the mean absolute error, mean square error, root mean square error, and mean absolute percentage error.

[0023] In (1) of step 4, the calendar is divided into:

[0024] 1) Holidays: New Year's Day, Spring Festival, Tomb-Sweeping Day, Labor Day, Dragon Boat Festival, Mid-Autumn Festival, National Day. Among them, the Spring Festival does not include the day before the festival, and the rest of the festivals include the day before the festival.

[0025] 2) Large-scale events: Lantern Festival, Women's Day, Christmas Eve, etc., without the day before the event.

[0026] 3) Working days: All normal working days (Monday - Friday) except holidays and large-scale events.

[0027] 4) Saturdays and Sundays: All normal Saturdays and Sundays except holidays and large-scale events.

[0028] In (2) of step 4, the holiday passenger flow prediction by the proportional coefficient method specifically includes the following steps:

[0029] 1) Passenger flow prediction for the day before the festival, and the expression is

[0030]

[0031] In the formula, Y represents the predicted passenger flow data for the day before the festival, X represents the passenger flow data for the day before the same or similar holidays in history, y represents the recent passenger flow data, x represents the passenger flow data for the same period of the same or similar holidays in history, and δ represents the correction coefficient;

[0032] 2) Passenger flow prediction during the festival, and the expression is

[0033]

[0034] In the formula, Y' represents the predicted passenger flow data during the festival, Y represents the predicted passenger flow data for the day before the festival, X represents the passenger flow data for the day before the same or similar holidays in history, x' represents the passenger flow data during the same or similar holidays in history, and η represents the correction coefficient.

[0035] Taking the recent normal day passenger flow as a reference, the prediction result is checked.

[0036] In (3) of step 4, the short-term passenger flow prediction model for predicting the normalized passenger flow includes the following steps:

[0037] 1) Fit and predict the passenger flow on historical holidays, denoted as x;

[0038] 2) Compare with the true value X of the holiday passenger flow to determine the fluctuation coefficient;

[0039]

[0040] 3) Conduct time series prediction of holiday passenger flow, denoted as y;

[0041] 4) Correct the predicted value of the time series passenger flow;

[0042] Y = y·η·ε

[0043] That is

[0044]

[0045] In the formula, Y is the predicted target value, y is the target time series predicted value, X is the true passenger flow value of historical same or similar holidays, x is the fitting value of the time series passenger flow prediction of historical same or similar holidays, and ε is the correction coefficient.

[0046] Advantages of the present invention:

[0047] First, the principle of the passenger flow prediction model of the present invention is based on the improved ARIMA (Autoregressive Integrated Moving Average Model), LSTM (Long Short-Term Memory) artificial neural network time series prediction theory; second, the prediction model is optimized in combination with the operation scenario, the system responds quickly to the passenger flow fluctuation, the model parameters are adaptively adjusted, and the manual intervention is less, and the operation is more convenient compared with the traditional prediction model.

[0048] For the existing operating stations, operating lines and network-level inbound volume, outbound volume, transfer volume, passenger volume and section passenger volume indicators, short-term passenger flow prediction is carried out according to daily, weekly, monthly and custom cycles, according to the morning peak, evening peak and full-day custom time periods, and according to the time granularity of 5 minutes, 15 minutes, 30 minutes, 1 hour and full day. The response to the passenger flow fluctuation is rapid, and it can effectively guide the operation dispatching command and on-site operation.

[0049] In practical applications, the passenger flow prediction results provide scientific support for the network capacity allocation of the operating enterprise during holidays and large-scale sports events, provide decision-making support for operation guarantee, train operation organization and passenger flow organization, and play an important guiding role in dispatching command and on-site operation. Brief Description of the Drawings

[0050] Figure 1 It is the modeling flow chart of the ARIMA model of the present invention.

[0051] Figure 2 It is the schematic diagram of the modeling steps of the ARMA stationary sequence of the present invention.

[0052] Figure 3 This is the flowchart for modeling the LSTM neural network model of the present invention. Detailed implementation manners

[0053] The present invention will be further described in detail below with reference to the accompanying drawings.

[0054] As Figures 1 - 3 shown: A subway network passenger flow prediction method is applied to short-term passenger flow prediction of urban rail transit. Passenger flow prediction indicators are selected, the spatial dimension and time granularity are set, and ACC passenger flow data is obtained as the prediction input data; a prediction period is set, and according to the prediction input data, a passenger flow prediction model is established; according to the passenger flow prediction model, a passenger flow prediction result is obtained and query and visualization display are performed.

[0055] Predict the inbound volume, outbound volume, transfer volume, passenger volume and sectional passenger flow indicators of existing operating stations, operating lines and network levels.

[0056] The prediction period is daily, weekly, monthly and custom periods.

[0057] The prediction time period is the morning peak, evening peak or the entire day's custom time period.

[0058] The time granularity is 5 minutes, 15 minutes, 30 minutes, 1 hour and the whole day.

[0059] According to different prediction requirements, select the corresponding historical passenger flow data and use the corresponding model for prediction. Specifically:

[0060] (1) For the passenger flow prediction on ordinary days, when using the ARIMA model, 14 days of historical data is default selected for single-day passenger flow prediction, and 30 days of historical data is default selected for weekly, monthly and custom periods; when using the LSTM artificial neural network model, 30 days of historical data is default selected.

[0061] (2) For the passenger flow prediction on holidays, the same historical holiday or a recent similar holiday is used as the similar day; when using the proportional coefficient method, the passenger flow on the day before the holiday is predicted first, and then the passenger flow during the holiday is predicted; when using the combined prediction model, the passenger flow is predicted according to ordinary days first, and then corrected according to the passenger flow growth ratio during holidays.

[0062] Perform stationarity test and pure randomness test on historical data, and perform residual white noise test on the prediction result; for weekly, monthly and custom periods, the rolling prediction method is used for model parameters.

[0063] Perform normalization and anti-normalization processing on historical data; for weekly, monthly and custom periods, the sliding time window training and prediction method is used.

[0064] Visualize the predicted passenger flow data and perform trend analysis, set a warning threshold to warn the predicted values, and generate a predicted data report.

[0065] Embodiment 1 of the present invention proposes a subway network passenger flow prediction method, including the following steps:

[0066] 1. According to the passenger flow prediction target, set prediction indicators, prediction cycles, time granularity, and spatial ranges, and obtain the passenger flow data of the ACC clearing system as the predicted input data.

[0067] 2. Preprocess the predicted input time series passenger flow data, mainly including: observing the trend chart of the time series passenger flow data; removing abnormal passenger flow data and replacing it with similar data to ensure the time continuity of the time series passenger flow data; performing a stationarity test on the time series data;

[0068] For example, calculate the moving average value and standard deviation of the time series passenger flow data, and perform a graphical test of stationarity; perform a unit root test. For a first-order autoregressive model, the DF test can be used for the unit root test, and for a p-order autoregressive model, the ADF test can be used for the unit root test; construct a test statistic for a pure randomness test, such as the Q statistic and the LB statistic, to test whether there is a correlation between the time series passenger flow values.

[0069] 3. The ARIMA model expression is as follows:

[0070] The autoregressive integrated moving average model, denoted as ARIMA(p,d,q), has the following expression:

[0071]

[0072] In the formula, is the autoregressive coefficient polynomial of the stationary invertible ARMA(p,q) model; Θ(B) = 1 - θ1B - … - θ q B q , is the moving average coefficient polynomial of the stationary invertible ARMA(p,q) model.

[0073] 4. If the predicted input data is a stationary non-white noise sequence after data preprocessing, use the ARMA model for sequence modeling. The modeling steps are as follows:

[0074] 1) Calculate the values of the autocorrelation function (ACF) and partial autocorrelation function (PACF) of the passenger flow time series.

[0075] 2) According to the properties of the ACF and PACF, select an ARMA(p,q) model with an appropriate order for fitting.

[0076] Table 1 Properties of the (Partial) Autocorrelation Function

[0077] Correlation coefficient AR(p) MA(q) ARMA(p,q) ACF Trailing Truncated after q terms Trailing PACF Truncated after p terms Trailing Trailing

[0078] 3) Estimate the values of the unknown parameters in the model using the method of moment estimation, maximum likelihood function method, and least squares method.

[0079] 4) Significance test of the model and parameters. If the model test fails, reselect the parameters for fitting.

[0080] 5) Model optimization. If the model passes the test, establish multiple fitting models and select the optimal model parameters, specifically including:

[0081] AIC criterion: AIC = -2ln(value of the maximum likelihood function of the model) + 2 (number of unknown parameters in the model)

[0082] BIC criterion: SBC = -2ln(value of the maximum likelihood function of the model) + (ln n) (number of unknown parameters in the model)

[0083] 6) Using the fitting model, use the recursive method to roll and predict the passenger flow data at any future moment.

[0084] 5. If the predicted input data is still a non-stationary sequence after data preprocessing, perform differencing processing, and the processing method is as follows:

[0085] 1) Perform first-order differencing on the original time series prediction input data, and conduct stationarity and pure randomness tests on the differenced time series data. If the time series data after first-order differencing is a stationary non-white noise sequence, use the ARMA model for modeling, and the steps are as in 3.

[0086] 2) After performing first-order differencing on the original time series prediction input data, if the stationarity condition is still not satisfied, perform second-order differencing, and conduct stationarity and pure randomness tests on the time series data after second-order differencing. If the time series data after second-order differencing is a stationary non-white noise sequence, use the ARMA model for modeling, and the steps are as in 3.

[0087] 3) After performing second-order differencing on the original time series prediction input data, if the stationarity condition is still not satisfied, do not continue with differencing processing, reselect the prediction input data and perform preprocessing, specifically including expanding the date selection range, removing bad values, etc., and repeat the above steps.

[0088] 4) Perform differencing reduction on the prediction results.

[0089] 6. If the original data or the processed time series data satisfies the stationarity condition but does not satisfy the non-white noise sequence condition, do not perform modeling, reselect the prediction input data, specifically including expanding the date selection range, removing bad values, etc., and repeat step 4.

[0090] 7. Analyze the model prediction results, specifically including: performing a fitting analysis on the model prediction results and evaluating the results through indicators such as MAE, MSE, RMSE, and MAPE; visually displaying the system of the model prediction results and giving early warnings for passenger flow.

[0091] Embodiment 2 of the present invention proposes a subway network passenger flow prediction method, including the following steps:

[0092] 1. According to the passenger flow prediction target, set the prediction indicators, prediction period, time granularity, and spatial range, and obtain the passenger flow data of the ACC clearing system as the prediction input data.

[0093] 2. Preprocess the predicted input time series passenger flow data, mainly including: observing the trend graph of the time series passenger flow data; removing abnormal passenger flow data and replacing it with similar data to ensure the time continuity of the time series passenger flow data; performing normalization processing.

[0094] Use the Min - Max method to perform normalization processing on the predicted input data, and its expression is: [[ID=?]] [[ID=?]]

[0095] [[ID=?]] [[ID=?]] [[ID=?]]

[0096] where x min is the minimum value in the data sequence, x max is the maximum value in the data sequence, and X is the normalized data sequence.

[0097] 3. Establish an LSTM model, and the model expression is as follows:

[0098] i t = σ(W xi x t + W hi h t-1 + W ci c t-1 + b i )

[0099] f t = σ(W xf x t + W hf h t-1 + W cf c t-1 + b f )

[0100] c t = f t c t-1 + i t tan h(W xc x t + W hc h t-1 + b It should be noted that there are some consecutive tags without clear content in the original text (such as -

[0096] ), which may affect the integrity of the understanding of the formula. If possible, it is recommended to check and supplement the complete information.c )

[0101] o t = σ(W xo x t + W ho h t-1 + W co c t + b o )

[0102] h t = o t tanh(c t )

[0103] Among them, i is the input gate, f is the forget gate, and o is the output gate. W and b are the corresponding weight coefficient matrices and bias terms respectively. σ and tanh are the sigmoid and hyperbolic tangent activation functions respectively.

[0104] 4. Perform batch training with a learning rate lr = 0.001, train 150 times, output the fitting result and perform inverse normalization.

[0105] 5. Perform model prediction and result evaluation, and conduct system visualization display and passenger flow early warning for the prediction results.

[0106] Embodiment 3 of the present invention proposes a subway network passenger flow prediction method, including the following steps:

[0107] 1. Selection of similar days

[0108] 1) When the objective conditions are the same, select the same holiday as the historical reference for passenger flow prediction.

[0109] 2) When the objective conditions change, select the recent similar holiday as the historical reference.

[0110] 3) For holidays or large-scale events with short dates, select the recent normal days as the historical reference.

[0111] 2. On the basis of determining the similar holiday as the historical reference, use the proportional coefficient method to predict the passenger flow on the day before the holiday.

[0112] 3. Using the predicted value of the passenger flow on the day before the holiday and the true value of the passenger flow on the recent normal days as references, predict the passenger flow data during the holiday.

[0113] Embodiment 4 of the present invention proposes a subway network passenger flow prediction method, including the following steps:

[0114] 1. Determine the similar days according to the method described in Embodiment 3.

[0115] 2. Use the short-term passenger flow prediction model to perform time series fitting prediction on the passenger flow of historical similar days.

[0116] 3. Compare with the true value of holiday passenger flow to determine the fluctuation coefficient.

[0117] 4. Use the same short-term passenger flow prediction model to conduct time series prediction on the normalized passenger flow of the target date.

[0118] 5. Correct the time series passenger flow prediction value according to the fluctuation coefficient.

[0119] The passenger flow prediction system based on autoregression and artificial neural network of the present invention has advanced theory, mature technology and high prediction accuracy. It can replace the traditional passenger flow prediction model and provide a scientific basis for transportation organization planning.

[0120] By using it in combination with the manual prediction method, the average absolute error of the line network passenger flow prediction during holidays and large-scale sports events is further reduced from 10% to within 5%. For example, on the day before the Labor Day in 2021, the predicted passenger volume was 4.535 million person-times, and the actual passenger volume was 4.4821 million person-times, with a prediction deviation of only about 1%; on the day before the National Day, the predicted passenger volume was 3.85 million person-times, and the actual passenger volume was 3.9599 million person-times, with a prediction deviation of only about 3%. It effectively helps the operation enterprise to make operation guarantee decision-making arrangements such as train operation organization, passenger transport organization, and personnel guarantee during major holidays, effectively prevents the predictable large passenger flow during holidays, and then improves the satisfaction of passengers.

[0121] With the continuous advancement of the wave of smart urban rail construction, through the form of independent research and joint development, the mastery of new technologies enables the operation enterprise to take the initiative in the development and use of the passenger flow prediction system.

[0122] The passenger flow prediction system and the passenger flow monitoring and early warning system complement each other, jointly constituting the subway passenger flow prediction system, initially realizing accurate passenger flow prediction and accurate matching of transport capacity, and further improving the transport control ability and operation service level.

[0123] Economic benefits:

[0124] The passenger flow prediction model based on autoregression and artificial neural network is independently researched by the line network control center, and the whole process of basic research, theoretical research, programming implementation and data verification is completed. After the system of the present invention is put into use, it reduces the time spent by the NCC and transportation technicians in passenger flow statistics and analysis, and improves the efficiency of transportation organization planning. Before the system was put into use, the time spent on passenger flow statistics and analysis was about 1 hour per day. After the system was launched, the data verification time was 0.5 hour per day, reducing the time cost of manual calculation and analysis by 0.5 hour per day and reducing the working hours by 50%.

Claims

1. A method for predicting the passenger flow of a subway network, characterized in that, Including the following steps; Step 1, Passenger flow data collection: The passenger flow data is obtained from the line network data warehouse, and the data source is the clearing data given by the ACC system. According to the prediction indicators, spatial granularity, time granularity, and lines or specific stations, the corresponding historical passenger flow data is selected; Step 2, Data preprocessing: Analyze the long-term trend changes, seasonal changes, weekly cycle changes, and abnormal change characteristics of the prediction input data. On this basis, preprocess the prediction input data; Step 3, Passenger flow prediction on normal days: (1) According to the time series data preprocessed in Step 2, establish an ARIMA passenger flow prediction model, conduct parameter calibration and model optimization tests, output the passenger flow prediction results and conduct index evaluation; (2) According to the preprocessed time series data, establish an LSTM artificial neural network passenger flow prediction model, conduct batch training, set the learning rate lr and the number of training epochs epoch, output the fitting results and denormalize them, conduct multiple rounds of training, calculate the evaluation indicators and optimize the model parameters, conduct passenger flow prediction and output the prediction results; Step 4, Passenger flow prediction on holidays: (1) Divide the urban rail transit calendar into holidays, large-scale events, weekdays, Saturdays and Sundays; (2) Based on the same holidays and similar holidays, and taking normal days as a reference, use the proportional coefficient method to predict the passenger flow on holidays; (3) First, use the short-term passenger flow prediction model to predict the normalized passenger flow, and then adjust it through the holiday fluctuation coefficient to predict the passenger flow on holidays.

2. The subway network passenger flow prediction method according to claim 1, characterized in that The preprocessing in Step 2 includes: data cleaning, data stationarization, and data normalization: Data cleaning specifically includes: eliminating the historical passenger flow data with large predictable passenger flows due to holidays and large-scale events, or historical data with sudden changes in passenger flow patterns affected by the COVID-19 pandemic. During the period of normalized COVID-19 prevention and control, the passenger flow data is used as the prediction input data; Data stationarization specifically includes: conducting a stationarity test on the prediction input time series data, observing the trend changes of the data, calculating the moving average and standard deviation of the data, conducting a unit root test. If it is stationary, conduct a pure randomness test. After passing the test, establish a model; if the data is not stationary, conduct differencing processing, and the maximum differencing order does not exceed 2 orders. If the data after differencing processing is stationary, conduct a pure randomness test. After passing the test, establish a model; if the data still does not meet the stationarity condition after differencing processing, reselect the prediction input data, expand the selection range, and repeat the data stationarization operation; Data normalization: For historical passenger flow data with a large amount of data and a wide time range, data normalization is required to improve the efficiency of model data processing and training.

3. The subway network passenger flow prediction method according to claim 1, characterized in that, The index evaluation in Step 3 specifically includes: The autocorrelation coefficient and partial autocorrelation coefficient are used for parameter calibration; The goodness of fit is measured by AIC and BIC; The model test uses the standard residual, sum of density, normal distribution, and autocorrelation function; The evaluation indicators include the mean absolute error, mean squared error, root mean squared error, and mean absolute percentage error.

4. The subway network passenger flow prediction method according to claim 1, wherein In (1) of Step 4, the calendar is divided into: 1) Holidays: New Year's Day, Spring Festival, Tomb-Sweeping Day, Labor Day, Dragon Boat Festival, Mid-Autumn Festival, National Day. For Spring Festival, the day before the festival is not included, while for the other festivals, the day before the festival is included; 2) Major events: Lantern Festival, Women's Day, Christmas Eve, without the day before the event; 3) Working days: All normal working days except holidays and major events; 4) Saturdays and Sundays: All normal Saturdays and Sundays except holidays and major events.

5. A subway network passenger flow prediction method according to claim 1, characterized in that The holiday passenger flow prediction by the proportional coefficient method in (2) of step 4 specifically includes the following steps: 1) Passenger flow prediction for the day before the festival, with the expression Wherein, represents the predicted passenger flow data on the day before the holiday, represents the passenger flow data on the day before the same or similar holidays in history, represents the recent passenger flow data, represents the passenger flow data during the same period of the same or similar holidays in history, represents the correction coefficient; 2) Passenger flow prediction during the festival, with the expression Wherein, represents the predicted passenger flow data during the festival period, represents the predicted passenger flow data on the day before the festival, represents the passenger flow data on the day before the same or similar historical holidays, represents the passenger flow data during the same or similar historical holidays, represents the correction coefficient; Taking the passenger flow on recent ordinary days as a reference, the prediction results are verified.

6. The subway network passenger flow prediction method according to claim 1, wherein The prediction of the normalized passenger flow by the short-term passenger flow prediction model in (3) of step 4 includes the following steps: 1) Fit and predict the passenger flow on historical holidays, denoted as ; 2) Compare with the true value of holiday passenger flow to determine the fluctuation coefficient; 3) Conduct the time series prediction of holiday passenger flow, denoted as ; 4) Correct the predicted value of the time-series passenger flow; That is In the formula, is the predicted target value, is the predicted value of the target time series, is the actual passenger flow value of the same or similar holidays in history, is the fitting value of the predicted time series of passenger flow for the same or similar holidays in history, is the correction coefficient.

Images