Determination method, device and equipment for flight ground service guarantee vehicle scheduling and medium

Abstract

The invention provides a flight ground service guarantee vehicle scheduling determination method and device, equipment and a medium, and the method comprises the steps: carrying out the problem solving of a mixed integer linear programming model of a target airport based on a greedy algorithm, and determining an initial solving result of the mixed integer linear programming model; if a GCN removal operator is selected based on a roulette algorithm, inputting feature information of an initial solving result into a graph neural network model, predicting a target vehicle node and removing an aircraft task associated with the target vehicle node; and processing the damaged initial solution result based on the repair operator and the simulated annealing algorithm criterion to obtain a new solution result, dynamically updating the performance score and the weight of the corresponding operator to enter the next iteration, and determining a target solution result of the mixed integer linear programming model until the number of iterations is reached. And the performance of the ALNS algorithm is accelerated and improved by using a GCN removal operator, so that efficient and reliable multi-garage vehicle collaborative scheduling is realized.

Description

Technical Field

[0001] The present application relates to the technical field of airport transportation data processing, and in particular to a method, device, equipment and medium for determining the scheduling of flight ground service support vehicles. Background Art

[0002] Intelligent dispatch of ground service vehicles for airport flights is an important part of the airport operation control and full-process support business system. Whether the scheduling of vehicle support resources is reasonable and efficient will have an impact on the overall operation efficiency of the airport. In real scenarios, airport ground service dispatch business usually involves the problem of multi-garage vehicle dispatch, that is, there are multiple garages in the airport apron, and the vehicle resources in different garages can be dispatched to each other, and various dispatch objectives must be met at the same time. Therefore, how to design an efficient solution algorithm to meet the multi-garage vehicle dispatch needs under large-scale flight missions is crucial to the field of intelligent dispatch of airport ground services. Existing technologies cannot achieve efficient solutions for complex multi-garage vehicle dispatch for airport ground services under a large number of flight missions, so how to achieve efficient and reliable multi-garage vehicle collaborative dispatch has become a technical issue that cannot be underestimated. Summary of the invention

[0003] In view of this, the purpose of the present application is to provide a method, device, equipment and medium for determining the scheduling of flight ground service support vehicles, using an adaptive large neighborhood search (ALNS) algorithm to directly solve the multi-garage ground service flight support vehicle scheduling problem, and using the destruction operator of the graph neural network model to accelerate and improve the performance of the ALNS algorithm, thereby realizing efficient and reliable multi-garage vehicle collaborative scheduling.

[0004] The present application embodiment provides a method for determining flight ground service vehicle scheduling, the method comprising:

[0005] The mixed integer linear programming model of the target airport is solved based on a greedy algorithm to determine the initial solution result of the mixed integer linear programming model; wherein the mixed integer linear programming model is constructed from the multi-garage ground service flight support vehicle scheduling problem; the initial solution result is the vehicle execution path for allocating the corresponding garage to each of the aircraft tasks and each vehicle arriving at the aircraft parking position corresponding to different aircraft tasks;

[0006] A destruction operator is selected based on a roulette algorithm. If a GCN removal operator is selected, the feature information of the initial solution result is input into a graph neural network model, the vehicle node is predicted, a target vehicle node is predicted, and the aircraft mission associated with the target vehicle node is removed to destroy the initial solution result; wherein the graph neural network model is obtained through supervised learning training;

[0007] Based on the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criteria, the destroyed initial solution result is processed to obtain a new solution result, and the performance score and weight of the corresponding operator are dynamically updated to enter the next iteration until the number of iterations is reached to determine the target solution result of the mixed integer linear programming model.

[0008] In a possible implementation, the multi-garage ground service flight support vehicle scheduling problem is constructed in the following manner:

[0009] Based on the decision variables, constraint rules and optimization objectives between the aircraft mission and the vehicle, a multi-garage ground service flight support vehicle scheduling problem for multiple aircraft missions is constructed;

[0010] The decision variables include path allocation decision variables and aircraft mission start time decision variables, the constraint rules include a constraint rule that one aircraft mission is served by one vehicle, a constraint rule on the demand for vehicles to perform aircraft missions, a constraint rule on vehicle continuous service, a constraint rule on vehicle travel, a time constraint rule that the vehicle satisfies the continuous service of aircraft missions, and a constraint rule that the aircraft mission is served within a specified time window, and the optimization target includes the shortest total vehicle travel distance.

[0011] In a possible implementation, the graph neural network model is determined by the following steps:

[0012] Converting the sample mixed integer linear programming model into a bipartite graph format, and determining a sample bipartite graph of the sample mixed integer linear programming model; wherein the sample bipartite graph includes a sample decision variable set, a sample constraint condition set, and a sample edge set formed by associations between the sample decision variables and the sample constraint conditions;

[0013] Input the extracted features of each sample decision variable, the features of each sample constraint condition, and the features of each sample edge into the initial graph neural network model for multiple graph convolution processes to determine the embedded features of each sample decision variable;

[0014] Aggregate and concatenate the embedded features of the plurality of sample decision variables and the features of the plurality of sample vehicles to determine a target embedded feature of each of the sample vehicles; wherein the target embedded feature is the embedded feature of all sample decision variables related to the sample vehicle;

[0015] Based on the multi-layer perceptron, the target embedded features of each of the sample vehicles are predicted and processed, the prediction probability of each of the sample vehicles is determined, the sample vehicles corresponding to the prediction probabilities exceeding the set threshold are selected, and the sample decision variables of the selected sample vehicles are destroyed;

[0016] The initial graph neural network model is subjected to supervised iterative training based on the loss values ​​between labels generated by the sample destruction operator and the pre-designed heuristic destruction operator of the sample mixed integer linear programming model to determine the graph neural network model.

[0017] In a possible implementation, multiple graph convolutions are performed using the following formula:

[0018]

[0019] in, and Respectively represent v i and c j The embedded features are processed by n layers of graph convolution, and and The corresponding original sample decision variable v is transformed by using a linear mapping i The characteristics and original sample constraints c j The features are projected into a 64-dimensional vector space to initialize, h 1 、h 2 Both represent a hidden layer with 128 dimensions, g 1 , g 2 Both represent MLPs with a 64-dimensional output layer. is the set of sample constraints in the historical bipartite graph, is the set of sample decision variables in the historical bipartite graph, e ij v i and c j The sample edges between them, ε is the set of sample edges in the historical bipartite graph.

[0020] In a possible implementation manner, the aggregating and concatenating the embedded features of the plurality of sample decision variables and the features of the plurality of sample vehicles to determine the target embedded feature of each of the sample vehicles includes:

[0021] Aggregating the embedded features of the multiple sample decision variables related to the same sample vehicle among the embedded features of the multiple sample decision variables based on the attention mechanism, and determining the embedded features of the sample decision variables after the aggregation of the sample vehicle;

[0022] The embedded features of the sample decision variables after the sample vehicles are aggregated are concatenated with the features of the sample vehicles to determine the target embedded features of each of the sample vehicles.

[0023] In a possible implementation, the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criterion are used to process the destroyed initial solution result to obtain a new solution result, and the performance score and weight of the corresponding operator are dynamically updated to enter the next iteration until the target solution result of the mixed integer linear programming model is determined when the number of iterations is reached, including:

[0024] Repairing the destroyed initial solution result based on the roulette wheel selection repair operator to determine the latest solution result of the mixed integer linear programming model;

[0025] The simulated annealing algorithm criterion determines whether to accept the latest solution result of the mixed integer linear programming model with a preset probability based on the probability between the latest solution result and the initial solution result, and if accepted, updates the initial solution result of the mixed integer linear programming model to the latest solution result; if not accepted, does not update the initial solution result;

[0026] The weights of the destruction operator and the repair operator are dynamically adjusted based on the latest solution result, and the mixed integer linear programming model is continuously iteratively solved based on the adjusted destruction operator and the adjusted repair operator to determine the target solution result.

[0027] In one possible implementation, the roulette algorithm selects a destruction operator other than the GCN removal operator to destroy the initial solution result in the following manner:

[0028] Selecting the vehicle execution path with the highest cost from the initial solution results, and removing all the aircraft tasks on the vehicle execution path;

[0029] Alternatively, the similarities between the aircraft tasks are determined based on the attributes of the aircraft tasks, and the aircraft tasks corresponding to the highest similarities are removed from the initial solution results based on the similarities, so as to destroy the initial solution results.

[0030] The embodiment of the present application also provides a device for determining the dispatch of flight ground service vehicles, the device comprising:

[0031] An initial solution generation module is used to solve the mixed integer linear programming model of the target airport based on a greedy algorithm to determine the initial solution result of the mixed integer linear programming model; wherein the mixed integer linear programming model is constructed from the multi-garage ground service flight support vehicle scheduling problem; the initial solution result is the vehicle execution path for allocating the corresponding garage to each of the aircraft tasks and each vehicle arriving at the aircraft parking position corresponding to different aircraft tasks;

[0032] A destruction operator determination module is used to select a destruction operator based on a roulette algorithm. If the GCN removal operator is selected, the feature information of the initial solution result is input into the graph neural network model, the vehicle node is predicted, the target vehicle node is predicted, and the aircraft mission associated with the target vehicle node is removed to destroy the initial solution result; wherein the graph neural network model is obtained through supervised learning training;

[0033] The target solution generation module is used to process the destroyed initial solution result based on the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criteria to obtain a new solution result, and dynamically update the performance score and weight of the corresponding operator to enter the next iteration until the target solution result of the mixed integer linear programming model is determined when the number of iterations is reached.

[0034] An embodiment of the present application also provides an electronic device, including: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the method for determining the scheduling of flight ground service support vehicles as described above are performed.

[0035] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for determining the scheduling of flight ground service support vehicles as described above are executed.

[0036] The embodiment of the present application provides a method, device, equipment and medium for determining flight ground service vehicle scheduling, which solves the mixed integer linear programming model of the target airport based on a greedy algorithm to determine the initial solution result of the mixed integer linear programming model; wherein the mixed integer linear programming model is constructed from the multi-garage ground service flight support vehicle scheduling problem; the initial solution result is the vehicle execution path for allocating the corresponding garage to each of the aircraft tasks and each vehicle arriving at the aircraft parking space corresponding to different aircraft tasks; based on the roulette algorithm, if the GCN removal operator is selected, the initial solution is removed. The characteristic information of the result is input into the graph neural network model, the vehicle node is predicted, the target vehicle node is predicted and the aircraft mission associated with the target vehicle node is removed to destroy the initial solution result; wherein the graph neural network model is obtained through supervised learning training; the initial solution result after destruction is processed based on the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criterion to obtain a new solution result, and the performance score and weight of the corresponding operator are dynamically updated to enter the next iteration until the target solution result of the mixed integer linear programming model is determined when the number of iterations is reached. The adaptive large neighborhood search (ALNS) algorithm is used to directly solve the multi-garage ground service flight support vehicle scheduling problem, and the destruction operator of the graph neural network model is used to accelerate and improve the performance of the ALNS algorithm, thereby realizing efficient and reliable multi-garage vehicle collaborative scheduling.

[0037] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0039] Figure 1 A flowchart of a method for determining flight ground service vehicle scheduling provided in an embodiment of the present application;

[0040] Figure 2 One of the structural schematic diagrams of a device for determining flight ground service vehicle scheduling provided in an embodiment of the present application;

[0041] Figure 3 A second structural diagram of a device for determining flight ground service vehicle scheduling provided in an embodiment of the present application;

[0042] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0043] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application usually described and shown in the drawings here 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 application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work belongs to the scope of protection of the present application.

[0044] First, the application scenarios to which the present application is applicable are introduced. The present application can be applied to the field of airport transportation data processing technology.

[0045] Research has found that the intelligent dispatch of airport flight ground service vehicles is an important part of the airport operation control and full-process support business system. Whether the vehicle support resource dispatch is reasonable and efficient will have an impact on the overall operation efficiency of the airport. In real scenarios, the airport ground service dispatch business usually involves the multi-garage vehicle dispatch problem, that is, there are multiple garages in the airport apron, and the vehicle resources in different garages can be dispatched to each other, and various dispatch objectives must be met at the same time. Therefore, how to design an efficient solution algorithm to meet the multi-garage vehicle dispatch needs under large-scale flight missions is crucial to the field of airport ground service intelligent dispatch. Existing technologies cannot achieve efficient solutions for the complex multi-garage vehicle dispatch of airport ground services under a large number of flight missions, so how to achieve efficient and reliable multi-garage vehicle collaborative dispatch has become a technical problem that cannot be underestimated.

[0046] Based on this, an embodiment of the present application provides a method for determining the scheduling of flight ground service vehicles, which uses an adaptive large neighborhood search (ALNS) algorithm to directly solve the multi-garage ground service flight support vehicle scheduling problem, and uses the destruction operator of the graph neural network model to accelerate and improve the performance of the ALNS algorithm, thereby achieving efficient and reliable multi-garage vehicle collaborative scheduling.

[0047] See also Figure 1 , Figure 1 This is a flow chart of a method for determining the dispatch of flight ground service vehicles provided in an embodiment of the present application. Figure 1 As shown in , the determination method provided in the embodiment of the present application includes:

[0048] S101: Solving the mixed integer linear programming model of the target airport based on a greedy algorithm to determine an initial solution result of the mixed integer linear programming model.

[0049] In this step, the mixed integer linear programming model of the target airport is solved according to the greedy algorithm in the adaptive large neighborhood search algorithm to determine the initial solution result of the mixed integer linear programming model.

[0050] Among them, the mixed integer linear programming model is constructed from the multi-garage ground service flight support vehicle scheduling problem; the initial solution result is the vehicle execution path that allocates the corresponding garage to each of the aircraft tasks and each vehicle arrives at the aircraft parking space corresponding to different aircraft tasks.

[0051] In a possible implementation, the multi-garage ground service flight support vehicle scheduling problem is constructed in the following manner:

[0052] Based on the decision variables, constraint rules and optimization objectives between the aircraft mission and the vehicle, a multi-garage ground service flight support vehicle scheduling problem for multiple aircraft missions is constructed;

[0053] The decision variables include path allocation decision variables and aircraft mission start time decision variables, the constraint rules include a constraint rule that one aircraft mission is served by one vehicle, a constraint rule on the demand for vehicles to perform aircraft missions, a constraint rule on vehicle continuous service, a constraint rule on vehicle travel, a time constraint rule that the vehicle satisfies the continuous service of aircraft missions, and a constraint rule that the aircraft mission is served within a specified time window, and the optimization target includes the shortest total vehicle travel distance.

[0054] Here, the path assignment decision variable is x i,j,k ∈{0,1}, if vehicle k∈K moves from position i∈L to position j∈L, then this variable is equal to 1, otherwise it is equal to 0, L is the index set of the visited positions; the decision variable for the start time of the aircraft mission is T fk ≥0, the starting time of vehicle k to the parking space corresponding to aircraft task f∈F.

[0055] The constraint rule that an aircraft mission is served by a vehicle is:

[0056]

[0057] Where k∈K is the set of vehicles, i∈L is the set of visited locations, and f∈F is the set of aircraft missions.

[0058] Among them, the constraint rules for the vehicle's demand for aircraft missions are:

[0059]

[0060] Where, d i is the i-th aircraft mission, is the capacity limit of vehicle k.

[0061] Among them, the vehicle continuous service constraint rule is that the vehicle must start from the current aircraft position to perform the next aircraft mission:

[0062]

[0063] The vehicle travel constraint rule is that all vehicles must start from a certain garage and eventually return to the garage:

[0064]

[0065] Where d∈D={n+1,n+2,...,n+m}: the index and set of garages, and m is the number of garages.

[0066] Among them, the time constraint rule for vehicles to meet the continuous service of aircraft missions is that all vehicles must meet the time constraint of continuous service of aircraft:

[0067] T jk ≥T ik +p i +c ij -M·(1-x ijk )

[0068] Where, T jk is the time when vehicle k arrives at aircraft task j to start the task, p i is the processing time of aircraft task i, c ij is the travel time required from the position of the aircraft parking stand corresponding to aircraft mission i to the position of the aircraft parking stand corresponding to aircraft mission j, M=10000.

[0069] Among them, the constraint rules for aircraft missions to be served within the specified time window are:

[0070]

[0071] In the formula, a i is the earliest start time of executing task i∈F, b i is the latest start time for executing task i∈F.

[0072] Among them, the shortest total vehicle travel distance is:

[0073]

[0074] S102: Select a destruction operator based on the roulette algorithm. If the GCN removal operator is selected, the feature information of the initial solution result is input into the graph neural network model, the vehicle node is predicted, the target vehicle node is predicted and the aircraft mission associated with the target vehicle node is removed to destroy the graph neural network model of the initial solution result.

[0075] In this step, the GCN removal operator is selected according to the roulette algorithm, and the feature information of the initial solution result is input into the graph neural network model, and the vehicle node prediction processing is performed on the feature information to predict the target vehicle node to be removed, and the aircraft mission associated with the target vehicle node is removed to complete the destruction of the initial solution result.

[0076] Here, the relevant information that will be degraded when the initial solution is used as input is converted into node features, and the information of the initial solution is extracted and input into the model as features to directly predict the vehicle node. The feature information includes the value of the solution, whether the vehicle is used in the solution, and the cosine similarity with the target.

[0077] The graph neural network model is obtained by performing supervised learning training on the initial graph neural network model.

[0078] In a possible implementation, the graph neural network model is determined by the following steps:

[0079] A: Convert the sample mixed integer linear programming model into a bipartite graph format, and determine a sample bipartite graph of the sample mixed integer linear programming model; wherein the sample bipartite graph includes a sample decision variable set, a sample constraint condition set, and a sample edge set formed by the association between the sample decision variables and the sample constraint conditions.

[0080] Here, the sample mixed integer linear programming model is converted into a bipartite graph format, and a sample bipartite graph of the sample mixed integer linear programming model is determined.

[0081] The sample bipartite graph includes a sample decision variable set, a sample constraint condition set, and a sample edge set formed by the association between the sample decision variables and the sample constraint conditions.

[0082] Here, the sample mixed integer linear programming model is represented as a bipartite graph in represents the sample decision variable, represents the sample constraint, e ij =(v i ,c j)∈ε represents the association (edge) between sample decision variables and sample constraints; all sample decision variables are represented by rows in the adjacency matrix of the bipartite graph of the mixed integer programming model and are represented as a feature matrix n v is the characteristic dimension of the sample decision variable, is the number of sample decision variables, all sample constraints are represented by the columns in the adjacency matrix of the bipartite graph of the mixed integer programming model and are represented as a feature matrix n c is the characteristic dimension of the sample constraint, is the number of sample constraints. All edges represent the association between decision variables and constraints. Currently, edges exist only when the coefficient of the decision variable is not 0, and are represented as a feature matrix n e is the dimension of the edge feature.

[0083] B: Input the extracted features of each sample decision variable, the features of each sample constraint condition, and the features of each sample edge into the initial graph neural network model for multiple graph convolution processes to determine the embedded features of each sample decision variable.

[0084] Here, the extracted features of each sample decision variable, the features of each sample constraint condition, and the features of each sample edge are input into the initial graph neural network model for multiple graph convolution processes to determine the embedded features of each sample decision variable.

[0085] In a possible implementation, multiple graph convolutions are performed using the following formula:

[0086]

[0087] in, and Respectively represent v i and c j The embedded features are processed by n layers of graph convolution, and and The corresponding original sample decision variable v is transformed by using a linear mapping i The characteristics and original sample constraints c j The features are projected into a 64-dimensional vector space to initialize, h 1 、h 2 Both represent a hidden layer with 128 dimensions, g 1 , g 2 Both represent MLPs with a 64-dimensional output layer. is the set of sample constraints in the historical bipartite graph, is the set of sample decision variables in the historical bipartite graph, e ij v i and c j The sample edges between them, ε is the set of sample edges in the historical bipartite graph.

[0088] Among them, there are two types of graph convolution processing: one is from constraints to decision variables, and the other is from decision variables to constraints.

[0089] C: Aggregate and concatenate the embedded features of multiple sample decision variables and the features of multiple sample vehicles to determine the target embedded features of each sample vehicle; wherein the target embedded features are the embedded features of all sample decision variables related to the sample vehicle.

[0090] Here, the embedding features of multiple sample decision variables are aggregated, and then the embedded features of the aggregated sample decision variables are concatenated with the features of the sample vehicles to determine the target embedding features of each sample vehicle.

[0091] In a possible implementation, the embedding features of the plurality of sample decision variables and the features of the plurality of sample vehicles are aggregated and concatenated to determine a target embedding feature of each of the sample vehicles, including:

[0092] a: Based on the attention mechanism, the embedded features of multiple sample decision variables related to the same sample vehicle are aggregated among the embedded features of the multiple sample decision variables to determine the embedded features of the sample decision variables after the sample vehicle is aggregated.

[0093] Here, according to the attention mechanism, the embedding features of multiple sample decision variables related to the same sample vehicle are aggregated among the embedding features of multiple sample decision variables to determine the embedding features of the sample decision variables after the sample vehicles are aggregated.

[0094] b: Concatenate the embedded features of the sample decision variables after the sample vehicles are aggregated with the features of the sample vehicles to determine the target embedded features of each of the sample vehicles.

[0095] Here, the embedded features of the sample decision variables after the sample vehicles are aggregated are concatenated with the features of the sample vehicles to determine the target embedded features of each sample vehicle.

[0096] Here, the target embedding feature of each sample vehicle is determined by the following formula:

[0097]

[0098] Among them, w his the 64-dimensional embeddings corresponding to the sample vehicle and the features obtained by initialization, η(·) represents the feature to be combined with w h The embedding features of all relevant sample decision variables are aggregated into a set, a i Embedded features v representing sample decision variables i The corresponding attention score.

[0099] Among them, the attention score is calculated by the following formula:

[0100]

[0101] D: Based on the multi-layer perceptron, the target embedded features of each of the sample vehicles are predicted and processed to determine the prediction probability of each of the sample vehicles, the sample vehicles corresponding to the prediction probabilities exceeding the set threshold are selected, and the sample decision variables of the selected sample vehicles are destroyed.

[0102] Here, the target embedded features of each sample vehicle are predicted and processed through a multi-layer perceptron to determine the prediction probability of each sample vehicle, select the sample vehicles corresponding to the prediction probability exceeding the set threshold, and destroy / remove the sample decision variables of the selected sample vehicles.

[0103] Among them, the target embedding feature of the sample vehicle represents all decision variables related to the vehicle. Finally, an MLP with a 128-dimensional hidden layer and a 1-dimensional output layer is used to predict the variables. The formula is as follows:

[0104] P(w h )=Sigmoid(h 3 (w h Target embedding features)),

[0105] Among them, h 3 (·) represents MLP, and finally each vehicle is obtained The predicted probability of the vehicle and its corresponding decision variables are selected for removal based on the probability. h ) is the predicted probability of the h-th vehicle.

[0106] E: Based on the predicted probability of the sample vehicle and the loss value between the labels generated by the heuristic destruction operator of the pre-designed sample mixed integer linear programming model, the initial graph neural network model is subjected to supervised iterative training to determine the graph neural network model.

[0107] Here, the initial graph neural network model is iteratively trained according to the predicted probability of the sample vehicle and the loss value between the labels generated by the heuristic destruction operator of the predetermined sample mixed integer linear programming model to determine the graph neural network model.

[0108] Among them, the heuristic destruction operator considers both the degree of destruction (how many vehicles are selected and the flight tasks associated with these vehicles are removed) and randomness (randomly selecting vehicles) to destroy the current solution repeatedly multiple times, and selects the one with the best effect after destruction (shortest total distance) as the training label.

[0109] Here, the loss value between the heuristic destruction operator and the sample destruction operator is determined by the following formula:

[0110]

[0111] in, is the loss value, when y(w h )=1 means that the vehicle w h A set of relevant decision variables is selected, y(w h )=0 means it is not selected. Through supervised learning, GCN has the ability to select and remove variables. h ) is the predicted probability of the vehicle (sample destruction operator), y(w H ) is the vehicle corresponding to the predetermined heuristic destruction operator.

[0112] Here, when the loss value is less than or equal to the preset threshold, the initial graph neural network model is used as the graph neural network model. When the loss value is greater than the preset threshold, the network parameters of the initial graph neural network model are changed and the initial graph neural network model continues to be trained until the loss value is less than or equal to the preset threshold, and the change of the network parameters of the initial graph neural network model is stopped.

[0113] S103: Based on the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criteria, the destroyed initial solution result is processed to obtain a new solution result, and the performance score and weight of the corresponding operator are dynamically updated to enter the next iteration until the number of iterations is reached to determine the target solution result of the mixed integer linear programming model.

[0114] In this step, the initial solution result after destruction is processed according to the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criteria to obtain a new solution result. At the same time, the performance score and weight of the corresponding operator are dynamically updated, and the next iteration is entered until the target solution result of the mixed integer linear programming model is determined when the number of iterations is reached.

[0115] In a possible implementation, the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criterion are used to process the destroyed initial solution result to obtain a new solution result, and the performance score and weight of the corresponding operator are dynamically updated to enter the next iteration until the target solution result of the mixed integer linear programming model is determined when the number of iterations is reached, including:

[0116] a: repairing the destroyed initial solution result based on the roulette wheel selection repair operator to determine the latest solution result of the mixed integer linear programming model.

[0117] Here, the damaged initial solution result is repaired according to the roulette wheel selection repair operator to determine the latest solution result of the mixed integer linear programming model.

[0118] b: The simulated annealing algorithm determines whether to accept the latest solution result of the mixed integer linear programming model with a preset probability based on the probability between the latest solution result and the initial solution result. If accepted, the initial solution result of the mixed integer linear programming model is updated to the latest solution result. If not accepted, the initial solution result is not updated.

[0119] Here, the simulated annealing algorithm criterion determines whether to accept the latest solution result of the mixed integer linear programming model with a preset probability based on the probability between the latest solution result and the initial solution result. If accepted, the initial solution result of the mixed integer linear programming model is updated to the latest solution result. If not accepted, the initial solution result is not updated.

[0120] c: dynamically adjusting the weights of the destruction operator and the repair operator based on the latest solution result, and continuing to iteratively solve the mixed integer linear programming model based on the adjusted destruction operator and the adjusted repair operator to determine the target solution result.

[0121] Here, the weights of the destruction operator and the repair operator are dynamically adjusted according to the latest solution result, and the mixed integer linear programming model is iteratively solved according to the adjusted destruction operator and the adjusted repair operator to determine the target solution result.

[0122] In a possible implementation manner, the initial solution result is destroyed in the following manner:

[0123] (1): Select the vehicle execution path with the highest cost from the initial solution results, and remove all the aircraft tasks on the vehicle execution path.

[0124] Here, the vehicle execution path with the highest cost is selected from the initial solution results, and all aircraft tasks on the vehicle execution path are removed.

[0125] (2): Determine the similarity between the aircraft tasks based on the attributes of the aircraft tasks, and remove the aircraft task corresponding to the highest similarity from the initial solution results based on the similarity, so as to destroy the initial solution results.

[0126] Here, the similarity between the aircraft tasks is determined according to the attributes of the aircraft tasks, and the aircraft task corresponding to the highest similarity is removed from the initial solution results according to the similarity, so as to destroy the initial solution results.

[0127] In the specific implementation example, step 1: construct a mixed integer programming model for the multi-garage vehicle scheduling problem, including decision variables (path allocation variables, aircraft mission start time decision variables, constraints (mission service satisfaction, vehicle capacity limit, entry and exit of the same garage, continuous service time, etc.) and optimization goals (shortest total driving distance), characterize the specific problem through the model, and design an algorithm for efficient solution. Step 2: for the above mixed integer programming model, design an adaptive large neighborhood search (ALNS) algorithm for efficient solution. The ALNS algorithm includes a destruction operator, a repair operator and a dynamic weight operator. Re-adjust and adaptively select. For the destruction operator, a variety of operators are designed to destroy the current solution, including GCN removal, worst removal, and similar removal. The GCN removal operator is a destruction operator based on graph neural network designed in this paper, which can accelerate the solution and improve the performance of the ALNS algorithm. It includes GCN model structure design and heuristic operator generation label and training of GCN model. First, the mixed integer programming model is represented as a bipartite graph, which includes two types of nodes: decision variable nodes and constraint nodes; secondly, the GCN network structure is designed and message passing and information are performed on the bipartite graph. The information is aggregated and the vehicle nodes are predicted, with the goal of selecting the flight mission nodes related to the vehicle for removal; finally, the heuristic operator is used to destroy the current solution, and after repeated multiple times, the best destruction solution is selected as the label, and the GCN model is trained with the label based on the cross entropy to guide it to obtain the best destruction solution. Step 3: Based on the trained GCN removal operator, the ALNS algorithm is solved as a whole. Among them, the role of the other two destruction operators is to increase generalization and avoid the results falling into the local optimum; for the repair operator, a variety of operators are also designed to repair the destroyed solution, including greedy Greedy insertion and regret value insertion; use simulated annealing Metropolis criterion to determine whether the current generated solution is accepted with a certain probability to avoid the algorithm falling into the local optimum, and use the maximum number of iterations as the termination condition of the algorithm. By designing multiple groups of destruction operators and repair operators, the search range of the solution space is expanded. Destruction and repair operators with good performance will get high scores accordingly, and the selection weight will be higher. In each iteration process, each destruction and repair operator is selected and weighted based on the performance before the iteration. The efficient combination method is used to improve the overall optimization ability of the algorithm and finally find the optimal solution.

[0128] The embodiment of the present application provides a method for determining the scheduling of flight ground service vehicles, the determination method comprising: solving a mixed integer linear programming model of a target airport based on a greedy algorithm to determine an initial solution result of the mixed integer linear programming model; wherein the mixed integer linear programming model is constructed from a multi-garage ground service flight support vehicle scheduling problem; the initial solution result is a vehicle execution path for allocating a corresponding garage to each of the aircraft tasks and each vehicle arriving at an aircraft parking space corresponding to a different aircraft task; if a GCN removal operator is selected based on a roulette algorithm, the feature information of the initial solution result is input into a graph neural network model, the vehicle node is predicted, the target vehicle node is predicted, and the aircraft task associated with the target vehicle node is removed to destroy the initial solution result; wherein the graph neural network model is obtained through supervised learning training; the initial solution result after destruction is processed based on the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criterion to obtain a new solution result, and the performance score and weight of the corresponding operator are dynamically updated to enter the next iteration until the target solution result of the mixed integer linear programming model is determined when the number of iterations is reached. The adaptive large neighborhood search (ALNS) algorithm is used to directly solve the multi-garage ground service flight support vehicle scheduling problem, and the destruction operator of the graph neural network model is used to accelerate and improve the performance of the ALNS algorithm, thereby achieving efficient and reliable multi-garage vehicle collaborative scheduling.

[0129] See also Figure 2 , Figure 3 , Figure 2 One of the structural schematic diagrams of a device for determining flight ground service vehicle scheduling provided in an embodiment of the present application; Figure 3 This is a second structural diagram of a device for determining flight ground service vehicle scheduling provided in an embodiment of the present application. Figure 3 As shown in , the determining device 200 includes:

[0130] The initial solution generation module 210 is used to solve the mixed integer linear programming model of the target airport based on a greedy algorithm to determine the initial solution result of the mixed integer linear programming model; wherein the mixed integer linear programming model is constructed from the multi-garage ground service flight support vehicle scheduling problem; the initial solution result is the vehicle execution path for allocating the corresponding garage to each of the aircraft tasks and each vehicle arriving at the aircraft parking position corresponding to different aircraft tasks;

[0131] A destruction operator determination module 220 is used to select a destruction operator based on a roulette algorithm. If the GCN removal operator is selected, the feature information of the initial solution result is input into a graph neural network model, the vehicle node is predicted, the target vehicle node is predicted, and the aircraft mission associated with the target vehicle node is removed to destroy the initial solution result; wherein the graph neural network model is obtained through supervised learning training;

[0132] The target solution generation module 230 is used to process the destroyed initial solution result based on the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criteria to obtain a new solution result, and dynamically update the performance score and weight of the corresponding operator to enter the next iteration until the target solution result of the mixed integer linear programming model is determined when the number of iterations is reached.

[0133] Further, such as Figure 3 As shown, the determination device 200 further includes a construction module 240, and the construction module 240 constructs the multi-garage ground service flight support vehicle scheduling problem in the following manner:

[0134] Based on the decision variables, constraint rules and optimization objectives between the aircraft mission and the vehicle, a multi-garage ground service flight support vehicle scheduling problem for multiple aircraft missions is constructed;

[0135] The decision variables include path allocation decision variables and aircraft mission start time decision variables, the constraint rules include a constraint rule that one aircraft mission is served by one vehicle, a constraint rule on the demand for vehicles to perform aircraft missions, a constraint rule on vehicle continuous service, a constraint rule on vehicle travel, a time constraint rule that the vehicle satisfies the continuous service of aircraft missions, and a constraint rule that the aircraft mission is served within a specified time window, and the optimization target includes the shortest total vehicle travel distance.

[0136] Further, such as Figure 3 As shown, the determination device 200 also includes a model training module 250, and the model training module 250 determines the graph neural network model through the following steps:

[0137] Converting the sample mixed integer linear programming model into a bipartite graph format, and determining a sample bipartite graph of the sample mixed integer linear programming model; wherein the sample bipartite graph includes a sample decision variable set, a sample constraint condition set, and a sample edge set formed by associations between the sample decision variables and the sample constraint conditions;

[0138] Input the extracted features of each sample decision variable, the features of each sample constraint condition, and the features of each sample edge into the initial graph neural network model for multiple graph convolution processes to determine the embedded features of each sample decision variable;

[0139] Aggregate and concatenate the embedded features of the plurality of sample decision variables and the features of the plurality of sample vehicles to determine a target embedded feature of each of the sample vehicles; wherein the target embedded feature is the embedded feature of all sample decision variables related to the sample vehicle;

[0140] Based on the multi-layer perceptron, the target embedded features of each of the sample vehicles are predicted and processed, the prediction probability of each of the sample vehicles is determined, the sample vehicles corresponding to the prediction probabilities exceeding the set threshold are selected, and the sample decision variables of the selected sample vehicles are destroyed;

[0141] The initial graph neural network model is subjected to supervised iterative training based on the predicted probability of the sample vehicle and the loss value between the labels generated by the heuristic destruction operator of the pre-designed sample mixed integer linear programming model to determine the graph neural network model.

[0142] Furthermore, the training module 250 performs multiple graph convolution processes using the following formula:

[0143]

[0144] in, and Respectively represent v i and c j The embedded features are processed by n layers of graph convolution, and and The corresponding original sample decision variable v is transformed by using a linear mapping i The characteristics and original sample constraints c j The features of h are projected into a 65-dimensional vector space to initialize, 1 、h 2 Both represent a hidden layer with 128 dimensions, g 1 , g 2 Both represent MLPs with a 65-dimensional output layer. is the set of sample constraints in the historical bipartite graph, is the set of sample decision variables in the historical bipartite graph, e ij v i and c j The sample edges between them, ε is the set of sample edges in the historical bipartite graph.

[0145] Furthermore, when the training module 250 is used to aggregate and concatenate the embedded features of the plurality of sample decision variables and the features of the plurality of sample vehicles to determine the target embedded features of each of the sample vehicles, the training module 250 is specifically used to:

[0146] Aggregating the embedded features of the multiple sample decision variables related to the same sample vehicle among the embedded features of the multiple sample decision variables based on the attention mechanism, and determining the embedded features of the sample decision variables after the aggregation of the sample vehicle;

[0147] The embedded features of the sample decision variables after the sample vehicles are aggregated are concatenated with the features of the sample vehicles to determine the target embedded features of each of the sample vehicles.

[0148] Further, the target solution generation module 230 processes the damaged initial solution result based on the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criterion to obtain a new solution result, and dynamically updates the performance score and weight of the corresponding operator to enter the next iteration until the target solution result of the mixed integer linear programming model is determined when the number of iterations is reached. The target solution generation module 230 is specifically used to:

[0149] Repairing the destroyed initial solution result based on the roulette wheel selection repair operator to determine the latest solution result of the mixed integer linear programming model;

[0150] The simulated annealing algorithm criterion determines whether to accept the latest solution result of the mixed integer linear programming model with a preset probability based on the probability between the latest solution result and the initial solution result, and if accepted, updates the initial solution result of the mixed integer linear programming model to the latest solution result; if not accepted, does not update the initial solution result;

[0151] The weights of the destruction operator and the repair operator are dynamically adjusted based on the latest solution result, and the mixed integer linear programming model is continuously iteratively solved based on the adjusted destruction operator and the adjusted repair operator to determine the target solution result.

[0152] Furthermore, the destruction operator determination module 220 destroys the initial solution result in the following manner:

[0153] Selecting the vehicle execution path with the highest cost from the initial solution results, and removing all the aircraft tasks on the vehicle execution path;

[0154] Alternatively, the similarities between the aircraft tasks are determined based on the attributes of the aircraft tasks, and the aircraft tasks corresponding to the highest similarities are removed from the initial solution results based on the similarities, so as to destroy the initial solution results.

[0155] An embodiment of the present application provides a device for determining flight ground service vehicle scheduling, the device comprising:

[0156] An initial solution generation module is used to solve the mixed integer linear programming model of the target airport based on a greedy algorithm to determine the initial solution result of the mixed integer linear programming model; wherein the mixed integer linear programming model is constructed from the multi-garage ground service flight support vehicle scheduling problem; the initial solution result is the vehicle execution path for allocating the corresponding garage to each of the aircraft tasks and each vehicle arriving at the aircraft parking position corresponding to different aircraft tasks;

[0157] The destruction operator determination module is used to input the characteristic information of the initial solution result into the graph neural network model if the GCN removal operator is selected based on the roulette algorithm, predict the vehicle node, predict the target vehicle node and remove the aircraft mission associated with the target vehicle node to destroy the initial solution result; wherein the graph neural network model is obtained through supervised learning training; the target solution generation module is used to process the destroyed initial solution result based on the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criterion to obtain a new solution result, and dynamically update the performance score and weight of the corresponding operator to enter the next iteration until the target solution result of the mixed integer linear programming model is determined when the number of iterations is reached. The adaptive large neighborhood search (ALNS) algorithm is used to directly solve the multi-garage ground service flight support vehicle scheduling problem, and the destruction operator of the graph neural network model is used to accelerate and improve the performance of the ALNS algorithm, thereby realizing efficient and reliable multi-garage vehicle collaborative scheduling.

[0158] See also Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 4 As shown in , the electronic device 400 includes a processor 410 , a memory 420 and a bus 430 .

[0159] The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 communicates with the memory 420 via the bus 430. When the machine-readable instructions are executed by the processor 410, the above-mentioned Figure 1 The specific implementation of the steps of the method for determining the flight ground service vehicle scheduling in the method embodiment shown can be found in the method embodiment, and will not be repeated here.

[0160] The present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1 The specific implementation of the steps of the method for determining the flight ground service vehicle scheduling in the method embodiment shown can be found in the method embodiment, and will not be repeated here.

[0161] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0162] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0163] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0164] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0165] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application can essentially or in other words, the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0166] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for determining the dispatch of flight ground service vehicles, characterized in that: The determination method comprises: The mixed integer linear programming model of the target airport is solved based on a greedy algorithm to determine the initial solution result of the mixed integer linear programming model; wherein the mixed integer linear programming model is constructed from the multi-garage ground service flight support vehicle scheduling problem; the initial solution result is the vehicle execution path for allocating the corresponding garage to each of the aircraft tasks and each vehicle arriving at the aircraft parking position corresponding to different aircraft tasks; A destruction operator is selected based on a roulette algorithm. If a GCN removal operator is selected, the feature information of the initial solution result is input into a graph neural network model, the vehicle node is predicted, a target vehicle node is predicted, and the aircraft mission associated with the target vehicle node is removed to destroy the initial solution result; wherein the graph neural network model is obtained through supervised learning training; Based on the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criteria, the destroyed initial solution result is processed to obtain a new solution result, and the performance score and weight of the corresponding operator are dynamically updated to enter the next iteration until the number of iterations is reached to determine the target solution result of the mixed integer linear programming model.

2. The determination method according to claim 1, characterized in that: The multi-garage ground service flight support vehicle scheduling problem is constructed in the following way: Based on the decision variables, constraint rules and optimization objectives between the aircraft mission and the vehicle, a multi-garage ground service flight support vehicle scheduling problem for multiple aircraft missions is constructed; The decision variables include path allocation decision variables and aircraft mission start time decision variables, the constraint rules include a constraint rule that one aircraft mission is served by one vehicle, a constraint rule on the demand for vehicles to perform aircraft missions, a constraint rule on vehicle continuous service, a constraint rule on vehicle travel, a time constraint rule that the vehicle satisfies the continuous service of aircraft missions, and a constraint rule that the aircraft mission is served within a specified time window, and the optimization target includes the shortest total vehicle travel distance.

3. The determination method according to claim 1, characterized in that: The graph neural network model is determined by the following steps: Converting the sample mixed integer linear programming model into a bipartite graph format, and determining a sample bipartite graph of the sample mixed integer linear programming model; wherein the sample bipartite graph includes a sample decision variable set, a sample constraint condition set, and a sample edge set formed by associations between the sample decision variables and the sample constraint conditions; Input the extracted features of each sample decision variable, the features of each sample constraint condition, and the features of each sample edge into the initial graph neural network model for multiple graph convolution processes to determine the embedded features of each sample decision variable; Aggregate and concatenate the embedded features of the plurality of sample decision variables and the features of the plurality of sample vehicles to determine a target embedded feature of each of the sample vehicles; wherein the target embedded feature is the embedded feature of all sample decision variables related to the sample vehicle; Based on the multi-layer perceptron, the target embedded features of each of the sample vehicles are predicted and processed, the prediction probability of each of the sample vehicles is determined, the sample vehicles corresponding to the prediction probabilities exceeding the set threshold are selected, and the sample decision variables of the selected sample vehicles are destroyed; The initial graph neural network model is subjected to supervised iterative training based on the predicted probability of the sample vehicle and the loss value between the labels generated by the heuristic destruction operator of the pre-designed sample mixed integer linear programming model to determine the graph neural network model.

4. The determination method according to claim 3, characterized in that: Multiple graph convolutions are performed using the following formula: in, and Respectively represent v i and c j The embedded features are processed by n layers of graph convolution, and and The corresponding original sample decision variable v is transformed by using a linear mapping i The characteristics and original sample constraints c j The features of are projected into a 64-dimensional vector space to initialize. h1 and h2 both represent MLPs with 128-dimensional hidden layers, and g1 and g2 both represent MLPs with 64-dimensional output layers. is the set of sample constraints of the historical bipartite graph, V is the set of sample decision variables of the historical bipartite graph, e ij v i and c j The sample edges between , ε is the set of sample edges of the historical bipartite graph.

5. The determination method according to claim 3, characterized in that: The step of aggregating and concatenating the embedded features of the plurality of sample decision variables and the features of the plurality of sample vehicles to determine a target embedded feature of each of the sample vehicles includes: Aggregating the embedded features of the multiple sample decision variables related to the same sample vehicle among the embedded features of the multiple sample decision variables based on the attention mechanism, and determining the embedded features of the sample decision variables after the aggregation of the sample vehicle; The embedded features of the sample decision variables after the sample vehicles are aggregated are concatenated with the features of the sample vehicles to determine the target embedded features of each of the sample vehicles.

6. The determination method according to claim 1, characterized in that: The repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criterion are used to process the destroyed initial solution result to obtain a new solution result, and the performance score and weight of the corresponding operator are dynamically updated to enter the next iteration until the target solution result of the mixed integer linear programming model is determined when the number of iterations is reached, including: Repairing the destroyed initial solution result based on the roulette wheel selection repair operator to determine the latest solution result of the mixed integer linear programming model; The simulated annealing algorithm criterion determines whether to accept the latest solution result of the mixed integer linear programming model with a preset probability based on the probability between the latest solution result and the initial solution result, and if accepted, updates the initial solution result of the mixed integer linear programming model to the latest solution result; if not accepted, does not update the initial solution result; The weights of the destruction operator and the repair operator are dynamically adjusted based on the latest solution result, and the mixed integer linear programming model is continuously iteratively solved based on the adjusted destruction operator and the adjusted repair operator to determine the target solution result.

7. The determination method according to claim 1, characterized in that: The roulette algorithm selects a destruction operator other than the GCN removal operator, and destroys the initial solution result in the following way: Selecting the vehicle execution path with the highest cost from the initial solution results, and removing all the aircraft tasks on the vehicle execution path; Alternatively, the similarities between the aircraft tasks are determined based on the attributes of the aircraft tasks, and the aircraft tasks corresponding to the highest similarities are removed from the initial solution results based on the similarities, so as to destroy the initial solution results.

8. A device for determining the dispatch of flight ground service vehicles, characterized in that: The determining device comprises: An initial solution generation module is used to solve the mixed integer linear programming model of the target airport based on a greedy algorithm to determine the initial solution result of the mixed integer linear programming model; wherein the mixed integer linear programming model is constructed from the multi-garage ground service flight support vehicle scheduling problem; the initial solution result is the vehicle execution path for allocating the corresponding garage to each of the aircraft tasks and each vehicle arriving at the aircraft parking position corresponding to different aircraft tasks; A destruction operator determination module is used to select a destruction operator based on a roulette algorithm. If the GCN removal operator is selected, the feature information of the initial solution result is input into the graph neural network model, the vehicle node is predicted, the target vehicle node is predicted, and the aircraft mission associated with the target vehicle node is removed to destroy the initial solution result; wherein the graph neural network model is obtained through supervised learning training; The target solution generation module is used to process the destroyed initial solution result based on the repair operator in the adaptive large neighborhood search algorithm and the simulated annealing algorithm criteria to obtain a new solution result, and dynamically update the performance score and weight of the corresponding operator to enter the next iteration until the target solution result of the mixed integer linear programming model is determined when the number of iterations is reached.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are run by the processor, the steps of the method for determining the scheduling of flight ground service support vehicles as described in any one of claims 1 to 7 are executed.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for determining the scheduling of flight ground service vehicles as described in any one of claims 1 to 7 are executed.

Images