Scheduling method for airport passenger security check channel resource optimization

Abstract

The invention discloses a scheduling method for airport passenger security check channel resource optimization. The method comprises the following steps: constructing a security check channel model; defining security check channel efficiency indexes; calculating a security check channel passing rate under each scheduling scheme; a security check resource multi-objective optimization model is constructed, multi-index balance optimization is used, and resources are reasonably allocated through a dynamic optimization-oriented resource allocation optimization algorithm; constructing a target function; and setting constraint conditions: solving the target function by using a deep reinforcement learning algorithm, and outputting an optimal opening scheme of each security check channel and a security check personnel scheduling scheme in each time period. Compared with the prior art, the method can comprehensively reflect the operation condition of each link of the security check system, improves the precision and practicability of dynamic resource scheduling, can dynamically respond to the fluctuation of passenger flow in each time period to realize flexible resource scheduling, supports multi-objective optimization, can adaptively output an optimal decision through the dynamic characteristics of a learning environment, and is suitable for popularization and application. The method has high practical value and wide application prospect.

Description

Technical Field

[0001] The present invention relates to the technical field of airport terminal operation management, and in particular to a scheduling method for optimizing airport passenger security inspection channel resources. Background Art

[0002] Data from the International Air Transport Association (IATA) indicates that global air passenger volume will exceed 5 billion for the first time in 2025. At the same time, with the acceleration of global economic integration and the booming civil aviation industry, domestic aviation networks are continuously improving, and more and more people are choosing convenient and efficient air travel. This has led to a continuous increase in airport flight takeoffs and landings and passenger throughput. As a super hub with an annual passenger throughput of over 45 million, Beijing Daxing International Airport needs to provide higher operational support capabilities to accommodate this growing passenger volume. In airport operations management, the passenger experience is directly related to airport service quality and passenger satisfaction, and is a key indicator of airport operational support. However, cumbersome and inefficient security screening processes and large passenger flows can lead to a large number of passengers being screened in a short period of time, resulting in congestion in security checkpoints and long wait times, seriously impacting the passenger experience. A study by the U.S. Transportation Security Administration (TSA) shows that wait times exceeding 20 minutes can lead to a 35% decrease in passenger satisfaction. Therefore, scientifically evaluating the inspection efficiency of security checkpoints and optimizing the allocation of security checkpoint resources have become the key to improving passenger travel experience and airport operation efficiency. The common problems currently faced by airport terminal security checkpoint systems are as follows:

[0003] (1) Fluctuations in passenger flow

[0004] Passenger flow fluctuates due to factors such as time and flight schedules. Peak hours may lead to congestion in security check channels, while off-peak hours may result in idle resources.

[0005] (2) Security inspection resource allocation needs to be optimized

[0006] Currently, the allocation of security resources, such as security inspector manpower allocation and channel opening and closing, relies primarily on empirical judgment, lacking a systematic, data-driven optimization strategy. This approach can lead to resource shortages and long queues during peak periods, while low resource utilization during off-peak periods, resulting in wasted operating costs. Therefore, the introduction of scientific optimization algorithms, through data analysis and dynamic adjustments, is crucial to improve the efficiency of security resource allocation, ensure smooth passenger flow, and reduce operating costs. This approach is of great practical significance.

[0007] (3) Balance between efficiency and operating costs

[0008] In airport security management, improving security efficiency and controlling operating costs are often in conflict. Excessively increasing security lanes and staffing can shorten passenger wait times and improve transit efficiency, but this can also lead to higher operating costs. Conversely, over-saving resources can lead to long passenger queues, impacting the travel experience and even the overall efficiency of the airport. Therefore, scientific optimization models and dynamic allocation strategies are needed to ensure efficient passenger transit while rationally allocating security resources, achieving an optimal balance between security efficiency and operating costs.

[0009] Therefore, to address these pain points in airport terminal security channel management, a systematic resource optimization and allocation algorithm is urgently needed. This algorithm can minimize the number of security channels open and the number of security personnel deployed while meeting passenger flow, thereby improving security resource utilization and reducing operating costs. The method for optimizing airport passenger security channel resources proposed in this paper can effectively address these issues. Summary of the Invention

[0010] The purpose of the present invention is to provide a scheduling method for optimizing airport passenger security inspection channel resources. Through scientific optimization models and dynamic allocation strategies, while ensuring the rapid passage of passengers, the rational allocation of security inspection resources is achieved, so that the security inspection efficiency and operating costs reach the optimal balance point. In the management of airport terminal security inspection channels, the number of security inspection channels opened and the configuration of security inspection personnel can be minimized, the utilization rate of security inspection resources can be improved and the operating costs can be reduced.

[0011] In order to achieve the above-mentioned purpose, the technical solution adopted by the present invention is: a scheduling method for optimizing airport passenger security inspection channel resources, the method steps are as follows:

[0012] Step 1: Model the passenger security inspection process in a multi-dimensional, multi-stage, integrated, and serial manner. A security inspection channel model is constructed that demonstrates the dynamic coupling relationship between each link. This security inspection channel model, based on factors influencing differences in passenger type, equipment type and its processing capacity, and the number of security inspectors and their operational efficiency, calculates and designs the channel flow rate model for channel k at different security inspection stages i under security inspector scheduling plan a, as well as the overall flow rate of the security inspection channel.

[0013] Step 2: To provide a scientific basis for optimizing the resource allocation of the security inspection system within the channel, improve airport security inspection efficiency and enhance the passenger experience, the security inspection channel efficiency indicator, the average passenger inspection time τ, is defined. ka and the work intensity of security inspectors ka ;

[0014] Step 3: Set up a set of security inspection channels, a set of security inspection posts, a set of statistical time periods, a set of passenger flow in those time periods, and a set of security inspector scheduling plans. Define the number of security inspectors required for each step of the security inspection process under different scheduling plans, and calculate the security inspection channel traffic rate under each scheduling plan using the security inspection channel model.

[0015] Step 4: Construct a multi-objective optimization model for security inspection resources and use multi-index balance optimization. Through the multi-objective optimization algorithm, the optimal balance point can be found between different objectives, security inspection resources can be reasonably allocated, and the joint optimization of dynamic channel configuration and static personnel scheduling can be achieved. The dynamic resource scheduling of security inspection channels and the static scheduling of security inspection personnel can be solved to achieve efficient operation of the overall security inspection system. Through the dynamic optimization-oriented resource allocation optimization algorithm, under the conditions of meeting dynamic passenger flow, the number of security inspector positions, and the workload limit of security inspectors, the average passenger inspection time, security inspection labor cost, and channel personnel configuration changes are minimized as optimization goals. The security inspection efficiency and human resource utilization are balanced and resources are reasonably allocated. Using the dynamic optimization-oriented resource allocation optimization algorithm, different terminal operation strategies can be implemented to meet different management needs by dynamically adjusting the labor cost target δ, the weight coefficient ψ of the personnel configuration stability target, and the weight coefficient ω of the passenger satisfaction index (i.e., the average passenger inspection time).

[0016] Step 5: Introduce binary integer variables to represent the selection of security inspector scheduling plans, forming a binary linear programming problem. Therefore, the constructed objective function is:

[0017]

[0018] where δ, ψ and They are the weight coefficients of security inspection manpower operation cost, change in channel personnel configuration in adjacent time periods, and average passenger inspection time, c a and N ka are the operating costs of the security inspection channel when using scheduling plan a and the total number of security inspectors required for channel k under scheduling plan a, is a decision variable, indicating whether channel k adopts the scheduling plan a in time period t, τ ka is the average passenger inspection time at channel k under scheduling plan a;

[0019] Step 6: Set constraints:

[0020] Step 7: Perform resource optimization design on the security inspection channel model, use the deep reinforcement learning algorithm to solve the objective function, and output the optimal opening plan for each security inspection channel in each time period and the security inspection personnel scheduling plan.

[0021] The scheduling method of the present invention supports coping with dynamic fluctuations in passenger flow at different time periods. It can reasonably allocate existing security inspection resources according to passenger flow conditions, dynamically adjust the opening status of channels and the configuration of security inspection personnel at different time periods, and at the same time, while considering security inspection efficiency, also pay attention to the utilization of security inspectors' human resources, balance the two, and achieve the purpose of scientific and reasonable optimization.

[0022] Preferably, in step 1, the specific construction method of the security inspection channel model is as follows:

[0023] First, the security inspection process is divided into multiple serial stages, including identity verification, passenger personal inspection, and luggage inspection key subsystems. Then, multiple key links in the entire chain of identity verification, passenger inspection, and luggage inspection are comprehensively modeled.

[0024] Then, based on the different influencing factors of passenger type, equipment type and its processing capacity, the number of security inspectors and their operating efficiency, a channel flow rate model for channel k at different security inspection stages i under security inspector scheduling plan a is established as follows:

[0025]

[0026] Where i represents different security check stages, including identity verification (id), passenger personal inspection (pass), and luggage inspection (bag), p zk represents the proportion of each passenger type Z in the total passenger flow in channel k, is the processing capacity of the security inspection equipment of channel k at different security inspection links i (unit: passenger / minute), is the impact factor of passenger type Z in channel k on security check link i, is the efficiency function of the security personnel, that is, in the security inspection link i of channel k, as the number of security personnel under the security inspection scheduling plan a at this time The increase of , the personnel efficiency function of the channel flow rate improvement;

[0027] Finally, the overall flow rate of the security inspection channel can be expressed as

[0028] Its security inspection process comprehensively models multiple key links across the entire chain, including identity verification, passenger screening, and baggage screening. It expands traditional single-link optimization into a multi-stage cascade model, introducing dynamic coupling relationships between each link. This modeling approach is more aligned with actual security inspection scenarios, fully reflecting the operational status of the security inspection system and providing a more accurate theoretical foundation for the analysis and optimization of security inspection channel resources. Its channel model comprehensively considers factors such as security inspection equipment throughput and personnel operational efficiency. It is a full-link, multi-dimensional, dynamically coupled joint model. It systematically conducts deep coupling modeling and optimization of each link in the baggage screening process, making the model more adaptable to complex scenarios.

[0029] Preferably, in step 1, the flow rate of the security inspection channel is modeled in combination according to the number of security inspectors configured in each link of the security inspection process to form a channel service capacity indicator under the scheduling plan.

[0030] As a preferred method, in step 1, the baggage inspection process model is refined to analyze the sub-processes of the baggage process, such as forward transmission, X-ray machine inspection (which is divided into single-view X-ray machine and dual-view X-ray machine), and secondary package inspection. The influencing factors of each process, such as equipment physical parameters, equipment performance, personnel operation speed, baggage quantity and type, are clarified, and the baggage model is improved to make it highly adaptable to complex scenarios.

[0031] Preferably, in step 2,

[0032] The average passenger inspection time τ ka , the expression is

[0033] This indicator reflects the average time it takes for passengers to enter the security check channel and complete the security check. The smaller the indicator value, the higher the security check efficiency and the shorter the waiting time for passengers.

[0034] The workload of security inspectors (number of passengers inspected per security inspector) ka , when channel k adopts scheduling scheme a, the indicator expression is

[0035] where λ k is the actual number of passengers passing security checks during the statistical period (unit: person), T op is the channel opening time during the statistical period, N ka It is the number of security inspectors. This indicator reflects the number of passengers checked by each security inspector per unit time during the statistical period. It can reflect the workload of the security inspectors and intuitively reflect the work intensity of the security inspectors.

[0036] Preferably, in step 6, the following constraints are set:

[0037] C1:

[0038] where, θ k is the emergency reserve coefficient for the passenger flow on channel k, which can be dynamically adjusted according to historical data and real-time conditions to cope with sudden changes in passenger flow. This constraint means that in each time period, the theoretical throughput of all open channels should be able to handle the passenger flow in that time period and meet the passenger volume demand, s is the theoretical throughput of channel k under the security inspector scheduling plan a; ka

[0039] C2: This condition is a constraint on the scheduling plan of channel selection, meaning that only one scheduling plan can be adopted in the current time period;

[0040] C3: represents the constraint of binary decision variables, with only the values 0 or 1;

[0041] C4: ensures that for each time period t and each post j, the total number of security inspectors assigned to that post does not exceed the total number of available security inspectors at that post;

[0042] C5: To avoid waste of human resources, ensure that the minimum number of passengers inspected per security inspector is not less than the minimum value To prevent the security inspectors from being overloaded and affecting the security inspection quality, it is required that the maximum number of passengers inspected per security inspector does not exceed the maximum workload

[0043] Preferably, in step 7, the method for resource optimization design and deep learning of the security inspection channel model is as follows

[0044] First, model the optimization objective problem as an optimal stationary decision-making Markov decision (MDP) process. The Markov decision process is represented by MDP = <S, A, P, R, Π>. The four elements S, A, P, R of the MDP model are state, action, state transition probability set, and reward function. The reward function R(s t , a t ) represents the reward that the system can obtain when the system is in state s t and takes action a t at a certain time period. Π is the policy, which represents the basis for the system to take actions, that is, actions are selected according to the policy;

[0045] Then, construct a DQN network, build a deep neural network to approximate the Q function. The input of this network is the state vector s t , and the output is the Q value of each action;

[0046] ​Finally, the DQN algorithm is used for deep learning. The DQN algorithm can handle complex scenarios such as resource allocation during security inspections very well, and make optimal decisions by learning the dynamic characteristics of the environment. It uses historical data for learning and optimizes strategies through continuous trial and error, thereby finding reasonable resource allocation plans under different passenger flows and other conditions.

[0047] As an advantage, the state set S is represented by s t =(Λ t , N a , S a , M j , C1, C2, C3, C4, C5) means,

[0048] Among them, s t ∈S represents the system state at time period t, where represents the passenger flow of channel k in time period t, N a ={N 1a , N 2a ,…,N Ka}, a∈A, represents the total number of security inspectors required for channel k under scheduling plan a, S a ={S 1a , S 2a ,…,S Ka}, a∈A, represents the theoretical throughput of channel k under the security inspector scheduling scheme a, M j ={M1, M2, ..., M J}, j∈J, is the total number of security inspectors available at position j, C1, C2, C3, C4, C5 respectively indicate whether the constraints C1-C5 are satisfied in the current state (satisfied is marked as 1, not satisfied is marked as 0);

[0049] The action set A is the action taken based on the current state in time period t, which can be expressed as The action consists of two parts: the first is the open state of each channel k in each time period t (Channel opening is marked as 1, and closing is marked as 0); second, each channel selects the scheduling plan a The size of the action space is 2 |K| ×2 |K|×|A| , where K is the channel set and A is the scheduling plan set;

[0050] The state transition probability P is solved using a model-free learning algorithm;

[0051] The reward function R, since the optimization goal is to find the minimum value, the reward function R is:

[0052]

[0053] The strategy π is represented by π∈π, which means a stable decision. Under the strategy π, the state is s t Action a will be taken t ,

[0054] When the strategy is uncertain, the probability of choosing each action is π(a t |s t )=P[A=a t |S=s t ].

[0055] Preferably, the DQN network has two sub-networks with the same structure, namely the target network and the prediction network. During the training process, the two sub-networks will iteratively update the network parameters according to the difference in the results between the two (i.e., the loss value). The target network is used to realize the actual value of Q, and the prediction network is used to obtain the estimated value of Q. After a certain number of iterations, the parameters of the prediction network will be updated to the target network. According to the Q value in the prediction network, the optimal strategy in each state is to select the action that maximizes the Q value. That is, after selecting the opening state of each channel k in each time period t and the security inspector scheduling plan for the channel, the environment feedbacks the instantaneous reward R (s t ,a t ), current state s t Transition to the next state s t+1 .

[0056] Compared with the prior art, the advantages of the present invention are:

[0057] (1) Through multi-dimensional, multi-stage serial comprehensive modeling of the entire security inspection process and dynamic coupling analysis of each link, the operation status of each link of the security inspection system can be fully reflected, and the accuracy and practicality of dynamic resource scheduling can be improved.

[0058] (2) The scheduling method of the present invention has the ability of adaptive adjustment, which can dynamically respond to the fluctuation of passenger flow in each time period to achieve flexible resource scheduling, increase manpower deployment to ensure service efficiency during peak passenger flow periods, and intelligently reduce redundant configurations to reduce operating costs during off-peak periods.

[0059] (3) The present invention supports multi-objective optimization and can simultaneously meet the indicators of minimizing labor costs, maintaining stability in security inspector scheduling configuration changes, and ensuring the sustainability of security inspection human resources. It also uses a deep reinforcement learning algorithm to automatically generate resource allocation plans and can adaptively output optimal decisions through the dynamic characteristics of the learning environment, significantly improving production efficiency compared to manual scheduling.

[0060] (4) Furthermore, the system model of the present invention is scalable, with parameterized configurations of job grouping, scheduling rules, channel attributes, etc., allowing for expansion and adaptation based on the actual job structure or airport scale. Therefore, the present invention has high practical value and broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is an overall flow chart of the method of the present invention;

[0062] Figure 2 It is a system model framework diagram of the present invention;

[0063] Figure 3 This is a schematic diagram of the security inspection channel post process of the present invention;

[0064] Figure 4 This is a schematic diagram of the DQN algorithm of the present invention;

[0065] Figure 5 This is a flow chart of the DQN algorithm of the present invention. DETAILED DESCRIPTION

[0066] The multi-index balanced optimization and dynamic optimization-guided security inspection resource allocation optimization algorithm proposed in this paper incorporates multiple goals into the optimization scope, including reducing security inspection labor costs, stabilizing channel staffing, meeting passenger flow requirements, and balancing security inspector workloads. The specific objectives are as follows:

[0067] (1) Minimizing the use of security inspection resources

[0068] Reduce the labor cost of security inspections and rationally optimize the number of security inspection channels and the number of security inspection personnel.

[0069] (2) Stable channel personnel configuration

[0070] This ensures efficient coordination of all aspects of the process, including identity verification, passenger screening, and baggage inspection, to avoid issues such as decreased efficiency or operational inefficiencies caused by frequent staffing changes. Furthermore, stable staffing ensures relatively even screening times for passengers, minimizing the impact on their travel experience. It also helps optimize security officer scheduling, reduces ineffective staffing, and improves human resource utilization, thereby reducing overall operating costs.

[0071] (3) Responding to Fluctuations in Passenger Flow

[0072] By dynamically adjusting the security inspector schedules and the number of open channels, we ensure sufficient resources during peak periods and reasonable control of manpower and equipment use during off-peak periods to cope with fluctuating passenger flows. Through multi-objective optimization algorithms, we can find the optimal balance between different objectives, reasonably allocate security inspection resources, and achieve joint optimization of dynamic channel configuration and static personnel scheduling, solving the dynamic resource scheduling of security inspection channels and the static scheduling of security inspectors, so as to achieve efficient operation of the overall security inspection system.

[0073] (IV) Sustainable development of security inspection human resources

[0074] By accurately quantifying the average number of passengers checked per security inspector in each security inspection channel and at each time period (i.e., the security inspector's workload), the fluctuations in the security inspectors' workload can be limited to a reasonable threshold range, avoiding the situation where some security inspectors have excessive workloads that affect the quality of security inspections, or some personnel have excessively low workloads that cause waste of human resources. This will achieve a balanced distribution of the security inspectors' workload, improve their job satisfaction and career sustainability, and ultimately achieve the coordinated optimization of security inspection efficiency, service quality, and human resource utilization, ensuring the efficient, stable, and safe operation of airport security inspections.

[0075] The present invention will be further described below. A scheduling method for optimizing airport passenger security inspection channel resources is described in detail below. Figures 1 to 5 ,

[0076] (1) Model assumptions

[0077] To facilitate the establishment of the model and the design of the computer simulation algorithm, the present invention makes the following assumptions:

[0078] (1) Passenger type assumption. Different types of passengers require different amounts of time to complete the security check process. It is assumed that passengers of the same type behave consistently during the security check process. For example, the degree of cooperation of ordinary passengers during security checks and the time it takes to check their luggage are similar, so that a unified analysis and processing can be performed on passenger groups.

[0079] (2) Security inspection process assumptions. Assume that the security inspection process is fixed, including identity verification, passenger inspection, luggage inspection, baggage inspection, etc., and the order of each link is fixed. The impact of special circumstances (such as temporary adjustment of the security inspection process due to emergencies) on the security inspection process is not considered.

[0080] The service time of each security check link obeys a specific probability distribution, such as exponential distribution.

[0081] (3) Security inspection equipment assumptions. The security inspection equipment in each link operates normally, the conveyor belt runs at a constant speed, and the luggage tray is used normally. There will be no failures or performance fluctuations during operation.

[0082] (4) Security inspector assumption: The security inspector's professional level is stable and remains stable throughout the work process. It will not fluctuate significantly due to factors such as working hours and fatigue level.

[0083] (5) Baggage assumption: The number of bags each passenger carries is fixed, and the security inspection process of one piece of baggage will not cause additional interference to the security inspection process of other passengers.

[0084] (2) Objective function setting

[0085] During the operation of the airport terminal, it is necessary to ensure the rapid passage of passengers while achieving a reasonable allocation of security inspection resources, so that the security inspection efficiency and operating costs reach the optimal balance point. Reducing the labor cost of security inspectors has always been one of the goals of airport operation and development. Stabilizing the allocation of channel personnel can avoid problems such as reduced efficiency or unskilled operation due to frequent changes in personnel deployment, and help optimize the scheduling mechanism of security inspectors, reduce ineffective personnel scheduling, and improve the utilization rate of human resources, thereby reducing overall operating costs. At the same time, limiting the fluctuation of the work intensity of security inspectors within a reasonable threshold range can achieve a balanced distribution of the workload of security inspectors, promote the sustainable development of human resources, and ensure the efficient, stable and safe operation of airport security inspection work. Therefore, the objective function setting is shown in formula (1)

[0086]

[0087] where δ, ψ and They are the weight coefficients of security inspection manpower operation cost, change in channel personnel configuration in adjacent time periods, and average passenger inspection time, c a and N ka are the operating costs of the security inspection channel when using scheduling plan a and the total number of security inspectors required for channel k under scheduling plan a, is a decision variable, indicating whether channel k adopts scheduling plan a in time period t. ka is the average passenger inspection time for channel k under scheduling plan a. T is the statistical time period, K is the set of all available security inspection channels in the airport security system, and A is the set of security personnel scheduling plans.

[0088] (3) Constraints

[0089] (1) In each time period, the theoretical throughput of all open channels must be able to handle the passenger flow in that time period and meet the passenger volume demand, while also being able to accommodate sudden changes in passenger flow.

[0090]

[0091] (2) Constraints on the scheduling plan for channel selection. Only one scheduling plan can be adopted for a channel during the current time period.

[0092]

[0093] (3) It is a constraint for binary decision variables, indicating whether scheduling plan a is adopted for channel k during period t. Therefore, only the values 0 or 1 can be selected.

[0094]

[0095] (4) Since the total number of security inspectors is fixed, the total number of security inspectors assigned to each post j during each time period t cannot exceed the total number of available security inspectors for that post.

[0096]

[0097] (5) To avoid waste of human resources and prevent security inspectors from being overburdened, which may affect the security inspection quality, it is required that the work intensity of security inspectors (the number of passengers inspected per capita) be restricted within a certain range.

[0098]

[0099] (IV) Design of the security inspection channel resource optimization algorithm based on DQN

[0100] Model the above optimization objective problem as a Markov decision (MDP) process of optimal stationary decision. The Markov decision process is represented by MDP = <S, A, P, R, Π>. The four elements S, A, P, and R of the MDP model are state, action, state transition probability set, and reward function respectively. The reward function R(s t , a t ) represents the reward that the system can obtain when the system is in state s t and takes action a t . Π is the policy, which represents the basis for the system to take actions, that is, actions are selected according to the policy. Therefore, it is a mapping from state to action.

[0101] (1) State set

[0102] Use s t ∈ S to represent the system state at time period t, where represents the passenger flow of channel k at time period t, N a = {N 1a , N 2a , …, N Ka}, a ∈ A, represents the total number of security inspectors required for channel k under scheduling plan a, S a = {S1a , S 2a ,…,S Ka}, a∈A, represents the theoretical throughput of channel k under the security inspector scheduling scheme a, M j ={M1, M2, ..., M I}, j∈J, is the total number of security inspectors available at position j, C1, C2, C3, C4, and C5 respectively indicate whether the constraints C1-C5 are satisfied in the current state (satisfied is marked as 1, and not satisfied is marked as 0).

[0103] (2) Action Set

[0104] The action taken based on the current state at time period t can be expressed as The action consists of two parts: the first is the open state of each channel k in each time period t (Channel opening is marked as 1, and closing is marked as 0); second, each channel selects the scheduling plan a The size of the action space is 2 |K| ×2 |K|×|A| , where K is the channel set and A is the scheduling plan set.

[0105] (3) State transition probability

[0106] The state transition profile of this problem is unknown, so it can be solved using a model-free learning algorithm (such as the Q-learning algorithm).

[0107] (4) Reward Function

[0108] Since the optimization goal is to find the minimum value, the reward function is:

[0109]

[0110] (5) Strategy

[0111] Use π∈Π to represent a stable decision. Under this strategy π, the state is s t Action a will be taken t , when the strategy is uncertain, the probability of choosing each action is π(a t |s t )=P[A=a t |s=s t ].

[0112] (6) DQN Network

[0113] Build a deep neural network to approximate the Q function, the input of which is the state vector s t , the output is the Q value of each action.

[0114] (7) Network structure

[0115] The network has two sub-networks with the same structure, namely the target network and the prediction network. During the training process, the two sub-networks will iteratively update the network parameters according to the difference between the results (i.e., the loss value). The target network is used to realize the actual value of Q, and the prediction network is used to obtain the estimated value of Q. After a certain number of iterations, the parameters of the prediction network will be updated to the target network. According to the Q value in the prediction network, the optimal strategy in each state is to select the action that maximizes the Q value. That is, after selecting the open state of each channel k in each time period t and the security inspector scheduling plan for the channel, the environment feedbacks the instantaneous reward R (s t , a t ), current state s t Transition to the next state s t+1 .

[0116] (8) Pseudocode of security inspection channel resource optimization algorithm based on DQN

[0117]

[0118]

[0119] The above is a detailed introduction to a scheduling method for optimizing airport passenger security channel resources provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Changes and improvements to the present invention will be possible without exceeding the concept and scope specified in the appended claims. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A scheduling method for optimizing airport passenger security inspection channel resources, characterized by: The steps are as follows: Step 1: Model the passenger security inspection process in a multi-dimensional, multi-stage, integrated, and serial manner. A security inspection channel model is constructed that demonstrates the dynamic coupling relationship between each link. This security inspection channel model, based on factors influencing differences in passenger type, equipment type and its processing capacity, and the number of security inspectors and their operational efficiency, calculates and designs the channel flow rate model for channel k at different security inspection stages i under security inspector scheduling plan a, as well as the overall flow rate of the security inspection channel. Step 2: Define the security inspection channel efficiency indicator: average passenger inspection time τ ka and security inspector work intensity Γ ka ; Step 3: Set up a set of security inspection channels, a set of security inspection posts, a set of statistical time periods, a set of passenger flow in those time periods, and a set of security inspector scheduling plans. Define the number of security inspectors required for each step of the security inspection process under different scheduling plans, and calculate the security inspection channel traffic rate under each scheduling plan using the security inspection channel model. Step 4: Build a multi-objective optimization model for security inspection resources. Using multi-index balanced optimization and a dynamic optimization-guided resource allocation optimization algorithm, the model aims to minimize average passenger inspection time, inspection labor costs, and channel staffing changes, while meeting dynamic passenger flow, inspection officer capacity limits, and inspector workload limits. This model balances inspection efficiency and human resource utilization, ensuring a reasonable allocation of resources. Step 5, construct the objective function: where δ, ψ, and ω are the weight coefficients of security inspection manpower operation cost, personnel configuration change in adjacent time periods, and average passenger inspection time, respectively. a and N ka are the operating costs of the security inspection channel when using scheduling plan a and the total number of security inspectors required for channel k under scheduling plan a, is a decision variable, indicating whether channel k adopts the scheduling plan a in time period t, τ ka is the average passenger inspection time at channel k under scheduling plan a; Step 6: Set constraints: Step 7: Perform resource optimization design on the security inspection channel model, use the deep reinforcement learning algorithm to solve the objective function, and output the optimal opening plan for each security inspection channel in each time period and the security inspection personnel scheduling plan.

2. The scheduling method for optimizing airport passenger security inspection channel resources according to claim 1 is characterized by: In step 1, the specific construction method of the security inspection channel model is as follows: First, the security inspection process is divided into multiple serial stages, including identity verification, passenger personal inspection, and luggage inspection key subsystems. Then, multiple key links in the entire chain of identity verification, passenger inspection, and luggage inspection are comprehensively modeled. Then, based on the different influencing factors of passenger type, equipment type and its processing capacity, the number of security inspectors and their operating efficiency, a channel flow rate model for channel k at different security inspection stages i under security inspector scheduling plan a is established as follows: Where i represents different security check stages, including identity verification (id), passenger personal inspection (pass), and luggage inspection (bag), p zk represents the proportion of each passenger type Z in the total passenger flow in channel k, is the processing capacity of the security inspection equipment of channel k at different security inspection links i (unit: passenger / minute), is the impact factor of passenger type Z in channel k on security check link i, is the efficiency function of the security personnel, that is, in the security inspection link i of channel k, the number of security personnel under the security inspection scheduling plan a at this time is The increase of , the personnel efficiency function of the channel flow rate improvement; Finally, the overall flow rate of the security inspection channel can be expressed as 3. The scheduling method for optimizing airport passenger security inspection channel resources according to claim 1 is characterized by: In step 2, the flow rate of the security inspection channel is combined and modeled according to the number of security inspectors configured in each link of the security inspection process to form a channel service capacity indicator under the scheduling plan.

4. The scheduling method for optimizing airport passenger security inspection channel resources according to claim 1 is characterized by: In step 1, the baggage inspection process model is refined, and sub-processes such as forward transmission, X-ray inspection, and secondary unpacking inspection are analyzed. The influencing factors of each process, such as equipment physical parameters, equipment performance, personnel operation speed, baggage quantity and type, are identified to improve the baggage model.

5. The scheduling method for optimizing airport passenger security inspection channel resources according to claim 1 is characterized by: In step 2, The average passenger inspection time τ ka , the expression is The work intensity of the security inspector (the number of passengers inspected per security inspector) Γ ka , when channel k adopts scheduling scheme a, the indicator expression is where λ k is the actual number of passengers passing security checks during the statistical period (unit: person), T op is the channel opening time during the statistical period, N ka is the number of security inspectors.

6. The scheduling method for optimizing airport passenger security inspection channel resources according to claim 1 is characterized by: In step 6, set the following constraints: C1： Among them, θ k is the passenger flow on channel k The emergency reserve coefficient can be adjusted dynamically based on historical data and real-time conditions to cope with sudden changes in passenger flow. ka is the theoretical throughput of channel k under security inspector scheduling scheme a; C2: This condition is a constraint on the scheduling scheme for channel selection; C3: It represents the constraints of binary decision variables, which can only have the value 0 or 1; C4: Ensure that in each time period t and each position j, the total number of security inspectors assigned to that position cannot exceed the total number of security inspectors available for that position; C5: To avoid wasting human resources, ensure that the number of passengers checked per security inspector is not less than the minimum In order to prevent the security inspectors from being overburdened with work and affecting the quality of security inspections, the number of passengers inspected per security inspector should not exceed the maximum workload.

7. The scheduling method for optimizing airport passenger security inspection channel resources according to claim 1 is characterized by: In step 7, the method for resource optimization design and deep learning of the security inspection channel model is as follows: First, the optimization problem is modeled as a Markov decision process (MDP) with optimal stationary decision. The Markov decision process is represented by MDP=<S,A,P,R,Π> Indicates that the four elements S, A, P, and R of the MDP model are state, action, state transition probability set, and reward function, respectively. The reward function R(s t , a t ) indicates that during the period when the system is in state s t and take action a t , the reward that the system can obtain, Π is the strategy, which represents the basis for the system to take actions; Then, construct a DQN network and build a deep neural network to approximate the Q function. The input of the network is the state vector s t , the output is the Q value of each action; Finally, the DQN algorithm is used for deep learning, historical data is used for learning, and the strategy is optimized through continuous trial and error to find a reasonable resource allocation plan under different passenger flows and other conditions.

8. The scheduling method for optimizing airport passenger security inspection channel resources according to claim 7 is characterized by: The state set S is used express, Among them, s t ∈S represents the system state at time period t, where represents the passenger flow of channel k during time period t, represents the total number of security inspectors required for channel k under scheduling plan a, s a ={S 1a , S 2a ,…,S Ka }, a∈A, represents the theoretical throughput of channel k under the security inspector scheduling scheme a, M j ={M1, M2, ..., M J }, j∈J, is the total number of security inspectors available at position j, c1, c2, c3, c4, c5 respectively indicate whether the constraints C1-C5 are satisfied in the current state (satisfied is marked as 1, not satisfied is marked as 0); The action set A is the action taken based on the current state in time period t, which can be expressed as The action consists of two parts: the first is the open state of each channel k in each time period t (Channel opening is marked as 1, and closing is marked as 0); second, each channel selects the scheduling plan a The size of the action space is 2 |K| ×2 |K|×|A| , where K is the channel set and A is the scheduling plan set; The state transition probability P is solved using a model-free learning algorithm; The reward function R is: The strategy π is represented by π∈π, which means a stable decision. Under the strategy π, the state is s t Action a will be taken t , when the strategy is uncertain, the probability of choosing each action is π(a r |s t )=P[A=a t |S=sx].

9. The scheduling method for optimizing airport passenger security inspection channel resources according to claim 7, characterized in that: The DQN network has two sub-networks with the same structure, namely the target network and the prediction network. During the training process, the two sub-networks will iteratively update the network parameters according to the difference between the results. The target network is used to realize the actual value of Q, and the prediction network is used to obtain the estimated value of Q. After a certain number of iterations, the parameters of the prediction network will be updated to the target network. According to the Q value in the prediction network, the optimal strategy in each state is to select the action that maximizes the Q value. That is, after selecting the opening state of each channel k in each time period t and the security inspector scheduling plan for the channel, the environment feedbacks the instantaneous reward R(s t , a t ), current state s t Transition to the next state s t+1 .

Images