Airport security check passenger flow prediction method and system

Abstract

The invention discloses an airport security check passenger flow prediction method, and relates to the technical field of airport passenger flow prediction. An SBC algorithm is combined with a heuristic iterative algorithm to obtain the passenger queuing time, and the time of arriving at a security check area is obtained; adopting a hierarchical clustering method to obtain each type of flight time period; and adopting an integrated learning algorithm based on a decision-making tree to fit the time distribution discrete points of arrival of the passengers in the security check area in each type of flight time period, obtaining the arrival distribution of the passengers, and completing the prediction of the security check passenger flow. According to the method, the difference between the security check time and the arrival time of passengers is considered, the time of arriving at the security check area is calculated to predict the passenger flow, and an optimization method combining an SBC algorithm of a non-stationary queuing theory and a heuristic iterative algorithm is provided to solve the time of arriving at the security check area. Flight time periods and passenger arrival distribution are clustered and fitted by adopting a hierarchical clustering method and an integrated learning algorithm based on a decision tree, and the accuracy of airport security check passenger flow prediction is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of airport passenger flow prediction, and more specifically, to a method and system for predicting airport security passenger flow. Background Art

[0002] The arrival process of passengers in the airport security check area has certain aggregation and randomness, and its arrival flow is mainly affected by the departure time and passenger capacity of the flight. An unstable arrival process can easily lead to a waste of service resources. Therefore, managers need to reasonably allocate the security check channel resources. Insufficient allocation will increase the queuing time of passengers and affect passenger satisfaction; excessive allocation will cause idleness and reduce channel utilization. When optimizing the allocation of airport security check channel resources, it is usually necessary to rely on the prediction of security check passenger flow. In order to dynamically configure the service resources of the airport security check area, it is necessary to accurately predict the passenger flow in each time period according to the flight plan. However, the arrival time of passengers is historical data that has already occurred. The existing methods for predicting the flow of departing passengers are all based on the condition that the arrival time of passengers is known, or the security check time is directly used as the arrival time. However, in actual situations, the airport security check area only verifies the identity information and records the security check time during the security check process. At this time, the queuing process has ended and the arrival time of passengers cannot be obtained; if the security check time is directly used as the arrival time for prediction, the prediction result will be lagging and the accuracy will be reduced because the passenger arrival time is earlier than the security check time. There are usually uncertain factors in the airport ground service process, such as the randomness of passenger arrival intervals, the non-homogeneity of passenger arrival rates, etc. To this end, the present invention studies the passenger flow in the airport security inspection area within a unit time period (with 60 minutes as a unit time period), and takes the actual data of a large airport as an example to provide an airport security inspection passenger flow prediction method and system, in order to provide a scientific basis for the reasonable allocation and scheduling of service resources in the airport security inspection area. Summary of the invention

[0003] The purpose of the present invention is to overcome the defects of the prior art that the difference between the time when passengers arrive at the security check area and the security check time and the non-stationarity of the process of passengers arriving at the security check area are not taken into account, thereby reducing the accuracy of the integrated passenger flow prediction, and to provide an airport security check passenger flow prediction method and system.

[0004] To achieve the above object, the technical solution of the present invention is as follows:

[0005] A method for predicting airport security passenger flow, comprising the following steps:

[0006] According to the airport security check queue mode and considering the non-stationary process of passengers arriving at the security check area, a general distribution non-stationary queue model is established;

[0007] The performance metric approximation of the non-stationary queuing model is calculated by using a backlog backward-shifting stationary approximation method, and the calculation process is embedded in a heuristic iterative algorithm to iteratively obtain the passenger queuing time, and the passenger arrival time at the security check area is obtained according to the passenger queuing time;

[0008] Considering the arrival distribution characteristics of passengers on a single flight, the hierarchical clustering method is used to classify flight time slots and obtain the flight time slots of each category;

[0009] Using an integrated learning algorithm based on a decision tree to fit the discrete points of the time distribution of the passengers arriving at the security inspection area in the flight time periods of each category, to obtain the passenger arrival distribution in the flight time periods of each category;

[0010] Based on the passenger arrival distribution of each category of flight time, the airport security passenger flow forecast is completed.

[0011] Preferably, the establishment of a general distributed non-stationary queuing model is specifically as follows: for each service system, when passengers arrive at the system, if the service desk is idle, they can directly receive service. On the contrary, if the service desk is busy, the passengers need to queue in a queue and wait for service. The system has infinite waiting capacity, the arrival process obeys the general distribution, the average arrival rate is λ(t), and the service time also obeys the general distribution with an average service rate of μ(t). The service rule of the service desk is "first come, first served". According to the airport queuing method, a multi-G(t) / G(t) / 1 parallel queuing model is constructed. According to Kendall notation, the queuing model is a G(t) / G(t) / 1 / ∞ model.

[0012] Preferably, the specific method of calculating the performance of the non-stationary queuing model by using the backlog-shifted stationary approximation method is:

[0013] The time interval of the unit time period is divided into T time periods, and the arrival rate λ of the i-th time period is calculated using the backlog backshift stationary approximation method i and service rate μ i :

[0014]

[0015] Starting from the first period, the analysis is carried out with a steady G / G / 1 / 1 loss model, defining the manual backlog b for each period. i-1 for:

[0016]

[0017]

[0018] Among them, the artificial backlog b i-1represents the steady-state rate of operations leaving the G / G / 1 / 1 loss system due to congestion in the i-1 period; P i-1 (B) represents the probability of congestion in the i-1th period, which should be replaced by the Erlang loss probability;

[0019] The backlog is transferred to the next period, thus getting the next period Manual arrival rate:

[0020]

[0021] At the same time, the expected utilization rate of the system in the next period is obtained. i ):

[0022]

[0023] The steady G / G / 1 / 1 loss system model cannot be used to approximate the expected queue length or the number of passengers in the system, so a modified arrival rate is used to approximate the corresponding performance metric for the G / G / 1 waiting system. The modified arrival rate for the i-th time period is for:

[0024]

[0025] Using Modified Reach Combining the approximation method of the G / G / 1 waiting system, we can get the approximate value E(W) of the required performance metric:

[0026]

[0027] Where E(W) is the theoretical time for the passenger queue to end, ρ is the traffic intensity, and are the squared coefficients of variation of the input process distribution and the service process distribution, respectively.

[0028] Preferably, the specific steps of the heuristic iterative method embedded in the backlog backward-shifted stationary approximation method calculation process are:

[0029] S21: Inputting the security check information including the passenger security check time read by the machine;

[0030] S22: setting an initial queuing time and an error threshold, and setting an initial random queuing time for the service station in the non-stationary queuing model according to different intervals of the service rate. The higher the service rate, the shorter the set initial queuing time;

[0031] S23: System performance analysis, using the backlog shift stationary approximation method to analyze and calculate the approximate value of the performance metric of the non-stationary queuing model, and obtain the theoretical time for the passenger to end the queue, that is, the theoretical time for the passenger to start security inspection

[0032] S24: calculating the error between the theoretical time for the passenger to start security check and the actual security check time under the current queuing time, and comparing the error with the error threshold; when the error is not less than the error threshold, adjusting the initial queuing time according to the comparison result;

[0033] S25: iteratively executing steps S23-S24 according to the adjusted initial queuing time;

[0034] S26: When the error is less than the error threshold, the current queuing time is output as the passenger queuing time, and the time when the passenger arrives at the security inspection area is obtained according to the passenger queuing time.

[0035] Preferably, the specific manner of adjusting the initial queuing time according to the comparison result is: if the theoretical time for passengers to start security inspection obtained according to the non-stationary queuing model is earlier than the actual time, the value of the initial queuing time is increased; otherwise, the value of the initial queuing time is reduced.

[0036] Preferably, the hierarchical clustering method is an agglomerative hierarchical clustering AGNES algorithm.

[0037] Preferably, the sample set used by the clustering algorithm is:

[0038] D = {x1, x2, ..., x m}

[0039] The sample set D contains m unlabeled samples, where x i ={x i1 , x i2 , …, x in} is an n-dimensional feature vector, representing the probability distribution of passengers arriving at the security check area for flights with a departure time in the i-th time period.

[0040] Preferably, the decision tree-based ensemble learning algorithm is a gradient boosting decision tree (GBDT) algorithm.

[0041] Preferably, the gradient boosting decision tree GBDT algorithm uses a coefficient of determination R 2 To measure the goodness of fit, the calculation formula is:

[0042]

[0043] Among them, SSR stands for the regression sum of squares, which is the sum of squares of the differences between the model prediction value and the mean of the dependent variable; SST stands for the sum of squares of deviations, which is the sum of squares of the differences between the dependent variable and its mean; y is the actual value, is the average value, is the predicted value.

[0044] To achieve the purpose of the invention, the present invention further provides an airport security inspection passenger flow prediction system, an airport security inspection passenger flow prediction method using the above technical solution, and the system comprises:

[0045] Modeling module, which is used to consider the non-stationary process of passengers arriving at the security check area and establish a non-stationary queue model with general distribution;

[0046] An algorithm iteration module, used for calculating the performance of the non-stationary queuing model by using a backlog backward-shifting stationary approximation method, and embedding the calculation process into a heuristic iteration algorithm, iteratively obtaining the passenger queuing time, and obtaining the time when the passenger arrives at the security inspection area according to the passenger queuing time;

[0047] The clustering module is used to consider the arrival distribution characteristics of passengers on a single flight and classify the flight time slots using a hierarchical clustering method to obtain the flight time slots for each category;

[0048] A fitting module, used to fit the discrete points of the time distribution of the passengers arriving at the security inspection area in the flight periods of each category by using an integrated learning algorithm based on a decision tree, so as to obtain the passenger arrival distribution in the flight periods of each category;

[0049] The prediction result output module is used to use the passenger arrival distribution of each category of flight time period as the prediction result of airport security passenger flow.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] The present invention considers the difference between the security check time and the arrival time of airport security check passengers, estimates the arrival time of passengers in the security check area, and predicts the security check passenger flow. An iterative optimization method combining the system performance index estimation method of non-stationary queuing theory with a heuristic iterative algorithm is proposed to solve the arrival time of passengers in the security check area. A hierarchical clustering method and an ensemble learning algorithm based on a decision tree are used to cluster and fit flight time periods and passenger arrival distributions, thereby effectively improving the accuracy of airport security check passenger flow prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A schematic diagram of the steps of a method for predicting airport security passenger flow provided in Example 1 of the present application;

[0053] Figure 2 A schematic diagram of the airport security check area security check process provided in Example 1 of the present application;

[0054] Figure 3 A schematic diagram of a multi-G(t) / G(t) / 1 parallel queuing model provided in Example 1 of the present application;

[0055] Figure 4A flowchart of a heuristic iterative algorithm for the embedded SBC approximation method provided in Example 1 of the present application;

[0056] Figure 5 A schematic diagram of the basic principle of the SBC approximation method provided in Example 1 of the present application;

[0057] Figure 6 A flow chart of the AGNES algorithm provided in Example 1 of the present application;

[0058] Figure 7 Schematic diagram of the basic principle of the GBDT algorithm provided in Example 1 of the present application. DETAILED DESCRIPTION

[0059] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0060] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0061] Example 1

[0062] See also Figure 1 Embodiment 1 of the present application provides a method for predicting airport security passenger flow, comprising the following steps:

[0063] S1: Based on the airport security queuing pattern and considering the non-stationary process of passengers arriving at the security check area, a general distribution non-stationary queuing model is established.

[0064] See also Figure 2 , Figure 2 The figure is a schematic diagram of the security check process in the airport security check area. During the airport security check, when the time when the device reads the ID information is used as the security check time, the passenger has completed the queuing process, and the time when the passenger arrives at the security check area is earlier than the security check time. However, there is often no way to accurately record the time when each passenger arrives at the security check area. This article infers the arrival time of the passenger by calculating the passenger queuing time. Generally, passengers queue up for security check when they arrive at the security check area, and rarely stay in the security check area. Without considering the passenger's stay, the time when a passenger arrives at the security check area can be determined as the time when the security check starts minus the time spent queuing for security check. The time of the passenger's security check is read by the machine and is known data, so you only need to calculate the passenger's queuing time to get the passenger's arrival time.

[0065] The number of passengers arriving in each period varies greatly and is non-stationary. The arrival process of passengers is random (the arrival interval of passengers is uncertain). The number of service channels and service rates change dynamically and are non-stationary. It is necessary to establish a general distribution non-stationary queue model. According to the airport queue mode, a multi-G(t) / G(t) / 1 parallel queue model is constructed, such as Figure 3 As shown. For each service system, when passengers arrive at the system, if the service desk is idle, they can receive service directly. On the contrary, if the service desk is busy, passengers need to wait in a queue for service. The system has infinite waiting capacity. The arrival process follows a general distribution with an average arrival rate of λ(t). The service time also follows a general distribution with an average service rate of μ(t). The service rule of the service desk is "first come, first served". According to Kendall notation, the queue model is the G(t) / G(t) / 1 / ∞ model.

[0066] S2: The performance of the non-stationary queuing model is calculated using a stationary backlog-carryover (SBC) approximation method, and the calculation process is embedded in a heuristic iterative algorithm to iteratively obtain the passenger queuing time, and the passenger arrival time at the security check area is obtained based on the passenger queuing time.

[0067] As mentioned in step S1, without considering the passenger's stay, the time when a passenger arrives at the security check area can be determined as the time when the security check starts minus the time spent queuing for the security check. Therefore, as long as the passenger's queuing time is obtained, the time when the passenger arrives at the security check area can be inferred. The queuing system first needs the passenger arrival rate to calculate the queuing time. Please refer to Figure 4 , Figure 4 This is a flowchart of a heuristic iterative algorithm embedded in the SBC approximation method provided in this embodiment 1. This embodiment 1 proposes a method for approximating the passenger queuing time by a heuristic iterative algorithm embedded in the SBC approximation method. The specific steps of the algorithm are as follows:

[0068] S21: Setting the initial queuing time: According to different intervals of the service rate, an initial random queuing time is set for the service station. Generally, the higher the service rate, the shorter the set initial queuing time.

[0069] S22: System performance analysis: Calculate the time it takes for passengers to arrive at the security check area (the time a passenger arrives at the security check area is determined as the time when the security check starts minus the time spent queuing for the security check). Use the SBC approximation method to analyze and calculate the approximate value of the performance metric of the non-stationary queuing model, thereby obtaining the time when the passenger ends the queue, that is, the time when the security check starts.

[0070] S23: Compare and adjust: Compare the security check start time in step S22 with the actual security check time, and calculate the error between the two. This error value is used to adjust the initial queue time value in step S21. If the security check time calculated by the queuing model is earlier than the actual time, the initial queue time value is increased; otherwise, the initial queue time value is reduced.

[0071] S24: Repeat iteration: Repeat steps S22 and S23 using new initial values, and perform system performance analysis and comparison iterations again.

[0072] S25: Error threshold determination: When the error calculated in step S23 is less than a certain set threshold, the iteration ends. The initial queue time at this time is considered to be a relatively accurate queue time and can be used as output.

[0073] The SBC algorithm approximate solution method for the G(t) / G(t) / 1 queuing time:

[0074] See also Figure 5 , Figure 5 The schematic diagram of the basic principle of the SBC approximation method is shown in FIG. 1. In the SBC approximation method, the entire time interval (60 minutes in this embodiment 1) is divided into T time periods, and the original arrival rate and service rate λ(t) and μ(t) are respectively replaced by the piecewise constant function λ i and μ i Substitution, λ i is the arrival rate of the ith period, μ i is the service rate of the ith period, which is given by equations (1) and (2):

[0075]

[0076] Phase 1: Starting from the first period, the analysis is conducted using the stable G / G / 1 / 1 loss model. In order to consider the connection between periods, the manual backlog b of each period is defined i-1 , represents the steady-state rate of operations leaving the G / G / 1 / 1 loss system due to congestion in the i-1th period, and is given by equation (3). i-1 (B) represents the probability of congestion in the i-1th time period, and this embodiment 1 uses the Erlang loss probability instead, which is given by equation (4). These backlogs are transferred to the next time period, thereby obtaining the next time period The manual arrival rate is given by formula (5). At the same time, the expected utilization rate of the system in the next period is obtained: E(U i ), given by formula (6):

[0077]

[0078] Phase II: The loss model proposed in Phase I cannot be used to approximate the expected queue length or number of passengers in the system, so a modified arrival rate is needed to approximate the corresponding performance metric for the G / G / 1 waiting system. The modified arrival rate for the i-th time period is Given by formula (7):

[0079]

[0080] Using Modified Reach Combined with the approximation method of the G / G / 1 waiting system, the approximate value of the required performance metric is obtained. An approximation method for the G / G / 1 queue model proposed by W.

[0081] W proposed the approximate values ​​of the performance metrics of the G / G / 1 queue model. The expected queue time E(W) is given by equations (8)-(9):

[0082]

[0083] Where ρ is the traffic intensity, and are the squared coefficients of variation of the input process distribution and the service process distribution, respectively.

[0084] S3: Considering the arrival distribution characteristics of passengers on a single flight, the hierarchical clustering method is used to classify flight time slots to obtain flight time slots for each category.

[0085] S4: Using a decision tree-based ensemble learning algorithm to fit the discrete points of the time distribution of the passengers arriving at the security check area in each category of flight time slots, to obtain the passenger arrival distribution in each category of flight time slots.

[0086] S5: According to the passenger arrival distribution of each category of flight time period, the prediction of airport security passenger flow is completed.

[0087] Example 2

[0088] This embodiment 2 is based on embodiment 1, and further describes step S3 and step S4 in embodiment 1, as follows:

[0089] S3: Considering the arrival distribution characteristics of passengers on a single flight, the hierarchical clustering method is used to classify flight time slots to obtain flight time slots for each category.

[0090] In previous studies, the classification of flight departure time is usually based on subjective judgment, dividing it into morning, afternoon and evening time periods. For example, a prior art simplifies flight departure time into two main categories when classifying flight departure time: morning peak period (9:00 to 12:00) and evening peak period (15:00 to 17:00). However, the arrival distribution of flights at adjacent departure times also has a large difference. Therefore, this embodiment 2 is based on a machine learning clustering algorithm to more objectively classify flight departure time, i.e., flight time period. Clustering algorithm is a class of unsupervised learning algorithms, which is mainly used to group a set of data into multiple classes or clusters measured by similarity. The goal of the clustering algorithm is to make the data points in the same cluster as similar as possible, while the data points in different clusters are as different as possible. The AGNES (Agglomerative Nesting) algorithm used in this embodiment 2 is a hierarchical clustering algorithm, which is an agglomerative hierarchical clustering method. In agglomerative hierarchical clustering, each sample is initially regarded as a separate cluster, and then the algorithm gradually merges the most similar clusters together until all samples belong to the same cluster or meet a certain termination condition.

[0091] In the AGNES algorithm, the sample set D = {x1, x2, ..., x m} contains m unlabeled samples, where x i ={x i11 , x i2 , …, x in} is an n-dimensional feature vector representing the probability distribution of passenger arrivals for flights with departure times in the i-th time period. Figure 6 , Figure 6 The following is a schematic diagram of the AGNES algorithm flow of Example 2. The basic flow of the algorithm is as follows:

[0092] S31: Initialization: Each sample x i Considered as a separate cluster C i .

[0093] S32: Distance calculation: Each cluster is a sample set, and the distance between sets is calculated using a specific link criterion. i With C j The distance between samples is usually calculated by Euclidean distance, which is given by formula (13). min d max d avg When calculating, the AGNES algorithm is called the "single link", "full link" or "even link" algorithm respectively.

[0094]

[0095]

[0096] S33: Merge clusters: select the clusters with the closest distance to merge and form a new cluster.

[0097] S34: Update distance: recalculate the distance between the merged cluster and other clusters.

[0098] S35: Repeat steps (3) and (4) until the stop condition is met: the above process is repeated until the preset number of clusters is reached.

[0099] S4: Using a decision tree-based ensemble learning algorithm to fit the discrete points of the time distribution of the passengers arriving at the security check area in each category of flight time period, to obtain the passenger arrival distribution in each category of flight time period.

[0100] In order to fit the distribution of passenger arrivals of multiple flights in each category, this paper adopts the Gradient Boosting Decision Tree (GBDT) algorithm to fit the discrete points of the distribution of passenger arrivals of multiple flights, and the fitting results are used as the distribution of passenger arrivals of the current category. The GBDT algorithm is an ensemble learning algorithm based on decision trees, which iterates the new learner through gradient descent. The GBDT algorithm uses the decision tree as the basic learner in the gradient boosting machine, first constructs a decision tree, and then continuously constructs another tree on the output residual of the existing model and the actual sample, iterates in sequence, and finally uses the results of all decision trees as the output. Its basic principle is as follows Figure 7 In each iteration, the GBDT algorithm tries to minimize the residuals of the model in the previous iteration and builds a new decision tree to fit these residuals.

[0101] Goodness of fit refers to the degree of fit of the regression line to the observed values. The statistic that measures goodness of fit is the coefficient of determination (also known as the coefficient of determination) R 2 . R 2 The maximum value is 1. 2 The closer the value of is to 1, the better the regression line fits the observed value; on the contrary, R 2 The smaller the value of is, the worse the regression line fits the observed value. The calculation method of goodness of fit is given by formula (14):

[0102]

[0103] Among them, SSR stands for the regression sum of squares, which is the sum of squares of the differences between the model prediction value and the mean of the dependent variable; SST stands for the sum of squares of deviations, which is the sum of squares of the differences between the dependent variable and its mean; y is the actual value, is the average value, is the predicted value.

[0104] The other steps of this embodiment 2 are the same as those of embodiment 1 and will not be repeated here.

[0105] Example 3

[0106] This embodiment 3 is based on embodiment 1 and embodiment 2, and applies an airport security inspection passenger flow prediction method in the above embodiment 1 and embodiment 2 to provide an airport security inspection passenger flow prediction system, including:

[0107] Modeling module, which is used to consider the non-stationary process of passengers arriving at the security check area and establish a non-stationary queue model with general distribution;

[0108] The algorithm iteration module is used to calculate the performance of the non-stationary queue model by using the backlog backward stationary approximation method, and embed the calculation process into the heuristic iteration algorithm to iteratively obtain the passenger queue time, and obtain the time when the passenger arrives at the security check area based on the passenger queue time;

[0109] The clustering module is used to consider the arrival distribution characteristics of passengers on a single flight and classify the flight time slots using a hierarchical clustering method to obtain the flight time slots for each category;

[0110] A fitting module, for fitting the discrete points of the time distribution of passengers arriving at the security check area in each category of flight period by using an integrated learning algorithm based on a decision tree, so as to obtain the passenger arrival distribution in each category of flight period;

[0111] The prediction result output module is used to use the passenger arrival distribution of each category of flight time as the prediction result of airport security passenger flow.

[0112] The other steps of this embodiment 3 are the same as those of embodiment 1 and will not be repeated here.

[0113] In summary, the present invention considers the difference between the security check time and the arrival time of airport security check passengers, estimates the time when passengers arrive at the security check area, and predicts the security check passenger flow. It proposes an iterative optimization method that combines the system performance indicator estimation method of non-stationary queuing theory with a heuristic iterative algorithm to solve the time when passengers arrive at the security check area. It also uses a hierarchical clustering method and an ensemble learning algorithm based on a decision tree to cluster and fit flight time periods and passenger arrival distributions, which effectively improves the accuracy of airport security check passenger flow prediction and provides a scientific basis for the reasonable allocation and scheduling of service resources in airport security check areas.

[0114] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the embodiments here. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the claims of the present invention.

Claims

1. A method for predicting airport security passenger flow, characterized in that: The following steps are involved: According to the airport security check queue mode and considering the non-stationary process of passengers arriving at the security check area, a general distribution non-stationary queue model is established; The performance of the non-stationary queue model is calculated by using a backlog backward-shifting stationary approximation method, and the calculation process is embedded in a heuristic iterative algorithm to iteratively obtain the passenger queue time, and the passenger arrival time at the security check area is obtained according to the passenger queue time; Considering the arrival distribution characteristics of passengers on a single flight, the hierarchical clustering method is used to classify flight time slots and obtain the flight time slots of each category; Using an integrated learning algorithm based on a decision tree to fit the discrete points of the time distribution of the passengers arriving at the security inspection area in the flight time periods of each category, to obtain the passenger arrival distribution in the flight time periods of each category; Based on the passenger arrival distribution of each category of flight time, the airport security passenger flow forecast is completed.

2. The method for predicting airport security passenger flow according to claim 1, characterized in that: The above-mentioned establishment of a general distribution non-stationary queuing model is as follows: for each service system, when passengers arrive at the system, if the service desk is idle, they can directly receive service. On the contrary, if the service desk is busy, passengers need to queue in a queue and wait for service. The system has infinite waiting capacity, and the arrival process obeys the general distribution. The average arrival rate is λ(t). The service time also obeys the general distribution with an average service rate of μ(t). The service rule of the service desk is "first come, first served". According to the airport queuing method, a multi-G(t) / G(t) / 1 parallel queuing model is constructed. According to Kendall notation, the queuing model is a G(t) / G(t) / 1 / ∞ model.

3. The method for predicting airport security passenger flow according to claim 2, characterized in that: The specific method of calculating the performance of the non-stationary queuing model using the backlog-shifted stationary approximation method is: The time interval of the unit time period is divided into T time periods, and the arrival rate λ of the i-th time period is calculated using the backlog backshift stationary approximation method i and service rate μ i : Starting from the first period, the analysis is carried out with a steady G / G / 1 / 1 loss model, defining the manual backlog b for each period. i-1 for: Among them, the artificial backlog b i-1 represents the steady-state rate of operations leaving the G / G / 1 / 1 loss system due to congestion in the i-1 period; P i-1 (B) represents the probability of congestion in the i-1th period, which is replaced by the Erlang loss probability; The backlog is transferred to the next period, thus getting the next period Manual arrival rate: At the same time, the expected utilization rate of the system in the next period is obtained. i ): The steady G / G / 1 / 1 loss system model cannot be used to approximate the expected queue length or the number of passengers in the system, so a modified arrival rate is used to approximate the corresponding performance metric for the G / G / 1 waiting system. The modified arrival rate for the i-th time period is for: Using Modified Reach Combining the approximation method of the G / G / 1 waiting system, we can get the approximate value E(W) of the required performance metric: Where E(W) is the theoretical time for passengers to queue, ρ is the traffic intensity, and are the squared coefficients of variation of the input process distribution and the service process distribution, respectively.

4. The method for predicting airport security passenger flow according to claim 3, characterized in that: The specific steps of the heuristic iterative method after embedding the backlog backward shift stationary approximation method calculation process are: S21: Inputting the security check information including the passenger security check time read by the machine; S22: setting an initial queuing time and an error threshold, and setting an initial random queuing time for the service station in the non-stationary queuing model according to different intervals of the service rate. The higher the service rate, the shorter the set initial queuing time; S23: System performance analysis: Based on the non-stationary queue model, the backlog-shifted stationary approximation method is used to analyze the performance of the calculation system to obtain the theoretical time when the passenger queue ends, that is, the theoretical time when the passenger starts security inspection. S24: calculating the error between the theoretical time for the passenger to start security check and the actual security check time under the current queuing time, and comparing the error with the error threshold; when the error is not less than the error threshold, adjusting the initial queuing time according to the comparison result; S25: iteratively executing steps S23-S24 according to the adjusted initial queuing time; S26: When the error is less than the error threshold, the current queuing time is output as the passenger queuing time, and the time when the passenger arrives at the security inspection area is obtained according to the passenger queuing time.

5. The method for predicting airport security passenger flow according to claim 4, characterized in that: The specific method of adjusting the initial queuing time according to the comparison result is: if the theoretical time for passengers to start security inspection obtained according to the non-stationary queuing model is earlier than the actual time, the value of the initial queuing time is increased; otherwise, the value of the initial queuing time is reduced.

6. The method for predicting airport security passenger flow according to claim 1, characterized in that: The hierarchical clustering method is the agglomerative hierarchical clustering AGNES algorithm.

7. The method for predicting airport security passenger flow according to claim 6, characterized in that: The sample set used by the clustering algorithm is: D={x1,x2,…,x m } The sample set D contains m unlabeled samples, where x i ={x i1 , x i2 , …, x in } is an n-dimensional feature vector, representing the probability distribution of passengers arriving at the security check area for flights with a departure time in the i-th time period.

8. The method for predicting airport security passenger flow according to claim 1, characterized in that: The decision tree-based ensemble learning algorithm is a gradient boosted decision tree (GBDT) algorithm.

9. The method for predicting airport security passenger flow according to claim 8, characterized in that: The gradient boosting decision tree GBDT algorithm uses the determination coefficient R 2 To measure the goodness of fit, the calculation formula is: Among them, SSR stands for the regression sum of squares, which is the sum of squares of the differences between the model prediction value and the mean of the dependent variable; SST stands for the sum of squares of deviations, which is the sum of squares of the differences between the dependent variable and its mean; y is the actual value, is the average value, is the predicted value.

10. An airport security check passenger flow prediction system, using an airport security check passenger flow prediction method as described in any one of claims 1 to 9, characterized in that: The system comprises: Modeling module, which is used to consider the non-stationary process of passengers arriving at the security check area and establish a non-stationary queue model with general distribution; An algorithm iteration module, used for calculating the performance of the non-stationary queuing model by using a backlog backward-shifting stationary approximation method, and embedding the calculation process into a heuristic iteration algorithm, iteratively obtaining the passenger queuing time, and obtaining the time when the passenger arrives at the security inspection area according to the passenger queuing time; The clustering module is used to consider the arrival distribution characteristics of passengers on a single flight and classify the flight time slots using a hierarchical clustering method to obtain the flight time slots for each category; A fitting module, used to fit the discrete points of the time distribution of the passengers arriving at the security inspection area in the flight periods of each category by using an integrated learning algorithm based on a decision tree, so as to obtain the passenger arrival distribution in the flight periods of each category; The prediction result output module is used to use the passenger arrival distribution of each category of flight time period as the prediction result of airport security passenger flow.

Images