Scheduling method, device and equipment of flight ground support vehicle and storage medium

Abstract

The invention provides a flight ground support vehicle scheduling method, apparatus and device, and a storage medium. The method comprises the steps of obtaining ground support service information in a to-be-scheduled airport operation environment; based on the ground support service information and a pre-constructed mixed integer linear programming model, determining a multi-target mixed integer linear programming model for scheduling of the current airport operation environment to be scheduled; constructing a task heterogeneous graph based on the multi-target mixed integer linear programming model, and performing feature extraction based on the task heterogeneous graph to obtain an embedded feature matrix corresponding to the operating environment of the airport to be scheduled; the embedded feature matrix is used for representing the global state of the operating environment of the airport to be scheduled; and inputting the embedded feature matrix into a pre-trained strategy model to enable the strategy model to output a target scheduling scheme. In this way, cooperative scheduling and global resource optimal allocation can be realized under complex and high-uncertainty conditions, and then the accuracy of scheduling scheme formulation is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of airport transportation information processing, in particular to a flight ground support vehicle scheduling method and device, equipment and a storage medium. BACKGROUND

[0002] In order to ensure the safety and efficiency of civil aviation transportation system, airport ground support service is necessary, which includes passenger service, resource allocation, cargo transportation, supply support, maintenance and other key links. With the continuous growth of air transportation demand, the airport needs to schedule support vehicles to ensure normal operation and passenger experience.

[0003] In related technologies, the scheduling scheme for support vehicles can be determined according to vehicle path planning or according to certain rules or corresponding scheduling algorithms. However, in the scheduling process, multiple business types and scheduling requirements are designed, and there are multiple constraint conditions. In the scheduling process, there may be resource conflicts or environmental disturbances and other sudden situations, resulting in low accuracy of the scheduling scheme. SUMMARY

[0004] Therefore, the purpose of the present application is to provide a flight ground support vehicle scheduling method, device, equipment and storage medium. According to the support service information of the to-be-scheduled airport operating environment and the mixed integer linear programming model, a multi-objective mixed integer linear programming model is determined, and the multi-objective mixed integer linear programming model is converted into a task heterogeneous graph for feature extraction to obtain an embedding feature vector representing the global state of the to-be-scheduled airport operating environment. The embedding feature vector is input into a policy model to obtain a corresponding target scheduling scheme, achieving collaborative scheduling and global resource optimal allocation under complex and high uncertainty conditions, thereby improving the accuracy of the scheduling scheme.

[0005] In a first aspect, an embodiment of the present application provides a flight ground support vehicle scheduling method, which comprises: Obtaining ground support service information in a to-be-scheduled airport operating environment; Based on the ground support service information and the pre-constructed mixed integer linear programming model, a multi-objective mixed integer linear programming model for scheduling the current to-be-scheduled airport operating environment is determined; Based on the multi-objective mixed integer linear programming model, a task heterogeneous graph is constructed, and feature extraction is performed based on the task heterogeneous graph to obtain an embedding feature matrix corresponding to the to-be-scheduled airport operating environment; the embedding feature matrix is used to represent the global state of the to-be-scheduled airport operating environment; The embedding feature matrix is input into a pre-trained policy model to make the policy model output a target scheduling scheme.

[0006] In a possible implementation, the mixed integer linear programming model is constructed by the following steps: determine a target function according to a pre-acquired flight set, a task set, a staff or vehicle set, and at least one target scenario; wherein the target function is a minimum time value calculated based on a weighted sum of a predicted total service time, a total idle time, and a predicted total delay time of flights; determine at least one constraint condition based on flight plan information, task timing information, resource capacity allocation condition information, space conflict information, and safety demand information; construct the mixed integer linear programming model based on the target function and the at least one constraint condition.

[0007] In a possible implementation, the task heterogeneous graph is constructed based on the multi-objective mixed integer linear programming model, including: determine at least two nodes based on flight information, task information, vehicle information, and personnel information in the multi-objective mixed integer linear programming model; determine an association relationship between each two nodes based on a task dependency relationship, a resource feasibility relationship, a space-time constraint relationship, and a priority relationship in the multi-objective mixed integer linear programming model; construct the task heterogeneous graph based on the at least two nodes and the association relationship between each two nodes; wherein an edge between each two nodes in the task heterogeneous graph represents the association relationship between the two nodes.

[0008] In a possible implementation, the feature extraction is performed based on the task heterogeneous graph to obtain an embedding feature matrix corresponding to the to-be-scheduled airport operating environment, including: input the task heterogeneous graph into a pre-trained graph convolutional neural network, so that a multi-layer convolutional neural network of the graph convolutional neural network aggregates local neighborhood information of each node in the task heterogeneous graph and forms a high-dimensional vector containing global structure semantics, and output the embedding feature matrix.

[0009] In a possible implementation, the strategy model is trained by the following steps: define a reinforcement learning environment of a Markov process based on node information in the embedding feature matrix, a task execution state, and sample ground support service information; wherein the reinforcement learning environment includes a state space, an action space, a transition function, and a reward function; The plurality of sample ground support service information is input into a Markov process for multiple rounds of training based on the reinforcement learning environment, and a parameter optimization is performed on the Markov process by using an experience replay mechanism and a target network update strategy. When the average cumulative reward determined based on the reward function converges in continuous multiple rounds of training and the fluctuation is lower than a preset fluctuation threshold, it is determined that the Markov process training is completed, the parameters of the current Markov process are saved, and the policy model is obtained.

[0010] In a possible implementation, the reward function is determined by the following steps: The reward function is determined based on the cumulative flight delay time, the driving distance of the ground support vehicle, the resource utilization rate index, and the pre-set penalty information for violating the priority or safety constraint.

[0011] In a possible implementation, after the embedding feature matrix is input into the pre-trained policy model so that the policy model outputs the target scheduling scheme, the scheduling method further includes: In response to a change in the ground support service information, the Markov process is retrained or locally updated according to the changed ground support service information, an updated policy model is obtained, and an updated scheduling scheme is output according to the updated policy model.

[0012] In a second aspect, the embodiments of the present application also provide a scheduling device for a flight ground support vehicle, and the scheduling device includes: An information acquisition module is configured to acquire ground support service information in a to-be-scheduled airport operating environment; A linear programming model determination module is configured to determine, based on the ground support service information and a pre-constructed mixed integer linear programming model, a multi-objective mixed integer linear programming model for scheduling of the current to-be-scheduled airport operating environment; A feature extraction module is configured to construct a task heterogeneous graph based on the multi-objective mixed integer linear programming model, and perform feature extraction based on the task heterogeneous graph to obtain an embedding feature matrix corresponding to the to-be-scheduled airport operating environment; the embedding feature matrix is used to represent the global state of the to-be-scheduled airport operating environment; A scheduling scheme determination module is configured to input the embedding feature matrix into a pre-trained policy model so that the policy model outputs a target scheduling scheme.

[0013] In a third aspect, the embodiments of the present application further provide an electronic device, comprising a processor, a storage medium and a bus, the storage medium stores machine readable instructions executable by the processor, when the electronic device is running, the processor communicates with the storage medium through the bus, and the processor executes the machine readable instructions to perform the steps of the scheduling method of the flight ground support vehicle according to any one of the first aspect.

[0014] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, the computer readable storage medium stores a computer program, when the computer program is run by a processor, the steps of the scheduling method of the flight ground support vehicle according to any one of the first aspect are executed.

[0015] The scheduling method, device, equipment and storage medium of the flight ground support vehicle provided by the embodiments of the present application obtain ground support service information in a to-be-scheduled airport operating environment; based on the ground support service information and a pre-constructed mixed integer linear programming model, a multi-objective mixed integer linear programming model is determined for the current to-be-scheduled airport operating environment; based on the multi-objective mixed integer linear programming model, a task heterogeneous graph is constructed, and feature extraction is performed based on the task heterogeneous graph to obtain an embedding feature matrix corresponding to the to-be-scheduled airport operating environment; the embedding feature matrix is used to represent the global state of the to-be-scheduled airport operating environment; and the embedding feature matrix is input into a pre-trained policy model to make the policy model output a target scheduling scheme. In this way, according to the support service information of the to-be-scheduled airport operating environment and the mixed integer linear programming model, a multi-objective mixed integer linear programming model is determined, the multi-objective mixed integer linear programming model is converted into a task heterogeneous graph for feature extraction, an embedding feature vector representing the global state of the to-be-scheduled airport operating environment is obtained, and the embedding feature vector is input into a policy model to obtain a corresponding target scheduling scheme, so that collaborative scheduling and global resource optimal allocation are realized under complex and high uncertainty conditions, and the accuracy of the scheduling scheme is improved.

[0016] In order to make the above objectives, characteristics and advantages of the present application more apparent, clear and easy to understand, the following preferred embodiments are described in detail below, and the accompanying drawings are described as follows. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can also be obtained without creative labor on the basis of these drawings.

[0018] Figure 1A flowchart of a flight ground support vehicle scheduling method provided by an embodiment of the present application is shown in FIG. 1. Figure 2 A structural schematic diagram of a mixed integer linear programming model provided by an embodiment of the present application is shown in FIG. 2. Figure 3 A structural schematic diagram of a task heterogeneous graph provided by an embodiment of the present application is shown in FIG. 3. Figure 4 A process schematic diagram of a policy model training provided by an embodiment of the present application is shown in FIG. 4. Figure 5 A structural schematic diagram of a flight ground support vehicle scheduling device provided by an embodiment of the present application is shown in FIG. 5. Figure 6 A structural schematic diagram of an electronic device provided by an embodiment of the present application is shown in FIG. 6. DETAILED DESCRIPTION

[0019] To make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described below in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. The components of the embodiments of the present application described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments of the present application, every other embodiment obtained by a person skilled in the art without creative work belongs to the scope of protection of the present application.

[0020] First, the application scenarios applicable to the present application are introduced. The present application can be applied to the field of airport transportation information processing technology.

[0021] In order to ensure the safety and efficiency of civil aviation transportation system, airport ground support service is necessary, which includes passenger service, resource allocation, cargo transportation, supply support, maintenance and other key links. With the continuous growth of air transportation demand, the airport needs to schedule support vehicles to ensure the normal operation of the airport and the passenger experience.

[0022] In the related art, the following problems need to be solved: ①The ground support operation scheduling task is dynamic and complex. The ground support operation involves multiple types of services, and there may be priority constraints between services. The task execution order is affected by the previous task and the flight state, showing conditional constraints and uncertainty. A simple static vehicle path problem cannot accurately describe the scheduling task mathematically. ②It is difficult to extract heterogeneous data features. The data of multiple subjects such as vehicles, flights, slots, and operation personnel are diverse in source and have significant differences in attributes. Traditional feature engineering cannot capture the topological relationship and spatiotemporal coupling features. ③The robustness of the decision strategy is insufficient. The existing scheduling algorithm based on rules or heuristics cannot realize adaptive optimization of scheduling. In the face of dynamic disturbance or resource conflict, the existing method is difficult to update in real time, resulting in low accuracy of scheduling scheme.

[0023] Based on this, the embodiment of the application provides a scheduling method for flight ground support vehicles to improve the accuracy of scheduling scheme.

[0024] Please refer to Figure 1 , Figure 1 The flowchart of the scheduling method for flight ground support vehicles provided by the embodiment of the application. As shown in Figure 1 The scheduling method for flight ground support vehicles provided by the embodiment of the application includes: S101, obtaining ground support service information in a to-be-scheduled airport operating environment.

[0025] S102, determining a multi-objective mixed integer linear programming model for scheduling the current to-be-scheduled airport operating environment based on the ground support service information and a pre-constructed mixed integer linear programming model.

[0026] S103, constructing a task heterogeneous graph based on the multi-objective mixed integer linear programming model, and extracting features based on the task heterogeneous graph to obtain an embedding feature matrix corresponding to the to-be-scheduled airport operating environment; the embedding feature matrix is used to represent the global state of the to-be-scheduled airport operating environment.

[0027] S104, inputting the embedding feature matrix into a pre-trained strategy model to make the strategy model output a target scheduling scheme.

[0028] The scheduling method of the flight ground support vehicle provided in the embodiments of the present application determines a multi-objective mixed integer linear programming model according to the support service information of the to-be-scheduled airport operating environment and a mixed integer linear programming model, converts the multi-objective mixed integer linear programming model into a task heterogeneous graph for feature extraction, obtains an embedding feature vector representing the global state of the to-be-scheduled airport operating environment, inputs the embedding feature vector into a policy model, and obtains a corresponding target scheduling scheme, thereby achieving collaborative scheduling and global resource optimal allocation under complex and high uncertainty conditions, and further improving the accuracy of the scheduling scheme.

[0029] The exemplary steps of the embodiments of the present application are described below: In S101, ground support service information in a to-be-scheduled airport operating environment is acquired.

[0030] In a possible implementation, in order to ensure the safety and efficiency of civil aviation transportation system, airport ground support service is necessary, which includes many key links such as passenger service, resource allocation, cargo and mail transportation, supply support, and maintenance inspection. With the continuous growth of aviation transportation demand, the airport needs to schedule support vehicles to ensure normal operation and passenger experience.

[0031] In the related art, the scheduling scheme for support vehicles can be determined according to vehicle path planning or according to a predetermined rule or a corresponding scheduling algorithm. However, in the scheduling process, multiple business types and scheduling requirements are designed, multiple constraint conditions exist, and in the scheduling process, sudden situations such as resource conflicts or environmental disturbances may occur, resulting in low accuracy of the scheduling scheme.

[0032] Therefore, in the embodiments of the present application, a multi-objective mixed integer linear programming model is determined according to the support service information of the to-be-scheduled airport operating environment and a mixed integer linear programming model, the multi-objective mixed integer linear programming model is converted into a task heterogeneous graph for feature extraction, an embedding feature vector representing the global state of the to-be-scheduled airport operating environment is obtained, the embedding feature vector is input into a policy model, and a corresponding target scheduling scheme is obtained, thereby achieving collaborative scheduling and global resource optimal allocation under complex and high uncertainty conditions, and further improving the accuracy of the scheduling scheme.

[0033] Here, the ground support service information in the to-be-scheduled airport operating environment can include flight plans, gate states, support vehicle positions, support personnel capabilities and qualifications, and other multi-source heterogeneous data. The ground support information is different in different to-be-scheduled airport operating environments, and the ground support service information corresponding to different times for the same to-be-scheduled airport operating environment is also different.

[0034] Further, after obtaining the ground support service information in the to-be-scheduled airport operation environment, a multi-objective mixed integer linear programming model for the current ground support service information can be determined based on a pre-constructed mixed integer linear programming model.

[0035] In S102, a multi-objective mixed integer linear programming model for scheduling of the current to-be-scheduled airport operation environment is determined based on the ground support service information and a pre-constructed mixed integer linear programming model.

[0036] Here, the mixed integer linear programming model is a combination of an objective function and a constraint condition, and the construction process of the mixed integer linear programming model will be described in detail below.

[0037] Specifically, the mixed integer linear programming model is constructed by the following steps: a1: determining an objective function according to a pre-acquired flight set, a task set, a staff or vehicle set, and at least one target scenario; wherein the objective function is a minimum time value calculated based on a weighted sum of a predicted total service time, a total idle time, and a predicted total delay time of a flight.

[0038] a2: determining at least one constraint condition based on flight plan information, task timing information, resource capacity allocation condition information, space conflict information, and safety demand information.

[0039] a3: constructing the mixed integer linear programming model based on the objective function and the at least one constraint condition.

[0040] In one possible implementation, please refer to Figure 2 , Figure 2 The structure diagram of the mixed integer linear programming model provided by the embodiments of the present application is shown in Figure 2 As shown, the optimization objectives of the mixed integer linear programming model include: optimization objective 1: shortest predicted total service time; optimization objective 2: shortest predicted total idle time; optimization objective 3: shortest predicted total delay time of a flight; and the constraint conditions include: time window constraint, task allocation constraint, work staff execution capability constraint, task processing time constraint, non-negativity and integrity constraint, and task priority constraint.

[0041] Here, before constructing the mixed integer linear programming model, a plurality of sets need to be determined, and then the mixed integer linear programming model is constructed according to different calculation formulas based on different sets.

[0042] Specifically, the sets can include a set of flights, a set of task types (e.g. tractor, boarding bridge, baggage handling, refueling, maintenance, tow truck, etc.), a set of workers, a set of scenarios, representing possible realizations of conditional task graphs; a set of precedence relations between task types.

[0043] Here, the set of flights is represented by the formula ; the set of task types is represented by the formula ; the set of workers is represented by the formula ; the set of scenarios is represented by the formula ; and the set of precedence relations between task types is represented by the formula .

[0044] In one possible implementation, in constructing the mixed integer linear programming model, a plurality of indices representing the combinatorial relations between the sets also need to be determined.

[0045] Specifically, the indices can include a sequence of tasks that flight f needs to complete under scenario s, represented by the formula ; a sequence of tasks i must be completed before the start of sequence of tasks j, represented by the formula ; and a set of task types that worker w can perform, represented by the formula .

[0046] In one possible implementation, in constructing the mixed integer linear programming model, a plurality of parameters and decision variables also need to be determined.

[0047] Specifically, the parameters include the arrival time of a flight, the scheduled departure time of a flight, the standard processing time of a task type, the efficiency coefficient of a worker when performing a task type (as a multiplier of the standard processing time), the occurrence probability of a scenario, and a preset constant large enough.

[0048] Exemplarily, represents the arrival time of flight f; represents the scheduled departure time of flight f; represents the standard processing time of task type T; represents the efficiency coefficient of worker W when performing task type task; represents the occurrence probability of scenario s; and M represents the constant.

[0049] Specifically, the decision variables include the assignment of a flight task to a worker under a scenario, the start and end times of a flight task execution under a scenario, the delay time of a flight under a scenario, and the total idle time of a worker in all scenarios.

[0050] Exemplarily,​​ denotes under the scenario assigns flight tasks to workers at this time, is set to 1, otherwise 0; 0 indicates under the scenario task execution start and end time; 0 indicates the flight delay time under the scenario s; denotes the total idle time of the worker W in all schemes.

[0051] In a possible implementation, according to the obtained flight set, task set, worker or vehicle set and at least one target scenario, the objective function of the mixed integer linear programming model can be determined, specifically, the objective function can be determined by the following formula: ; Wherein, minZ is the objective function; Z1 represents the total expected service time, Z2 represents the total idle time, and Z3 represents the total expected flight delay time; α, β and γ represent weight coefficients, and the importance of each optimization target is balanced through different weight coefficients. The importance of different optimization targets is different according to different business scenarios.

[0052] In a possible implementation, the total expected service time is determined by the following formula: ; Wherein, Z1 represents the total expected service time; denotes the scenario s in the scenario set; denotes the flight ; denotes under the scenario task execution end time; denotes the flight arrival time.

[0053] In a possible implementation, the total idle time is determined by the following formula: ; Wherein, Z2 represents the total idle time; denotes the worker W in the worker set; denotes the total idle time of the worker W in all schemes.

[0054] In a possible implementation, the total expected flight delay time is determined by the following formula: ; Z3 indicates the estimated total flight delay time; Represents scene s in the scene set; Indicates flights in the flight set This represents the probability of scenario s occurring; This indicates the flight delay time under scenario s.

[0055] Furthermore, after determining the objective function, at least one constraint can be determined based on flight schedule information, mission timing information, resource capacity allocation conditions, spatial conflict information, and security requirements. Then, a mixed-integer linear programming model can be constructed based on the objective function and at least one constraint.

[0056] In one possible implementation, the constraints include task assignment constraints, which ensure that for each scenario, each task must be assigned to one worker. Specifically, the task assignment constraints can be determined using the following formula: ; in, This represents worker W in the set of workers; express In the scene The flight mission was assigned to operations personnel W. At this time... Set to 1 if it is set to 1, otherwise set to 0; Represents scene s in the scene set; Indicates flights in the flight set ; This represents a series of tasks that flight f needs to complete in scenario s.

[0057] In another possible implementation, the constraints also include worker performance capability constraints, which ensure that workers can only be assigned tasks they are capable of performing. Specifically, worker performance capability constraints can be determined using the following formula: ; in, This represents worker W in the set of workers; express In the scene The flight mission was assigned to operations personnel W. At this time... Set to 1 if it is set to 1, otherwise set to 0; Represents scene s in the scene set; Indicates flights in the flight set ; This represents a series of tasks that flight f needs to complete in scenario s; Represents scene s in the scene set; This represents the index function.

[0058] In another possible implementation, the constraints also include a processing time limit constraint, where the task's end time is the sum of the task's start time and the effective processing time. Specifically, the processing time limit constraint can be determined using the following formula: ; in, express Flights in the scenario Task start and end times; Indicates flights in the flight set ; This represents a series of tasks that flight f needs to complete in scenario s; This means that task sequence i must be completed before task sequence j begins; This represents scene s in the scene set.

[0059] In another possible implementation, the constraints also include task priority constraints, that is, for each task, a task with a lower priority level must not start before a task with a higher priority level is completed. Specifically, the task priority constraints can be determined by the following formula: ; ; in, express Flights in the scenario Task start and end times; Indicates flights in the flight set ; This represents a series of tasks that flight f needs to complete in scenario s; Indicates flight Arrival time; Indicates flight Planned departure time; This indicates the flight delay time under scenario s; This represents scene s in the scene set.

[0060] In another possible implementation, the constraints also include a time window constraint, namely, the start time of each task cannot be before the flight arrival time, and the last task must be completed before the scheduled departure time. Specifically, the time window constraint can be determined by the following formula: ; in, express Flights in the scenario The start and end times of tasks i and j; M represents a constant; express In the scene Flight task i is assigned to operations personnel W. At this time... Set to 1; express In this scenario, flight task j is assigned to operator W. Set to 1; Indicates flights in the flight set g; Represents flights i and g in the task set; This represents worker W in the set of workers; This represents scene s in the scene set.

[0061] In another possible implementation, the constraints also include worker time scheduling constraints to ensure that workers are not assigned to two tasks simultaneously. M ensures that if a worker is assigned to two tasks, their times do not overlap. Specifically, the worker time scheduling constraints can be determined using the following formula: ; in, express In the scene Flight missions assigned to operations personnel ; express Flights in the scenario Task start and end times; This indicates the flight delay time under scenario s; This represents the total idle time of worker W across all scenarios.

[0062] In another possible implementation, the constraints also include nonnegativity and integrity constraints to ensure that the relevant parameters conform to the actual reality.

[0063] Furthermore, after the mixed-integer linear programming model is constructed, the multi-source heterogeneous data such as flight plans, gate status, support vehicle locations, and support personnel qualifications from the acquired ground support service information are filled into the objective function and various constraint formulas of the mixed-integer linear programming model to obtain a multi-objective mixed-integer linear programming model. After determining the multi-objective mixed-integer linear programming model, a task heterogeneity graph can be constructed based on the multi-objective mixed-integer linear programming model to perform subsequent accurate feature extraction.

[0064] S103. Based on the multi-objective mixed integer linear programming model, construct a task heterogeneous graph, and perform feature extraction based on the task heterogeneous graph to obtain the embedded feature matrix corresponding to the airport operating environment to be scheduled; the embedded feature matrix is ​​used to characterize the global state of the airport operating environment to be scheduled.

[0065] In one possible implementation, the information of multiple elements in the airport operation environment to be scheduled can be modeled as a task heterogeneous graph based on a multi-objective mixed integer linear programming model.

[0066] Specifically, the step "constructing a task heterogeneous graph based on the multi-objective mixed-integer linear programming model" includes: b1: Based on the flight information, mission information, vehicle information, and personnel information in the multi-objective mixed integer linear programming model, determine at least two nodes.

[0067] b2: Based on the task dependency relationship, resource feasibility relationship, spatiotemporal constraint relationship and priority relationship in the multi-objective mixed integer linear programming model, determine the association relationship between every two nodes.

[0068] b3: Construct the task heterogeneous graph based on at least two nodes and the association between each pair of nodes; wherein, the edge between each pair of nodes in the task heterogeneous graph represents the association between the two nodes.

[0069] In one possible implementation, the task heterogeneity graph can be represented by the following formula: ; Where G is the task heterogeneous graph; V represents the set of nodes; and E is used to describe the relationships between nodes.

[0070] Here, nodes include flight nodes, task nodes, vehicle nodes, and personnel nodes, which respectively represent flight entities, support tasks, and operational personnel; the relationships between nodes include at least one of the following: task dependency relationship, resource feasibility relationship, spatiotemporal constraint relationship, and priority relationship.

[0071] For example, task dependencies may include sequential execution, parallel execution, and mutual exclusion relationships; resource feasibility relationships may include executable mappings between tasks and vehicles or personnel; and spatiotemporal constraints may include conflicts between machine location and vehicle path.

[0072] In one possible implementation, please refer to Figure 3 , Figure 3 This is a schematic diagram of the structure of the task heterogeneity graph provided in the embodiments of this application, such as... Figure 3As shown, the heterogeneous task graph includes nodes such as flight nodes, task nodes, vehicle nodes, and personnel nodes, and the relationships between nodes are described by the edges between them.

[0073] Furthermore, after determining the task heterogeneous graph, features can be extracted from the task heterogeneous graph using a graph convolutional neural network to obtain the embedded feature matrix.

[0074] Specifically, the step "based on the task heterogeneous graph, feature extraction is performed to obtain the embedding feature matrix corresponding to the operating environment of the airport to be scheduled" includes: c1: Input the heterogeneous task graph into a pre-trained graph convolutional neural network, so that the multi-layer convolutional neural network of the graph convolutional neural network aggregates local neighborhood information for each node in the heterogeneous task graph and forms a high-dimensional vector containing global structural semantics, and outputs the embedding feature matrix.

[0075] In one possible implementation, the heterogeneous task graph is input into a pre-trained graph convolutional neural network (GCN). This GCN comprises multiple convolutional layers. During each convolutional operation, the features of a node are updated by aggregating the features of its neighboring nodes. Through the stacking of multiple GCNs, nodes can progressively aggregate local neighborhood information and form a high-dimensional representation containing global structural semantics, ultimately obtaining an embedding matrix H. This embedding matrix serves as a global state representation of the airport's operational environment, and the subsequent target scheduling strategy is determined based on the extracted embedding matrix H.

[0076] S104. Input the embedded feature matrix into the pre-trained policy model so that the policy model outputs the target scheduling scheme.

[0077] In one possible implementation, the embedded feature matrix can be input into a pre-trained policy model. After processing by the policy model, the policy model outputs a corresponding target scheduling scheme to realize task resource allocation and path optimization in dynamic scenarios. Here, the policy model is a Markov process and needs to be trained according to the Markov process. The training process of the policy model in the embodiment of this application will be described in detail below.

[0078] Specifically, the policy model is trained through the following steps: d1: Based on the node information, task execution status, and sample ground support service information in the embedded feature matrix, define the reinforcement learning environment of the Markov process; wherein, the reinforcement learning environment includes a state space, an action space, a transition function, and a reward function.

[0079] d2: Input multiple sample ground support service information into the Markov process and perform multiple rounds of training based on the reinforcement learning environment. Utilize the experience replay mechanism and target network update strategy to optimize the parameters of the Markov process. When the average cumulative reward determined based on the reward function converges in multiple consecutive rounds of training and the fluctuation is lower than a preset fluctuation threshold, the Markov process training is determined to be complete. Save the parameters of the current Markov process to obtain the policy model.

[0080] Here, the Markov Decision Process (MDP) is a mathematical model for sequential decision-making, used to simulate stochastic policies and rewards achievable by an agent in an environment where the system state has Markov properties. An agent changes its state and receives rewards by continuously interacting with the environment and choosing actions accordingly. Its goal is to learn a policy that maximizes long-term cumulative rewards.

[0081] Specifically, the reinforcement learning environment includes a state space, an action space, a transition function, and a reward function. The state space S contains node embedding vectors generated by the graph embedding module, task execution states, vehicle availability matrices, flight time windows, and resource usage information. Each state reflects the global situation of the entire airport ground operations system at the corresponding time. The action space A represents the state at time s. t The set of executable decisions under a given state, including vehicle-task matching decisions and path selection actions, where each action can be defined as vehicle v. i Assigned to task t j and along path p i，j Execution; the state transition function P is determined by the task execution process and constraints, describing the system's execution of action a. t The subsequent state change pattern; the reward function R quantifies the timely impact of decision-making behavior on system performance and applies it to strategy optimization.

[0082] In one possible implementation, a composite reward function is set up to simultaneously consider task completion efficiency, vehicle and equipment operating costs, and system coordination.

[0083] Specifically, the reward function is determined through the following steps: e1: Determine the reward function based on the cumulative flight delay time, the driving distance of ground support vehicles, resource utilization indicators, and pre-set penalty information for violating priority or safety constraints.

[0084] Here, the cumulative flight delay time, the driving distance of ground support vehicles, resource utilization indicators, and pre-set penalty information for violating priority or safety constraints can be weighted to determine the reward function. Specifically, the reward function can be determined using the following function: ; Among them, R t For the reward function; This is the total flight delay time; To ensure the driving distance of vehicles on the ground; As an indicator of resource utilization rate; This refers to pre-defined penalty information for violating priority or security constraints; ω1, ω2, ω3, and ω4 are weight coefficients that can be set according to training requirements and reward function settings.

[0085] Here, even the reward function can guide the model to balance task timeliness, energy consumption cost and scheduling security during the optimization process, forming a comprehensive scheduling strategy that takes into account efficiency, security and stability.

[0086] Furthermore, based on the established policy network combining Markov processes and attention mechanisms, a mapping from high-order embedded features to reinforcement learning action distributions is achieved through an encoder-decoder structure. The encoder takes the node vectors embedded by the GCN as input, aggregates dynamic environmental information, and generates a global state representation; the decoder employs a multi-head attention mechanism to focus on key task nodes, outputting the corresponding action probability distributions for overall policy network configuration.

[0087] Here, to enhance the stability of training and the feasibility of constraints, a scenario-constrained near-end policy optimization algorithm is used. Task order constraints and resource conflict constraints are introduced into the policy update. By combining the constraint penalty function with the policy pruning mechanism, the scheduling algorithm is prevented from generating action sequences that violate business logic, thereby improving the processing accuracy of the policy model.

[0088] Furthermore, multiple samples of ground support service information are input into the Markov process for multi-round training based on the reinforcement learning environment. The Markov process is optimized by using an experience replay mechanism and a target network update strategy. When the average cumulative reward determined based on the reward function converges in multiple rounds of training and the fluctuation is lower than a preset fluctuation threshold, the Markov process training is determined to be complete. The parameters of the current Markov process are saved to obtain the policy model.

[0089] For example, please refer to Figure 4 , Figure 4 This is a schematic diagram illustrating the strategy model training process provided in the embodiments of this application, such as... Figure 4As shown, based on the flight information arriving at the airport in succession, the task information in the task pool, and the personnel information in the operation environment, a static graph is generated and embedded into the reinforcement learning policy network. The training process of the policy model is completed through dynamic observation and action and reward information.

[0090] Furthermore, if it is determined that the ground support service information has changed, the strategy model can be dynamically adjusted, and then the subsequent prediction process can be carried out based on the updated and adjusted strategy model to improve the accuracy of the target scheduling plan generation.

[0091] Specifically, after the step of "inputting the embedded feature matrix into a pre-trained policy model so that the policy model outputs a target scheduling scheme", the scheduling method further includes: f1: In response to changes in the ground support service information, the Markov process is retrained or partially updated according to the changed ground support service information to obtain an updated policy model, and an adjusted update scheduling scheme is output based on the updated policy model.

[0092] In one possible implementation, if it is determined that the ground support service information has changed, for example, when there is a flight delay or a change in resources, the Markov process can be retrained or partially updated based on the changed ground support service information to obtain an updated strategy model. The subsequent prediction process can then be performed based on the updated and adjusted strategy model to improve the accuracy of the target scheduling scheme generation.

[0093] The scheduling method for ground support vehicles for flights provided in this application embodiment obtains ground support service information in the operating environment of the airport to be scheduled; based on the ground support service information and a pre-constructed mixed-integer linear programming model, a multi-objective mixed-integer linear programming model for scheduling the current operating environment of the airport to be scheduled is determined; based on the multi-objective mixed-integer linear programming model, a task heterogeneous graph is constructed, and feature extraction is performed based on the task heterogeneous graph to obtain an embedded feature matrix corresponding to the operating environment of the airport to be scheduled; the embedded feature matrix is ​​used to characterize the global state of the operating environment of the airport to be scheduled; the embedded feature matrix is ​​input into a pre-trained strategy model so that the strategy model outputs a target scheduling scheme. In this way, based on the support service information of the operating environment of the airport to be scheduled and the mixed-integer linear programming model, a multi-objective mixed-integer linear programming model is determined, and the multi-objective mixed-integer linear programming model is converted into a task heterogeneous graph for feature extraction to obtain an embedded feature vector characterizing the global state of the operating environment of the airport to be scheduled; the embedded feature vector is input into the strategy model to obtain the corresponding target scheduling scheme, realizing collaborative scheduling and optimal allocation of global resources under complex and highly uncertain conditions, thereby improving the accuracy of scheduling scheme formulation.

[0094] Based on the same inventive concept, this application also provides a scheduling device for flight ground support vehicles corresponding to the scheduling method for flight ground support vehicles. Since the principle of the device in this application is similar to the scheduling method for flight ground support vehicles described above in this application, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0095] Please see Figure 5 , Figure 5 This is a schematic diagram of the structure of a dispatching device for flight ground support vehicles provided in an embodiment of this application. Figure 5 As shown, the scheduling device 500 includes: Information acquisition module 510 is used to acquire ground support service information in the operating environment of the airport to be dispatched; The linear programming model determination module 520 is used to determine a multi-objective mixed-integer linear programming model for scheduling the current airport operation environment based on the ground support service information and the pre-constructed mixed-integer linear programming model. The feature extraction module 530 is used to construct a task heterogeneous graph based on the multi-objective mixed integer linear programming model, and to perform feature extraction based on the task heterogeneous graph to obtain the embedded feature matrix corresponding to the airport operating environment to be scheduled; the embedded feature matrix is ​​used to characterize the global state of the airport operating environment to be scheduled. The scheduling scheme determination module 540 is used to input the embedded feature matrix into a pre-trained policy model so that the policy model outputs a target scheduling scheme.

[0096] In one possible implementation, the linear programming model determination module 520 is used to construct the mixed-integer linear programming model through the following steps: Based on the pre-obtained set of flights, tasks, staff or vehicles, and at least one target scenario, an objective function is determined; wherein, the multi-objective function is the minimum time value calculated by weighted summation of the expected total service time, total idle time, and expected total flight delay time; Based on flight plan information, mission timing information, resource capacity allocation information, spatial conflict information, and security requirements information, at least one constraint condition is determined. Based on the objective function and at least one constraint, the mixed-integer linear programming model is constructed.

[0097] In one possible implementation, when the feature extraction module 530 is used to construct a task heterogeneous graph based on the multi-objective mixed-integer linear programming model, the feature extraction module 530 is used to: Based on the flight information, mission information, vehicle information, and personnel information in the multi-objective mixed integer linear programming model, at least two nodes are determined; Based on the task dependency relationship, resource feasibility relationship, spatiotemporal constraint relationship and priority relationship in the multi-objective mixed integer linear programming model, the association relationship between every two nodes is determined. The task heterogeneous graph is constructed based on at least two nodes and the relationships between each pair of nodes; wherein, the edge between each pair of nodes in the task heterogeneous graph represents the relationship between the two nodes.

[0098] In one possible implementation, when the feature extraction module 530 is used to extract features based on the task heterogeneity graph to obtain the embedded feature matrix corresponding to the airport operating environment to be scheduled, the feature extraction module 530 is used to: The task heterogeneous graph is input into a pre-trained graph convolutional neural network, so that the multi-layer convolutional neural network of the graph convolutional neural network aggregates local neighborhood information for each node in the task heterogeneous graph and forms a high-dimensional vector containing global structural semantics, and outputs the embedding feature matrix.

[0099] In one possible implementation, the scheduling device 500 further includes a model training module (not shown in the figure), which is used to train the policy model through the following steps: Based on the node information, task execution status, and sample ground support service information in the embedded feature matrix, a reinforcement learning environment for the Markov process is defined; wherein, the reinforcement learning environment includes a state space, an action space, a transition function, and a reward function; Multiple samples of ground support service information are input into the Markov process for multi-round training based on the reinforcement learning environment. The Markov process is optimized by using an experience replay mechanism and a target network update strategy. When the average cumulative reward determined based on the reward function converges in multiple rounds of training and the fluctuation is lower than a preset fluctuation threshold, the Markov process training is determined to be complete. The parameters of the current Markov process are saved to obtain the policy model.

[0100] In one possible implementation, the model training module is used to determine the reward function through the following steps: The reward function is determined based on the cumulative flight delay time, the driving distance of ground support vehicles, resource utilization indicators, and pre-set penalty information for violating priority or safety constraints.

[0101] In one possible implementation, the scheduling device 500 further includes a model update module (not shown in the figure), the model update module being used for: In response to changes in the ground support service information, the Markov process is retrained or partially updated according to the changed ground support service information to obtain an updated policy model, and an adjusted update scheduling scheme is output based on the updated policy model.

[0102] The scheduling device for ground support vehicles provided in this application acquires ground support service information in the operating environment of the airport to be scheduled; based on the ground support service information and a pre-constructed mixed-integer linear programming model, a multi-objective mixed-integer linear programming model for scheduling the current operating environment of the airport to be scheduled is determined; based on the multi-objective mixed-integer linear programming model, a task heterogeneous graph is constructed, and feature extraction is performed based on the task heterogeneous graph to obtain an embedded feature matrix corresponding to the operating environment of the airport to be scheduled; the embedded feature matrix is ​​used to characterize the global state of the operating environment of the airport to be scheduled; the embedded feature matrix is ​​input into a pre-trained strategy model so that the strategy model outputs a target scheduling scheme. Thus, based on the support service information of the operating environment of the airport to be scheduled and the mixed-integer linear programming model, a multi-objective mixed-integer linear programming model is determined, and the multi-objective mixed-integer linear programming model is converted into a task heterogeneous graph for feature extraction to obtain an embedded feature vector characterizing the global state of the operating environment of the airport to be scheduled. The embedded feature vector is then input into the strategy model to obtain the corresponding target scheduling scheme, achieving collaborative scheduling and optimal allocation of global resources under complex and highly uncertain conditions, thereby improving the accuracy of scheduling scheme formulation. Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 6 As shown, the electronic device 600 includes a processor 610, a memory 620, and a bus 630.

[0103] The memory 620 stores machine-readable instructions executable by the processor 610. When the electronic device 600 is running, the processor 610 and the memory 620 communicate via the bus 630. When the machine-readable instructions are executed by the processor 610, they can perform the operations described above. Figure 1 The specific implementation of the flight ground support vehicle scheduling method in the method embodiment shown can be found in the method embodiment, and will not be repeated here.

[0104] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, can perform the above-described actions. Figure 1 The specific implementation of the flight ground support vehicle scheduling method in the method embodiment shown can be found in the method embodiment, and will not be repeated here.

[0105] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0106] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0107] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0108] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0109] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] Finally, it should be noted that the above-described embodiments are merely specific implementations of this application, used to illustrate the technical solutions of this application, and not to limit them. The scope of protection of this application is not limited thereto. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some of the technical features, within the scope of the technology disclosed in this application. Such modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for dispatching ground support vehicles for flights, characterized in that, The scheduling method includes: Obtain ground support service information from the airport's operational environment; Based on the ground support service information and the pre-constructed mixed-integer linear programming model, a multi-objective mixed-integer linear programming model for scheduling the current airport operation environment is determined. Based on the multi-objective mixed integer linear programming model, a task heterogeneous graph is constructed, and feature extraction is performed based on the task heterogeneous graph to obtain the embedded feature matrix corresponding to the airport operating environment to be scheduled; the embedded feature matrix is ​​used to characterize the global state of the airport operating environment to be scheduled. The embedded feature matrix is ​​input into a pre-trained policy model so that the policy model outputs a target scheduling scheme.

2. The scheduling method according to claim 1, characterized in that, The mixed-integer linear programming model is constructed using the following steps: Based on the pre-obtained set of flights, tasks, staff or vehicles, and at least one target scenario, a target function is determined; wherein, the target function is the minimum time value calculated by weighted summation of the expected total service time, total idle time, and expected total flight delay time; Based on flight plan information, mission timing information, resource capacity allocation information, spatial conflict information, and security requirements information, at least one constraint condition is determined. Based on the objective function and at least one constraint, the mixed-integer linear programming model is constructed.

3. The scheduling method according to claim 1, characterized in that, The construction of the task heterogeneous graph based on the multi-objective mixed-integer linear programming model includes: Based on the flight information, mission information, vehicle information, and personnel information in the multi-objective mixed integer linear programming model, at least two nodes are determined; Based on the task dependency relationship, resource feasibility relationship, spatiotemporal constraint relationship and priority relationship in the multi-objective mixed integer linear programming model, the association relationship between every two nodes is determined. The task heterogeneous graph is constructed based on at least two nodes and the relationships between each pair of nodes; wherein, the edge between each pair of nodes in the task heterogeneous graph represents the relationship between the two nodes.

4. The scheduling method according to claim 1, characterized in that, Based on the heterogeneous task graph, feature extraction is performed to obtain the embedded feature matrix corresponding to the operating environment of the airport to be scheduled, including: The task heterogeneous graph is input into a pre-trained graph convolutional neural network, so that the multi-layer convolutional neural network of the graph convolutional neural network aggregates local neighborhood information for each node in the task heterogeneous graph and forms a high-dimensional vector containing global structural semantics, and outputs the embedding feature matrix.

5. The scheduling method according to claim 1, characterized in that, The policy model is trained using the following steps: Based on the node information, task execution status, and sample ground support service information in the embedded feature matrix, a reinforcement learning environment for the Markov process is defined; wherein, the reinforcement learning environment includes a state space, an action space, a transition function, and a reward function; Multiple samples of ground support service information are input into the Markov process for multi-round training based on the reinforcement learning environment. The Markov process is optimized by using an experience replay mechanism and a target network update strategy. When the average cumulative reward determined based on the reward function converges in multiple rounds of training and the fluctuation is lower than a preset fluctuation threshold, the Markov process training is determined to be complete. The parameters of the current Markov process are saved to obtain the policy model.

6. The scheduling method according to claim 5, characterized in that, The reward function is determined through the following steps: The reward function is determined based on the cumulative flight delay time, the driving distance of ground support vehicles, resource utilization indicators, and pre-set penalty information for violating priority or safety constraints.

7. The scheduling method according to claim 5, characterized in that, After inputting the embedded feature matrix into a pre-trained policy model to enable the policy model to output a target scheduling scheme, the scheduling method further includes: In response to changes in the ground support service information, the Markov process is retrained or partially updated according to the changed ground support service information to obtain an updated policy model, and an adjusted update scheduling scheme is output based on the updated policy model.

8. A dispatching device for flight ground support vehicles, characterized in that, The scheduling device includes: The information acquisition module is used to acquire ground support service information in the operating environment of the airport to be dispatched; The linear programming model determination module is used to determine a multi-objective mixed-integer linear programming model for scheduling the current airport operation environment based on the ground support service information and a pre-constructed mixed-integer linear programming model. The feature extraction module is used to construct a task heterogeneous graph based on the multi-objective mixed integer linear programming model, and to extract features based on the task heterogeneous graph to obtain the embedded feature matrix corresponding to the airport operating environment to be scheduled; the embedded feature matrix is ​​used to characterize the global state of the airport operating environment to be scheduled. The scheduling scheme determination module is used to input the embedded feature matrix into a pre-trained policy model so that the policy model outputs a target scheduling scheme.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus. The memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the memory via the bus. When the machine-readable instructions are executed by the processor, they perform the steps of the flight ground support vehicle scheduling method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the flight ground support vehicle scheduling method as described in any one of claims 1 to 7.

Images