A Passenger Flow Organization Decision-making Method Based on Station Passenger Flow Prediction

Abstract

The present invention discloses a passenger flow organization decision-making method based on station passenger flow prediction, comprising the following steps: collecting historical data; respectively constructing a station passenger flow prediction model according to different situations; predicting the passenger flow volume of the station based on the station passenger flow prediction model; classifying the predicted passenger flow volume into passenger flow levels; and giving suggestions for passenger flow organization decision-making according to the passenger flow levels. The above technical solution constructs a corresponding inbound and outbound passenger flow prediction model for each station according to different situations, determines the passenger flow level according to the predicted inbound and outbound passenger flow volume, and thus gives suggestions for passenger flow organization. By performing short-term passenger flow prediction on rail transit stations and using the prediction results as a reference for the passenger flow organization plan, the algorithm is used to classify and determine the predicted passenger flow values, so as to quickly and accurately give suggestions for passenger flow organization decision-making, improving the disadvantages of the cumbersome and delayed manual observation of passenger flow volume and then making decisions.

Description

Technical Field

[0001] The present invention relates to the technical field of urban rail transit transportation, and particularly relates to a passenger flow organization decision-making method based on station passenger flow prediction. Background Art

[0002] With the networked development of rail transit, the operation scale has increased sharply, and the passenger volume has continued to rise. In order to ensure the balance between the demand and supply of rail transit resources, improve the operation efficiency, and avoid resource waste, while vigorously developing urban rail transit, it is necessary to effectively allocate system resources and timely adjust operation strategies.

[0003] The occurrence possibility, occurrence time, and consequences of urban rail transit operation emergencies are uncertain. Under the networked conditions, the correlation between urban rail transit operation lines and the complexity of operation risks are intensified, and operation emergencies may affect the normal operation of other lines through transfer stations. Therefore, in order to effectively respond to sudden accidents and other safety derivative events and do a good job in corresponding emergency handling and post-disaster recovery work, a highly safe and reliable emergency command decision-making is required to ensure the safe, normal, and effective operation of urban rail transit.

[0004] Station passenger flow is a very important and fundamental link in rail transit operation. Therefore, the organization and service work of station passenger flow is particularly important. Moreover, when the station encounters a large passenger flow, it will face greater pressure than usual. Therefore, predicting the passenger flow trend in the future time period of the station can provide a decision-making buffer time for the station passenger flow organization work and can effectively reduce the risk of urban rail transit accidents caused by large passenger flows.

[0005] There is information showing that the current passenger flow control methods and measures are mainly judged and implemented based on the subjective experience of station operation management personnel, lacking systematic and objective theoretical research on when to start passenger flow control and how to arrange control measures. Therefore, to a certain extent, it increases the difficulty of the station passenger flow organization work during peak periods.

[0006] In order to better adapt to the characteristics of the dynamic changes of passenger flow in urban rail transit stations, improve the matching between rail transit operation arrangements and passenger flow demands, and reduce the travel costs of passengers and the operation costs of enterprises, it is necessary to construct a short-term passenger flow prediction model based on the dynamically changing station passenger flow, and do a good job in the station passenger flow organization work in advance according to the predicted passenger volume to ensure the efficient and safe operation of rail transit.

[0007] Chinese patent document CN105404850B discloses a "station passenger flow monitoring system, station passenger flow monitoring method and station passenger flow control system, station passenger flow control method". The monitoring system includes a data acquisition system, a data processing system, etc.; the data acquisition system includes multiple infrared counting, video recognition counting and gate counting devices set up at the station; these devices are all connected to the data processing system; the monitoring method includes: dividing the station into a non-paid area, a paid area, and a platform area, and further separating the three areas by setting up security inspection equipment, entry and exit gates, etc., and then transmitting the collected data to the data processing system for processing through infrared counting, gate counting and other equipment to obtain the two-way passenger flow in and out of the three areas. The above technical solution only adopts a single monitoring scheme, and does not consider the impact of different passenger flow prediction schemes under different circumstances on accuracy. Summary of the invention

[0008] The present invention mainly solves the technical problem that the original technical solution is difficult to effectively deal with the complicated correlation between urban rail transit operating lines by using a single monitoring solution. It provides a passenger flow organization decision-making method based on station passenger flow prediction, builds a corresponding inbound and outbound passenger flow prediction model for different situations of each station, determines the passenger flow level according to the predicted inbound and outbound passenger flow, and thus gives suggestions for passenger flow organization. By making short-term passenger flow predictions for rail transit stations, the prediction results are used as a reference for the passenger flow organization plan, instead of adjusting the passenger flow organization plan after the passenger flow changes. An algorithm is used to grade the predicted passenger flow values, so as to quickly and accurately give suggestions for passenger flow organization decisions, thereby improving the shortcomings of cumbersome and delayed decision-making based on manual observation of passenger flow, and greatly improving the targeted processing efficiency of stations in different situations.

[0009] The above technical problem of the present invention is mainly solved by the following technical solution: The present invention comprises the following steps:

[0010] S1 collects historical data;

[0011] S2 constructs station passenger flow prediction models according to different situations;

[0012] S3 predicts the passenger flow of the station based on the station passenger flow prediction model;

[0013] S4 classifies the predicted passenger flow into passenger flow grades;

[0014] S5 provides suggestions for passenger flow organization decisions based on passenger flow levels.

[0015] Preferably, the historical data collected in step S1 includes counting the daily passenger flow in and out of stations, counting the date types of each day and the events occurring on that day within a certain historical period, and counting the concourse and platform areas of each station in the network, the number of entrances and exits, and the passenger flow levels corresponding to the passenger flow in and out of stations in different 10-minute intervals within a certain historical period.

[0016] Preferably, in step S2, the short-term passenger flow prediction models for stations are constructed according to different situations, and the different situations specifically include: days with emergencies, days when large-scale events are held, ordinary holidays without emergencies and large-scale events, and ordinary non-holidays without emergencies and large-scale events. The above different situations are ranked from high to low in priority.

[0017] Preferably, for the days with emergencies, the gain method is used to construct the short-term passenger flow prediction model for urban rail transit stations in case of emergencies. For the passenger flow in and out of stations under emergencies at different stations, it is obtained by multiplying the passenger flow in and out of stations under the normal operation of the station by a gain coefficient.

[0018] Preferably, for the days when large-scale events are held, SVM is used to construct the short-term passenger flow prediction model for urban rail transit stations in case of large-scale events.

[0019] The optimization objective function of the SVM model is The passenger flow data points (X i , Y i ) of large-scale events are fitted to a linear model . Defining a constant ε>0, the loss function metric of the SVM model is:

[0020]

[0021] Preferably, for ordinary holidays without emergencies and large-scale events, the ARIMA model is used to construct the short-term passenger flow prediction model for urban rail transit stations in case of holidays. The specific steps of the process are as follows: S3.31 Obtain the passenger flow time series data in case of holidays;

[0022] S3.32 Plot the data to observe whether it is a stationary time series. If it is a non-stationary time series, perform a d-order difference operation first. The difference method is as follows, and it is transformed into a stationary time series;

[0023] d = 0, y t = Y t

[0024] d = 1, y t = Y t - Y t-1

[0025] d = 2, yt =(Y t -Y t-1 )-(Y t-1 -Y t-2 )

[0026] ……

[0027] S3.33 Calculate the autocorrelation coefficient ACF and partial autocorrelation coefficient PACF of the stationary time series respectively. Through the analysis of the autocorrelation plot and partial autocorrelation plot, obtain the order p and order q;

[0028] S3.34 According to q and p, obtain the ARMA model. The mathematical expression of the model is as follows:

[0029]

[0030] Among them, φ represents the coefficient of AR, and θ represents the coefficient of MA;

[0031] S3.35 Conduct a white noise test on the obtained model. If the white noise test is not passed, update p and q;

[0032] S3.36 Until the model that passes the white noise test is the optimal ARMA model.

[0033] Preferably, for ordinary non-holiday periods without emergencies and large-scale events, an LSTM neural network is used to construct a short-term passenger flow prediction model for urban rail transit stations. The specific steps are as follows:

[0034] S3.41 The neural network structure consists of an input layer, a hidden layer, and an output layer;

[0035] S3.42 In the network training algorithm, select the Sigmoid function and Tanh function as the activation functions for the network hidden layer; use the adam algorithm to optimize the weights of the network.

[0036] The formula of the Sigmoid function is as follows:

[0037]

[0038] The formula of the Tanh function is as follows:

[0039]

[0040] Preferably, in step S4, a KNN model is used to construct a model for determining the classification of station passenger flow. The specific steps are as follows:

[0041] S4.1 Select the inbound volume, outbound volume, number of entrances and exits, and the area of the station concourse and platform as the eigenvalue of the model, and the passenger flow level as the output of the model;

[0042] S4.2 calculates the distance between the points in the historical passenger flow level sample set and the current point. The distance formula is as follows:

[0043]

[0044] S4.3 Sort in order of increasing distance;

[0045] S4.4 selects the k value and selects the k nearest neighbors, that is, each sample selects the k closest neighbors to represent;

[0046] S4.5 returns the category with the highest frequency of the first k points as the passenger flow level of the current point. When k=3, the category of the square point to be classified is the triangle category. When k=5, the category of the square point to be classified is the circle category.

[0047] Preferably, the passenger flow level is divided into 4 levels: Level 1 indicates that the incoming and outgoing volumes are normal; Level 2 indicates that the outgoing volume is excessive and the incoming volume is normal; Level 3 indicates that the incoming volume is excessive and the outgoing volume is normal; Level 4 indicates that both the incoming and outgoing volumes are excessive.

[0048] Preferably, the step S5 provides suggestions for passenger flow organization decision according to the passenger flow level, specifically including:

[0049] The recommendations for passenger flow organization decisions for passenger flow level 1 are: the station adopts the normal entry and exit ticket checking mode, but strengthens guidance and inspection at escalators and entry and exit gates to avoid congestion;

[0050] The following are the recommendations for passenger flow organization decisions for passenger flow level 2: When outbound passenger flow is congested at the outbound gate, the station will change some inbound gates from the inbound ticket inspection mode to the outbound ticket inspection mode and increase the number of outbound gates. If the congestion cannot be alleviated, the station will open the gate side door to allow passengers with one-way tickets to pass without inspection, and staff will manually collect the one-way tickets at the gate;

[0051] The following are the suggestions for passenger flow organization decisions for level 3 passenger flow: the station operation management personnel shall set up obstacles at various places in the entrance to guide the passenger flow to detour. When the passenger flow is congested at the entrance gate, and the passenger density in the platform and the paid area of ​​the station hall is not large, the station shall change the mode of some exit gates, increase the number of entrance gates, or set the entrance gates to the inspection-free mode, and open the side doors at the same time, so that passengers can enter the station without tickets and buy tickets when leaving the station. When the number of passengers stranded in the paid area of ​​the station hall reaches a certain number, the ticket sales shall be slowed down, and isolation fences shall be added in the entrance area. The staff shall organize passengers to queue up in an orderly manner to detour into the station. When the passengers on the platform cannot get on the train in time, causing passenger congestion, the station shall control the speed of passengers getting off the platform, the speed of passengers entering the paid area in time, close some entrance gates and temporary ticket booths, and arrange staff to intercept at the stairs, while maintaining the order of passengers waiting for the train on the platform.

[0052] For the passenger flow organization decision-making suggestions for passenger flow level 4: When the passenger flow aggregation volume at the entrance and exit channels is large, further increase may lead to difficulties for passengers to enter and leave the station. It is recommended to adjust the functions of some entrances and exits of the station, intercept the inbound passenger flow at the entrances and exits, and reopen the inbound channels after the congestion in the station is relieved. In terms of strengthening guidance and information dissemination, strengthen the patrol and guidance at key parts such as the station concourse, escalators, and platforms to avoid congestion and disputes among passengers and prevent safety accidents; continuously broadcast the safety broadcast for large passenger flows throughout the station to remind passengers to pay attention to safety, and use the PIS system at the platform and the inbound area to remind passengers of the current train operation interval and the latest train operation information.

[0053] The beneficial effects of the present invention are: constructing corresponding inbound and outbound passenger flow prediction models according to the different situations of each station, determining the passenger flow level based on the predicted inbound and outbound passenger flow volumes, and thus giving suggestions for passenger flow organization. By conducting short-term passenger flow prediction for rail transit stations and using the prediction results as a reference for the passenger flow organization plan, rather than adjusting the passenger flow organization plan only after the passenger flow changes, using algorithms to classify and determine the predicted passenger flow values, so as to quickly and accurately give suggestions for passenger flow organization decision-making, improving the shortcomings of the cumbersome and delayed decision-making by manually observing the passenger flow volume, and greatly improving the targeted processing efficiency of stations in different situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 is a flowchart of the present invention.

[0055] Figure 2 is a schematic diagram of the principle of the support vector machine of the present invention.

[0056] Figure 3 is a flowchart of the ARIMA algorithm of the present invention.

[0057] Figure 4 is a structural diagram of the LSTM neural network of the present invention.

[0058] Figure 5 is an effect diagram of the KNN model of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0059] The technical solutions of the present invention will be further specifically described below through embodiments in conjunction with the accompanying drawings.

[0060] Embodiment: A passenger flow organization decision-making method based on station passenger flow prediction according to this embodiment, as Figure 1 shown, includes the following steps:

[0061] S1 performs historical data collection. The collected historical data includes counting the daily passenger flow in and out of the station, counting the date type of each day and the events that occurred on that day within a certain historical period, as well as counting the area of the concourse and platform of each station in the network, the number of entrances and exits, and the passenger flow levels corresponding to the passenger flow in and out of the station in different 10-minute intervals within a certain historical period.

[0062] Preprocess the passenger flow data. Taking 10 minutes as the time granularity, count the daily passenger flow in and out of the station in a period before the prediction date of the prediction station. Organize the passenger flow data of each day as shown in Table 1.

[0063] Table 1 Passenger Flow Statistics Table of XX Station on XX / XX / XX

[0064]

[0065]

[0066] Count the date type of each day and the events that occurred on that day (if any, such as major events, concerts or major accidents, etc.) within a certain historical period. Organize the information of the date and the events that occurred as shown in Table 2.

[0067] Table 2 Date and Event Statistics Table of XX Station

[0068]

[0069] Count the area of the concourse and platform of each station in the network, the number of entrances and exits, and the passenger flow levels corresponding to the passenger flow in and out of the station in different 10-minute intervals within a certain historical period.

[0070] Table 3 Passenger Flow Level Statistics Table of XX Station

[0071]

[0072] The passenger flow levels are divided into 4 levels: Level 1 - indicating normal inbound and outbound volumes; Level 2 - indicating excessive outbound volume and normal inbound volume; Level 3 - indicating excessive inbound volume and normal outbound volume; Level 4 - indicating excessive inbound and outbound volumes.

[0073] S2 constructs a passenger flow prediction model for the station according to different situations respectively; constructs a passenger flow prediction model for the station according to different situations respectively. The different situations specifically include: days with emergencies, days with large-scale events, ordinary holidays without emergencies and large-scale events, and ordinary non-holidays without emergencies and large-scale events. The above different situations are ranked from high to low in priority.

[0074] On days with emergencies, the gain method is used to construct a short-term passenger flow prediction model for urban rail transit stations under emergency situations. For the inbound and outbound passenger flows of different stations under emergencies, it is obtained by multiplying the inbound and outbound passenger flows of the station under normal operation by a gain coefficient.

[0075] On days when large-scale events are held, the SVM is used to construct a short-term passenger flow prediction model for urban rail transit stations under large-scale event situations. The principle is as Figure 2 shown.

[0076] The optimization objective function of the SVM model is Fitting the passenger flow data points (X i , Y i ) of large-scale events to a linear model . Defining a constant ε > 0, the loss function metric of the SVM model is:

[0077]

[0078] Figure 2 The loss value of the points with outer circles in is Other points are points without loss.

[0079] On ordinary holidays without emergencies and large-scale events, the ARIMA model is used to construct a short-term passenger flow prediction model for urban rail transit stations under holiday situations. The process is as Figure 3 shown. The specific steps of the process are as follows:

[0080] S3.31 Obtain the passenger flow time series data under holiday situations;

[0081] S3.32 Plot the data to observe whether it is a stationary time series. If it is a non-stationary time series, first perform d-order difference operation (the difference method is as follows), and convert it into a stationary time series;

[0082] d = 0, y t = Y t

[0083] d = 1, y t = Y t - Y t-1

[0084] d = 2, y t = (Y t - Y t-1 ) - (Y t-1 - Y t-2 )

[0085] ……

[0086] S3.33 Calculate the autocorrelation coefficient ACF and partial autocorrelation coefficient PACF of the stationary time series respectively. Through the analysis of the autocorrelation graph and partial autocorrelation graph, obtain the order p and order q;

[0087] S3.34 According to q and p, obtain the ARMA model. The mathematical expression of the model is as follows:

[0088]

[0089] Among them, φ represents the coefficient of AR, and θ represents the coefficient of MA;

[0090] S3.35 Conduct a white noise test on the obtained model. If the white noise test is not passed, update p and q;

[0091] S3.36 Until the model that passes the white noise test is the optimal ARMA model.

[0092] For ordinary non-holiday days without emergencies and large-scale events, use the LSTM neural network to construct a short-term passenger flow prediction model for urban rail transit stations under non-holiday conditions. The specific steps are as follows:

[0093] S3.41 The neural network structure is the input layer, hidden layer, and output layer. The structure is as Figure 4 shown;

[0094] In the network training algorithm, select the Sigmoid function and Tanh function as the activation functions of the network hidden layer; use the adam algorithm to optimize the weights of the network.

[0095] The formula of the Sigmoid function is as follows:

[0096]

[0097] The formula of the Tanh function is as follows:

[0098]

[0099] S3 predicts the passenger flow of the station based on the station passenger flow prediction model; according to the actual situation on the prediction day, select the corresponding model for prediction. First, consider whether there is an emergency. If there is an emergency, select the emergency passenger flow prediction model. If not, then consider whether there is a large-scale event. If there is a large-scale event, select the large-scale event passenger flow prediction model. If not, then finally consider whether it is a holiday. If it is a holiday, select the holiday passenger flow prediction model. Otherwise, select the normal passenger flow prediction model.

[0100] Adopt the fitted or trained passenger flow prediction model, and select the correct prediction factors to predict the inbound and outbound passenger flow data of the target station within the target time period.

[0101] S4 performs passenger flow level classification on the predicted passenger flow; constructs a model for determining the passenger flow level of the station using the KNN model, and the specific steps are as follows:

[0102] S4.1 Selects the inbound volume, outbound volume, number of entrances and exits, and the area of the station concourse and platform as the feature values of the model, and the passenger flow level as the output of the model;

[0103] S4.2 Calculates the distance between the points in the historical passenger flow level sample set and the current point, and the distance formula is as follows:

[0104]

[0105] S4.3 Sorts in ascending order of distance;

[0106] S4.4 Selects the value of k, selects k nearest neighbors, that is, each sample selects k closest neighbors to represent;

[0107] S4.5 Returns the category with the highest frequency of occurrence among the top k points as the passenger flow level of the current point. As Figure 5 shown, when k = 3, the category of the point to be classified (square) is the triangle class, and when k = 5, the category of the point to be classified (square) is the circle class.

[0108] S5 Gives suggestions on passenger flow organization decisions according to the passenger flow level.

[0109] Suggestions on passenger flow organization decisions for passenger flow level 1 are as follows: The station adopts the normal inbound and outbound ticket checking mode, but strengthens the guidance and patrol at the escalators and inbound and outbound turnstiles to avoid congestion.

[0110] Suggestions on passenger flow organization decisions for passenger flow level 2 are as follows: When congestion occurs at the outbound turnstiles for outbound passenger flow, the station can change some inbound turnstiles from the inbound ticket checking mode to the outbound ticket checking mode, increase the number of outbound turnstiles. If the congestion situation fails to ease, the station can open the side door of the turnstile for passengers holding single - journey tickets to pass through without inspection, and the staff manually collects the single - journey tickets at the turnstiles.

[0111] The decision-making suggestions for passenger flow organization at level 3 are as follows: the station operation management personnel set up obstacles at various places in the entrance and guide the passenger flow to detour. When the passenger flow is congested at the entrance gate, and the passenger density in the platform and the paid area of ​​the station hall is not large, the station can change the mode of some exit gates and increase the number of entrance gates. It can also set the entrance gate to the inspection-free mode and open the side door at the same time, so that passengers can enter the station without tickets and pay for tickets when leaving the station. When the number of passengers stranded in the paid area of ​​the station hall reaches a certain number, the ticket sales will be slowed down, isolation fences will be added in the entrance area, and staff will organize passengers to queue up in an orderly manner to detour into the station. When passengers on the platform cannot get on the train in time, causing passenger congestion, the station needs to control the speed of passengers getting off the platform, promptly control the speed of passengers entering the paid area, close some entrance gates and temporary ticket booths, and arrange staff to intercept at the stairs, while maintaining the order of passengers waiting for the train on the platform.

[0112] The following are the recommendations for passenger flow organization decisions for passenger flow level 4: When the passenger flow at the entrance and exit channels is large, further increase may cause difficulty for passengers to enter and exit the station. It is recommended to adjust the functions of some entrances and exits of the station, intercept the passenger flow at the entrances and exits, and reopen the entrance channels after the congestion in the station is alleviated. In terms of strengthening guidance and information transmission, strengthen inspections and guidance of key areas such as station halls, escalators, and platforms to avoid passenger congestion and disputes and safety accidents; broadcast large passenger flow safety broadcasts throughout the station to remind passengers to pay attention to safety, and use the PIS system on the platform and at the entrance to remind passengers of the current driving intervals and the latest driving information.

[0113] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

[0114] Although the terms passenger flow prediction model, passenger flow level, etc. are used more frequently in this article, the possibility of using other terms is not excluded. These terms are used only to more conveniently describe and explain the essence of the present invention; interpreting them as any additional restrictions is contrary to the spirit of the present invention.

Claims

1. A passenger flow organization decision-making method based on station passenger flow prediction, characterized in that, it includes the following steps: S1 Conduct historical data collection. The historical data includes counting the daily inbound and outbound passenger flow, counting the date type of each day and the events that occurred on that day within a certain historical period, and counting the concourse and platform areas of each station in the network, the number of entrances and exits, and the passenger flow levels corresponding to the inbound and outbound passenger flow in different 10-minute intervals within a certain historical period; S2 Construct station passenger flow prediction models according to different situations respectively; On days with emergencies, the gain method is used to construct the model. For the inbound and outbound passenger flow of different stations under emergencies, it is obtained by multiplying the inbound and outbound passenger flow under the normal operation of the station by a gain coefficient; On the days of holding large-scale events, an SVM model is constructed. The optimization objective function of the SVM model is The passenger flow data points (X i , Y i ) of the large-scale event are fitted to a linear model , and a constant ε > 0 is defined; For ordinary holidays without emergencies and large-scale events, the ARIMA model is used to obtain the passenger flow time series data under holiday conditions; convert it into a stationary time series; respectively obtain its autocorrelation coefficient ACF and partial autocorrelation coefficient PACF for the stationary time series, and through the analysis of the autocorrelation diagram and partial autocorrelation diagram, obtain the order p and order q; according to q and p, obtain the ARMA model; For ordinary non-holidays without emergencies and large-scale events, the LSTM neural network is used to construct the model. The Sigmoid function and Tanh function are selected as the activation functions of the network hidden layer; the adam algorithm is used to optimize the weights of the network. The inbound volume and outbound volume of the most recent period are used as the input of the model, and the inbound volume and outbound volume of the prediction period are used as the output of the model, and the passenger flow data under the corresponding conditions collected is used to train the model; S3 Predict the station passenger flow according to the station passenger flow prediction model; S4 Classify the passenger flow levels of the predicted passenger flow, and use the KNN model to construct a model for judging the station passenger flow classification; select the inbound volume, outbound volume, number of entrances and exits, and the concourse and platform areas of the station as the characteristic values of the model, and the passenger flow level as the output of the model; calculate the distance between the points in the historical passenger flow level sample set and the current point; for each sample, select k closest neighbors to represent; the category with the highest frequency of occurrence among the first k points is used as the passenger flow level of the current point; S5 Give suggestions on passenger flow organization decision-making according to the passenger flow level.

2. The passenger flow organization decision-making method based on station passenger flow prediction according to claim 1, characterized in that, in step S2, station passenger flow prediction models are constructed according to different situations respectively. The different situations specifically include: days with emergencies, days with large-scale events, ordinary holidays without emergencies and large-scale events, and ordinary non-holidays without emergencies and large-scale events. The above different situations are arranged in order of priority from high to low.

3. The passenger flow organization decision-making method based on station passenger flow prediction according to claim 2, characterized in that, on the days with emergencies, the gain method is used to construct a short-term passenger flow prediction model for urban rail transit stations under emergency situations. For the inbound and outbound passenger flow of different stations under emergencies, it is obtained by multiplying the inbound and outbound passenger flow under the normal operation of the station by a gain coefficient.

4. A passenger flow organization decision-making method based on station passenger flow prediction according to claim 2, characterized in that, the days of holding large-scale events use SVM to construct a short-term passenger flow prediction model for urban rail transit stations under the condition of large-scale events, The optimization objective function of the SVM model is For the passenger flow data points (x i , y i ) of large-scale events, fitting them to a linear model and defining a constant ε > 0, the loss function metric of the SVM model is as follows:

5. A passenger flow organization decision-making method based on station passenger flow prediction according to claim 2, characterized in that, for ordinary holidays without emergencies and large-scale events, an ARIMA model is used to construct a short-term passenger flow prediction model for urban rail transit stations under holiday conditions. The specific steps of the process are as follows: S3.31 Obtain the passenger flow time series data under holiday conditions; S3.32 Plot the data to observe whether it is a stationary time series. If it is a non-stationary time series, perform d-order difference operation first to convert it into a stationary time series; d = 0, y t = Y t d = 1, y t = Y t -Y t-1 d = 2, y t = (Y t - Y t-1 ) - (Y t-1 - Y t-2 ) S3.33 Calculate the autocorrelation coefficient ACF and partial autocorrelation coefficient PACF of the stationary time series respectively. Through the analysis of the autocorrelation diagram and partial autocorrelation diagram, obtain the order p and order q; S3.34 According to q and p, obtain the ARMA model. The mathematical expression of the model is as follows: where, φ represents the coefficient of AR, and θ represents the coefficient of MA; S3.35 Conduct a white noise test on the obtained model. If the white noise test is not passed, update p and q; S3.36 Until the model that passes the white noise test is the optimal ARMA model.

6. A passenger flow organization decision-making method based on station passenger flow prediction according to claim 2, characterized in that, for ordinary non-holidays without emergencies and large-scale events, an LSTM neural network is used to construct a short-term passenger flow prediction model for urban rail transit stations under non-holiday conditions. The specific steps are as follows: S3.41 The neural network structure is an input layer, a hidden layer, and an output layer; S3.42 In the network training algorithm, select the Sigmoid function and Tanh function as the activation functions of the network hidden layer; use the adam algorithm to optimize the weights of the network; The formula of the Sigmoid function is as follows: The formula of the Tanh function is as follows: S3.43 Use the inbound volume and outbound volume in the recent period as the input of the model, and use the inbound volume and outbound volume in the prediction period as the output of the model, and train the model with the passenger flow data collected under the corresponding conditions.

7. A passenger flow organization decision-making method based on station passenger flow prediction according to claim 1, characterized in that, in step S4, a KNN model is used to construct a model for judging the classification of station passenger flow. The specific steps are as follows: S4.1 Select the inbound volume, outbound volume, number of entrances and exits, and the area of the station concourse and platform as the characteristic values of the model, and the passenger flow level as the output of the model; S4.2 Calculate the distance between the points in the historical passenger flow level sample set and the current point. The distance formula is as follows: S4.3 Sort in ascending order of distance; S4.4 Select the value of k, and select k nearest neighbors, that is, each sample selects k closest neighbors to represent; S4.5 Return the category with the highest frequency of occurrence among the first k points as the passenger flow level of the current point.

8. A passenger flow organization decision-making method based on station passenger flow prediction according to claim 7, characterized in that, The passenger flow levels are divided into four levels: Level 1 indicates normal inbound and outbound volumes; Level 2 indicates excessive outbound volume and normal inbound volume; Level 3 indicates excessive inbound volume and normal outbound volume; Level 4 indicates excessive inbound and outbound volumes.

9. A passenger flow organization decision-making method based on station passenger flow prediction according to claim 8, characterized in that the specific suggestions for passenger flow organization decision-making given according to the passenger flow level in step S5 include: The suggestions for passenger flow organization decision-making for Level 1 passenger flow are: The station adopts the normal inbound and outbound ticket checking mode, but strengthens guidance and patrols at escalators and inbound and outbound turnstiles to avoid congestion; The suggestions for passenger flow organization decision-making for Level 2 passenger flow are: When congestion occurs at the outbound turnstiles for outbound passenger flow, the station changes some inbound turnstiles from the inbound ticket checking mode to the outbound ticket checking mode, increases the number of outbound turnstiles. If the congestion situation fails to ease, the station opens the side door of the turnstile for passengers holding single-journey tickets to pass through without inspection, and the staff manually collects the single-journey tickets at the turnstile; The suggestions for passenger flow organization decision-making for Level 3 passenger flow are: The station operation management personnel set obstacles at various places inside the inbound entrance to guide the passenger flow to detour. When congestion forms at the inbound turnstiles and the passenger density in the platform or paid area of the concourse is not large, the station changes the mode of some outbound turnstiles, increases the number of inbound turnstiles, or sets the inbound turnstiles to the non-inspection mode, and at the same time opens the side door for passengers to enter the station free of charge and make up the ticket when leaving the station. When the number of passengers staying in the paid area of the concourse reaches a certain amount, slow down ticket sales, add isolation fences in the inbound entrance area, and the staff organize the passengers to queue up in an orderly manner to enter the station in a detour. When the passengers on the platform cannot board the train in time and cause passenger flow congestion, the station needs to control the speed of passengers getting off the platform, timely control the speed of passengers entering the paid area, close some inbound turnstiles and temporary ticket booths, and arrange staff to intercept at the staircase entrance, while maintaining good order for passengers waiting on the platform; The suggestions for passenger flow organization decision-making for Level 4 passenger flow are: When the passenger flow agglomeration volume at the entrance and exit channels is large and further increase will cause difficulties for passengers to enter and leave the station, it is recommended to adjust the functions of some entrances and exits of the station, intercept the inbound passenger flow at the entrance and exit, and reopen the inbound channel after the congestion phenomenon in the station is relieved. In terms of strengthening guidance and information dissemination, strengthen the patrol and guidance in the station concourse, escalators, and platform to avoid congestion and disputes among passengers and prevent safety accidents; Continuously broadcast the safety broadcast for large passenger flows throughout the station to remind passengers to pay attention to safety, and use the PIS system on the platform and at the inbound entrance to remind passengers of the current train operation interval and the latest train operation information.

Images