Training method of gate position distribution model and gate position distribution method

Abstract

The invention belongs to the technical field of intelligent scheduling, and particularly relates to a training method of a gate position distribution model and a gate position distribution method. The training method of the parking space distribution model comprises the following steps: determining the similarity of parking space distribution problems of a first airport and a second airport; determining a model migration strategy based on the similarity of gate position distribution problems; and according to the model migration strategy and the historical parking space distribution data of the second airport, training the first parking space distribution model of the first airport to obtain a second parking space distribution model suitable for the second airport. According to the method, the distribution model can automatically adapt to airport operation changes, the construction complexity of the distribution model is reduced, and the universality of the distribution model is improved. The invention further provides a gate position distribution method, the method is based on the gate position distribution model obtained through training of the training method of the gate position distribution model, dynamic response is conducted on the gate position distribution problem of the airport, and efficient distribution of gate position resources is achieved.

Description

Technical Field

[0001] The present application belongs to the field of intelligent scheduling technology, and specifically relates to a training method for a parking space allocation model and a parking space allocation method. Background Art

[0002] Airport parking stand allocation is a key issue in aviation operations management, involving the rational allocation of arriving and departing flights to the airport's parking stands. Effective parking stand allocation can improve airport operational efficiency, reduce flight delays, optimize ground services, and enhance the overall passenger experience.

[0003] Conventional technologies often use mathematical optimization models, heuristic algorithms, and artificial intelligence techniques to solve the airport parking stand allocation problem. However, existing parking stand allocation models must be independently developed for each airport, making their development cumbersome and, in turn, complex. Summary of the Invention

[0004] The present application provides a training method for a parking stand allocation model and a parking stand allocation method to solve the technical problem that the process of independently constructing a parking stand allocation model for each airport is too cumbersome and easily leads to complex model construction in traditional parking stand allocation models.

[0005] In a first aspect, the present application provides a method for training an aircraft parking stand allocation model, comprising:

[0006] Determine the similarity of the parking slot allocation problems at the first airport and the second airport;

[0007] Determining a model migration strategy based on the similarity of the parking slot allocation problem;

[0008] According to the model migration strategy and the historical parking stand allocation data of the second airport, the first parking stand allocation model of the first airport is trained to obtain a second parking stand allocation model suitable for the second airport, wherein the first parking stand allocation model is a trained model, the input of the first parking stand allocation model is parking stand environment status information, and the output of the first parking stand allocation model is a parking stand allocation action.

[0009] In one possible implementation, determining the similarity between the parking slot allocation problems of the first airport and the second airport includes:

[0010] Obtaining multimodal data related to the parking stand allocation problem at each of the first airport and the second airport, wherein the multimodal data includes multiple items of the following data: airport images, airport layout topology maps, parking stand-related facility data, flight dynamic data, environmental data, and parking stand allocation optimization objectives;

[0011] A similarity of the parking stand allocation problem is determined based on the multimodal data of each of the first airport and the second airport.

[0012] In one possible implementation, determining the similarity of the parking stand allocation problem based on the multimodal data of the first airport and the second airport includes:

[0013] Inputting the multimodal data of each of the first airport and the second airport into a large model, extracting initial features of each item of data in the multimodal data using the large model, mapping the initial features of each item of data into a unified feature space, obtaining feature vectors corresponding to each item of data of the first airport and the second airport, and determining attention weights between different items of data of the first airport and the second airport;

[0014] Determining multimodal fusion features of the first airport and the second airport based on the feature vectors of the respective items of data corresponding to the first airport and the second airport, and the attention weights between the different items of data of the first airport and the second airport;

[0015] Determine the similarity of the parking space allocation problem based on the multimodal fusion features of the first airport and the second airport.

[0016] In one possible implementation, determining the similarity of the parking stand allocation problem based on the multimodal data of the first airport and the second airport includes:

[0017] determining a similarity between each item of multimodal data of the first airport and the second airport;

[0018] The similarity of the parking space allocation problem is determined according to the weight of each data item and the similarity of each data item.

[0019] In one possible implementation, determining a model migration strategy based on the similarity of the parking slot allocation problem includes:

[0020] When the similarity of the parking space allocation problem is greater than or equal to a first threshold, determining that the model migration strategy is state alignment and model output layer adjustment;

[0021] The step of training the first parking stand allocation model for the first airport according to the model migration strategy and the historical parking stand allocation data of the second airport to obtain a second parking stand allocation model suitable for the second airport includes:

[0022] When the model migration strategy is state alignment and adjustment of the model output layer, freezing the parameters of the feature extraction layer and the intermediate hidden layer of the first parking slot allocation model;

[0023] The state space applicable to the second airport is mapped to the state space of the first parking stand allocation model. The first parking stand allocation model is trained using historical parking stand allocation data of the second airport to adjust output layer parameters of the first parking stand allocation model to obtain the second parking stand allocation model.

[0024] In one possible implementation, determining a model migration strategy based on the similarity of the parking slot allocation problem includes:

[0025] When the similarity of the parking space allocation problem is greater than or equal to a second threshold and less than a first threshold, determining that the model migration strategy is state alignment and model layer adjustment;

[0026] The step of training the first parking stand allocation model for the first airport according to the model migration strategy and the historical parking stand allocation data of the second airport to obtain a second parking stand allocation model suitable for the second airport includes:

[0027] When the model migration strategy is state alignment and model hierarchical adjustment, the state space applicable to the second airport is mapped to the state space of the first parking stand allocation model, and the first parking stand allocation model is trained in stages using historical parking stand allocation data of the second airport to obtain the second parking stand allocation model;

[0028] In the first stage, parameters of the feature extraction layer and the intermediate hidden layer of the first parking space allocation model are frozen, and parameters of the output layer of the first parking space allocation model are adjusted;

[0029] In the second stage, parameters of a feature extraction layer of the first parking stand allocation model are frozen, and parameters of an intermediate hidden layer and an output layer of the first parking stand allocation model are adjusted;

[0030] In the third stage, the parameters of each layer of the first parking stand allocation model are adjusted.

[0031] In one possible implementation, determining a model migration strategy based on the similarity of the parking slot allocation problem includes:

[0032] When the similarity of the parking space allocation problem is greater than or equal to a third threshold and less than a second threshold, determining that the model migration strategy is to add a new model branch, and hierarchically adjusting the new model branch and the original model branch;

[0033] The step of training the first parking stand allocation model for the first airport according to the model migration strategy and the historical parking stand allocation data of the second airport to obtain a second parking stand allocation model suitable for the second airport includes:

[0034] In a case where the model migration strategy is to add a new model branch and perform hierarchical adjustment on the new model branch and the original model branch, a dedicated model branch is added on the basis of the original model branch of the first parking stand allocation model, wherein the dedicated model branch is used to process the state space and action space having differences between the first airport and the second airport;

[0035] The first parking stand allocation model with the dedicated model branch added is trained in stages to obtain the second parking stand allocation model;

[0036] In the first stage, the original model branch of the first parking stand allocation model is frozen, and the parameters of the dedicated model branch are adjusted;

[0037] In the second stage, the feature extraction layer of the original model branch of the first parking stand allocation model is frozen, the parameters of the intermediate hidden layer and the output layer of the original model branch are adjusted, and the parameters of the dedicated model branch are adjusted;

[0038] In the third stage, the parameters of the original model branch and the dedicated model branch are adjusted.

[0039] In a second aspect, the present application provides a parking stand allocation method, comprising:

[0040] Acquiring parking stand environment status information of the second airport, wherein the parking stand environment status information includes parking stand information, taxiway information, runway information, and unassigned flight status information;

[0041] The parking stand environmental status information of the second airport is input into a second parking stand allocation model to obtain a parking stand allocation action output by the second parking stand allocation model, wherein the second parking stand allocation model is trained using the parking stand allocation model training method described in the first aspect and various possible implementation methods of the first aspect.

[0042] In a third aspect, the present application provides a training device for an aircraft parking stand allocation model, comprising:

[0043] The determining module is used to determine the similarity between the parking space allocation problems of the first airport and the second airport.

[0044] The determination module is further configured to determine a model migration strategy based on the similarity of the parking space allocation problem.

[0045] and a processing module, configured to train a first parking stand allocation model for the first airport according to the model migration strategy and historical parking stand allocation data for the second airport to obtain a second parking stand allocation model suitable for the second airport, wherein the first parking stand allocation model is a trained model, the input of the first parking stand allocation model is parking stand environment status information, and the output of the first parking stand allocation model is a parking stand allocation action.

[0046] In a fourth aspect, the present application provides a parking stand allocation device, comprising:

[0047] The acquisition module is used to acquire parking stand environment status information of the second airport, where the parking stand environment status information includes parking stand information, taxiway information, runway information, and unassigned flight status information.

[0048] a processing module, configured to input the parking stand environmental status information of the second airport into a second parking stand allocation model, and obtain a parking stand allocation action output by the second parking stand allocation model, wherein the second parking stand allocation model is trained using the parking stand allocation model training method described in the first aspect and various possible implementations of the first aspect.

[0049] In a fifth aspect, the present application provides an electronic device, including:

[0050] a processor, and a memory communicatively connected to the processor;

[0051] The memory stores computer-executable instructions;

[0052] The processor executes the computer-executable instructions stored in the memory to implement the training method of the parking stand allocation model as described in the first aspect and various possible implementations of the first aspect, or the parking stand allocation method as described in the second aspect and various possible implementations of the second aspect.

[0053] In a sixth aspect, the present application provides a computer storage medium having computer execution instructions stored thereon, and the computer execution instructions are executed by a processor to implement the training method of the parking stand allocation model as described in the first aspect and various possible implementations of the first aspect, or the parking stand allocation method as described in the second aspect and various possible implementations of the second aspect.

[0054] In a seventh aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the training method for the parking stand allocation model as described in the first aspect and various possible implementations of the first aspect, or the parking stand allocation method as described in the second aspect and various possible implementations of the second aspect.

[0055] The training method of the parking stand allocation model provided in the present application determines the similarity of the parking stand allocation problems of the first airport and the second airport; based on the similarity of the parking stand allocation problems, determines the model migration strategy; and trains the first parking stand allocation model of the first airport according to the model migration strategy and the historical parking stand allocation data of the second airport to obtain a second parking stand allocation model suitable for the second airport. While maintaining the personalized adaptability of the allocation model, this method significantly reduces the model development and maintenance costs, enables the allocation model to automatically adapt to changes in airport operations, reduces the complexity of constructing allocation models between different airports, and improves the versatility of the allocation model. The present application also provides a parking stand allocation method, which dynamically responds to the parking stand allocation problem of the corresponding airport based on the parking stand allocation model trained by the parking stand allocation model training method, further optimizes and improves the accuracy and effectiveness of parking stand allocation, and ensures that the allocation strategy is always synchronized with the airport operation status. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0057] Figure 1 This is a schematic diagram of a scenario of the parking stand allocation method provided by this application;

[0058] Figure 2 This is a flow chart of the training method for the parking space allocation model provided by this application. Figure 1 ;

[0059] Figure 3 This is a flow chart of the training method for the parking space allocation model provided by this application. Figure 2 ;

[0060] Figure 4 It is a flowchart of the parking space allocation method provided by this application;

[0061] Figure 5 It is a structural diagram of the training device of the parking stand allocation model provided by this application;

[0062] Figure 6 It is a structural diagram of the parking space allocation device provided by this application;

[0063] Figure 7 It is a structural diagram of the electronic device provided in this application.

[0064] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0065] To make the objectives, technical solutions, and advantages of this application more clear, the technical solutions in this application will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0066] The terms "first," "second," "third," "fourth," and so forth (if any) in the description and claims of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequential sequence. It should be understood that the terms used in this manner are interchangeable where appropriate, such that the embodiments of the present invention described herein can, for example, be practiced in an order other than that illustrated or described herein.

[0067] In the embodiments of this application, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described in this application as "exemplary" or "for example" should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0068] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards, and corresponding operation entrances must be provided for users to choose to authorize or refuse.

[0069] First, the terms involved in this application are explained.

[0070] Transformer: It is a deep learning architecture based on the self-attention mechanism, which is used to process sequential data (such as text, time series, etc.); its core idea is to dynamically calculate the association weights between different parts of the input data through the attention mechanism, replacing traditional recurrent neural networks or convolutional neural networks.

[0071] Jaccard similarity: used to measure the similarity between two sets, defined as the size of the intersection of the two sets divided by the size of the union.

[0072] Cosine similarity: Cosine similarity measures the similarity between two vectors by calculating the cosine of the angle between them.

[0073] Euclidean distance: Euclidean distance is a measure of the straight-line distance between two points in Euclidean space.

[0074] Weighted Euclidean distance: Weighted Euclidean distance is an extension of Euclidean distance that allows different weights to be given to each dimension.

[0075] Structural similarity: It is an indicator to measure the similarity between two images, taking into account brightness, contrast and structural information; the value range is between -1 and 1, and the larger the value, the higher the similarity.

[0076] Ratio similarity: Usually used to measure whether the proportional relationship between two numbers or vectors is similar; it can be obtained by calculating the ratio of corresponding elements of two vectors and comparing these ratios.

[0077] Model fine-tuning: It is a commonly used technique in deep learning, mainly used to optimize the performance of pre-trained models on specific tasks. The core idea of ​​fine-tuning is to use the model pre-trained on a large-scale dataset as a basis, and adjust the model parameters by further training on the dataset of a specific task, thereby improving the performance on the task.

[0078] Airport parking stand allocation is a key issue in aviation operations management, involving the rational allocation of arriving and departing flights to the airport's parking stands. Effective parking stand allocation can improve airport operational efficiency, reduce flight delays, optimize ground services, and enhance the overall passenger experience.

[0079] In the existing technology, airport parking space allocation is a complex and dynamic optimization problem involving multiple constraints and objectives; its complexity is reflected in the need to consider multiple factors such as flight time, aircraft type, parking space restrictions, airline preferences, and safety requirements.

[0080] Conventional technologies often use mathematical optimization models, heuristic algorithms, and artificial intelligence techniques to solve the airport parking stand allocation problem. However, existing parking stand allocation models must be independently developed for each airport, making their development cumbersome and, in turn, complex.

[0081] To address the above issues, the present application provides a parking stand allocation model training method and a parking stand allocation method.

[0082] First, the implementation scenarios involved in this application are described.

[0083] Figure 1 This is a schematic diagram of the parking space allocation method provided by this application. Figure 1As shown, multiple aircraft 101 are communicatively connected to an airport ground management system 102, and the multiple aircraft 101 include but are not limited to aircraft in flight, aircraft parked at parking stands, and aircraft waiting to take off. The airport ground management system 102 is provided with a parking stand dynamic management and control module, which can be trained for the parking stand allocation model required by the current airport, and based on the parking stand allocation model, the parking stand problems of multiple aircraft in the airport are allocated in real time.

[0084] Exemplarily, multiple aircraft 101 include: aircraft A, aircraft B and aircraft C, and aircraft A, aircraft B and aircraft C can all communicate with the airport ground management system 102; the parking stand dynamic management and control module in the airport ground management system 102, based on the real-time acquired dynamic data information of aircraft A, aircraft B and aircraft C, specifically including: the time to take-off, whether the runway is occupied, whether the taxiway is occupied, and whether the parking stand has been parked, inputs the obtained information into the parking stand dynamic management and control module, and the parking stand dynamic management and control module outputs the parking stand allocation information for aircraft A, aircraft B and aircraft C, for example, aircraft A drives from stand 1-1 to taxiway 1, aircraft B drives from taxiway 2 to stand 2-1, and aircraft C drives from taxiway 3 to runway 2.

[0085] This application provides a training method for a parking stand allocation model. Based on an existing parking stand allocation model, this method determines the similarity between the parking stand allocation problems of the airport for which an allocation model is currently being constructed and the airport for which an existing parking stand allocation model is already constructed. Based on the different degrees of similarity, corresponding model migration strategies are determined. The existing parking stand allocation model is then migrated according to the model migration strategies corresponding to the different degrees of similarity, thereby achieving rapid migration of the corresponding model. This method optimizes the allocation model construction strategy, improves model construction efficiency, and reduces the complexity of allocation model construction.

[0086] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0087] Figure 2 This is a flow chart of the training method for the parking space allocation model provided by this application. Figure 1 The execution subject of this embodiment may be, for example, an airport ground management system equipped with a dynamic parking space control module. Figure 2 As shown, the training method of the parking space allocation model provided in this embodiment includes:

[0088] S201: Determine the similarity between the parking space allocation problems of the first airport and the second airport.

[0089] The similarity is used to describe the similarity between the first and second airports in terms of characteristics, challenges, and solutions for parking slot allocation.

[0090] It can be understood that the airport parking stand allocation problem refers to the reasonable allocation of limited parking stand resources to each flight during the airport operation process based on factors such as flight arrival and departure time, aircraft type, airline, special needs (such as jet bridge requirements), so as to ensure that the flights can operate safely and efficiently; although the parking stand allocation problems of different airports have their own characteristics due to differences in airport scale, layout and operating rules, there are similarities in core issues, goals, constraints and data processing; specifically, the parking stand allocation problems of different airports all need to allocate parking locations for different flights under limited parking stand resources, and must comprehensively consider core factors such as flight schedules, aircraft types, and airline needs. For example, both hub airports and regional airports need to resolve the problem of parking stand conflicts during periods of high flight density; the parking stand allocation problems of different airports all pursue improving operational efficiency (such as reducing taxi time), ensuring safety (avoiding aircraft conflicts), improving service quality (such as prioritizing jet bridge allocation) and reducing costs (optimizing resource utilization).

[0091] The parking stand allocation problems of the first airport and the second airport are determined respectively, and data processing is performed on the parking stand allocation problem of the first airport and the parking stand allocation problem of the second airport respectively, corresponding problem information is extracted according to different dimensions, and based on the problem information of multiple dimensions, the similarity of the corresponding dimensions is calculated respectively.

[0092] For example, the first airport may be Airport A, and the second airport may be Airport B. The parking space allocation problems of Airport A include: Airport A is a large hub airport with high flight density, complex route network and multiple types of parking spaces (close-in parking spaces, remote parking spaces, combined parking spaces, etc.). It needs to improve parking space utilization, reduce aircraft taxiing time, ensure safety intervals, optimize passenger experience (such as jet bridge allocation), and improve transfer efficiency; the parking space allocation problems of Airport B include: Airport B is a regional airport with relatively few flights, but it needs to take into account the needs of domestic and international flights, improve parking space utilization, reduce aircraft taxiing time, ensure safety intervals, optimize passenger experience (such as jet bridge allocation), and improve transfer efficiency. The system can ensure safe intervals and optimize passenger experience (such as jet bridge allocation) to ensure the normal operation of flights; the data of the parking stand allocation problem of Airport A and the parking stand allocation problem of Airport B are processed separately, and the corresponding problem information is extracted according to different dimensions, and the corresponding similarity is calculated based on different dimensions. The dimensions of the corresponding problem information include but are not limited to: the number and type of parking stands, the number and type of flights, the distribution of aircraft models, the adaptation requirements of aircraft models, the exclusive requirements of airlines, and the priority of jet bridge allocation; specifically, the similarity corresponding to different dimensions includes: "the number and type of parking stands: Airport A has 200 (50 jet bridges), Airport B has 120 (30 jet bridges), the calculated similarity of the number of parking spaces (such as cosine similarity) is 0.2", "Flight volume and type: Airport A has 1,500 flights (50% international), Airport B has 800 flights (30% international), and the calculated similarity of flight volume and type (such as weighted Euclidean distance) is 0.25", "Aircraft type adaptation requirements: Airport A requires Class E for wide-body aircraft, and Airport B requires Class E for wide-body aircraft. The parking space constraints are completely consistent, and the corresponding similarity is calculated to be 10.15", "Safety interval rules: Airport A has a 5-minute interval between flights, and Airport B has a 5-minute interval between flights. The constraints are complete. The similarity between the two problems is completely consistent, and the calculated similarity is 10.15", "Airline exclusive parking stands: Airport A has 10 exclusive parking stands, and Airport B has 5 exclusive parking stands. The calculated similarity between the number of exclusive parking stands and the airline is 0.1", "Jet bridge allocation priority: Airport A is: international flights>wide-body aircraft>domestic flights, and Airport B is: international flights>wide-body aircraft>domestic flights. The priority rules are completely consistent, and the calculated similarity is 10.15"; Therefore, based on the similarities calculated from multiple problem dimensions, a weighted calculation method can be used to determine that the similarity between the parking stand allocation problems of Airport A and Airport B is 0.81. This application does not impose any special restrictions on the parking stand allocation problem, and this application does not impose any special restrictions on the method for calculating the similarity of parking stand allocation problems at different airports.

[0093] S202: Determine a model migration strategy based on the similarity of the parking slot allocation problem.

[0094] S203: Training the first parking stand allocation model for the first airport according to the model migration strategy and the historical parking stand allocation data of the second airport to obtain a second parking stand allocation model suitable for the second airport.

[0095] The first parking stand allocation model is a trained model; the input of the first parking stand allocation model is parking stand environment status information, and the output is a parking stand allocation action.

[0096] As you can understand, model transfer strategy usually refers to the method of applying an already trained model to another related but different task or dataset in the field of machine learning and deep learning. Model transfer strategy is suitable for situations where the amount of data for the target task is small, the annotation cost is high, or the cost of training the model from scratch is too high. By leveraging the knowledge of pre-trained models, the performance of new tasks can be effectively improved, and training time and data requirements can be reduced.

[0097] In the embodiment of the present application, since different airports have significant differences in structural and operational characteristics in the parking stand allocation problem, and these differences are reflected in multiple key factors, such as flight density, number of parking stands, airline operating model, and whether there are dedicated parking stands; the differences in key factors lead to a certain degree of inconsistency in the modeling methods, constraints, and optimization goals of the parking stand allocation problem at various airports. If the model trained at a certain airport is directly migrated to another airport, it may lead to problems such as poor adaptability, inaccurate prediction, or decreased scheduling efficiency; therefore, the similarity of the parking stand allocation problem can be divided into different similarity levels according to the degree of similarity from high to low, and different similarity levels adopt corresponding model migration strategies.

[0098] Based on the similarity of the parking stand allocation problem, the similarity level corresponding to the currently obtained similarity is determined. According to the model migration strategy corresponding to the similarity level, combined with the historical parking stand allocation data of the second airport, the first parking stand allocation model of the first airport is trained to obtain the second parking stand allocation model suitable for the second airport.

[0099] Exemplarily, the similarity levels determined based on the similarity of the parking stand allocation problem include: high similarity (similarity greater than or equal to 0.8), moderate similarity (similarity greater than or equal to 0.5 but less than 0.8) and low similarity (similarity less than 0.5); when airport A already has a corresponding parking stand allocation model and it is necessary to construct a parking stand allocation model corresponding to airport B, if the similarity between the parking stand allocation problems of airport A and airport B is 0.81, 0.81>0.8, it indicates that the similarity between the parking stand allocation problems of airport A and airport B is high, and the corresponding similarity level is highly similar, and a similar optimization algorithm or system architecture can be used. At this time, the parking stand data information of airport B can be input into the parking stand allocation model of airport A, and the parking stand data information of airport B can be used to fine-tune the parking stand allocation model corresponding to airport A, and the fine-tuned model can be used as the parking stand allocation model of airport B; if the similarity between the parking stand allocation problems of airport A and airport B is 0.6, 0.5<0.6 If the similarity is <0.8, it indicates that the similarity between the parking stand allocation problems of Airport A and Airport B is moderate, and the corresponding similarity level is moderate. Some constraints or objective functions need to be adjusted in a targeted manner. In this case, the differences between Airport A and Airport B in constraints (such as airline-exclusive parking rules) and objective functions (such as jet bridge allocation priority) should be analyzed, and the constraint parameters or weights in the algorithm should be adjusted in a targeted manner. The core algorithm framework (such as rule-based allocation logic) should be migrated, but some modules (such as dynamic adjustment rules) need to be retrained or adjusted. The adjusted parking stand allocation model for Airport A should then be trained based on the parking stand data information of Airport B, and the trained model should be used as the parking stand allocation model for Airport B. If the similarity between the parking stand allocation problems of Airport A and Airport B is 0.3, and 0.3 is <0.5, it indicates that the similarity between the parking stand allocation problems of Airport A and Airport B is low, and the corresponding similarity level is low. In this case, the similarity between the two airports is too low to perform model migration, and independent allocation strategies need to be designed.

[0100] In the embodiment of the present application, when the similarity between parking stand allocation models at different airports is low, it is not possible to migrate the model based on the existing parking stand allocation model. In other words, the training method for the parking stand allocation model provided in the present application is applicable to situations with high or medium similarity, but not to situations with too low similarity. Such situations with too low similarity can be, for example, situations with a similarity less than 0.5. In addition, the similarity value corresponding to such situations with too low similarity can be dynamically set for different usage scenarios. The present application does not impose any special restrictions on the specific similarity value corresponding to such situations with too low similarity.

[0101] The parking stand allocation model training method provided in this embodiment determines the similarity between the parking stand allocation problems at a first airport and a second airport; based on this similarity, a model migration strategy is determined; and a first parking stand allocation model for the first airport is trained using the model migration strategy and the historical parking stand allocation data of the second airport to obtain a second parking stand allocation model suitable for the second airport. This method improves the efficiency of allocation model construction, enables the allocation model to automatically adapt to changes in airport operations, and reduces the complexity of building allocation models between different airports.

[0102] Figure 3 This is a flow chart of the training method for the parking space allocation model provided by this application. Figure 2 .like Figure 3 As shown, this embodiment Figure 2 Based on the embodiment, the training method of the parking stand allocation model is described in detail. The training method of the parking stand allocation model shown in this embodiment includes:

[0103] S301: Obtain multimodal data related to the parking stand allocation problem of each of the first airport and the second airport.

[0104] S302: Determine the similarity of the parking stand allocation problem based on the multimodal data of the first airport and the second airport.

[0105] Among them, multimodal data is used to indicate an information collection of data of multiple types or sources.

[0106] It can be understood that multimodal data includes airport-related data of multiple types or dimensions, and each type or dimension of data represents a specific "modality", that is, data of different dimensions, that is, data of different modes; therefore, based on multimodal data, it is possible to analyze and process the parking space allocation problem of the first airport and the second airport from different dimensions.

[0107] In some embodiments, based on the multimodal data of each of the first airport and the second airport, the similarity of the parking stand allocation problems between the corresponding airports is determined, including: inputting the multimodal data of each of the first airport and the second airport into a large model, extracting the initial features of each data in the multimodal data through the large model, and mapping the initial features of each data to a unified feature space to obtain the feature vectors of each of the data corresponding to the first airport and the second airport, and determining the attention weights between different items of data of each of the first airport and the second airport; based on the feature vectors of each of the data corresponding to the first airport and the second airport, and the attention weights between different items of data of each of the first airport and the second airport, determining the multimodal fusion features of each of the first airport and the second airport; and determining the similarity of the parking stand allocation problems according to the multimodal fusion features of each of the first airport and the second airport.

[0108] Among them, the initial features are used to indicate the basic features extracted from the parking space allocation problem, and the attention weight is used to measure the contribution of different modal data of the same parking space allocation problem to the final similarity calculation; the large model can be, for example, a multimodal pre-trained model based on the Transformer architecture.

[0109] In the embodiments of the present application, a large model refers to a deep learning model with superb generalization capabilities that is trained through massive amounts of data and enormous computing power, such as GPT (Generative Pre-trained Transformer), BERT (Bidirectional Encoder Representations from Transformers), and Transformer. In the scenario of airport parking stand allocation, the large model has multimodal processing capabilities and can simultaneously parse heterogeneous data such as numerical values ​​(flight volume), text (rules), spatiotemporal data (flight dynamics), and images (airport layout maps). It can also map different multimodal data corresponding to different airports into a unified feature space, that is, convert various types of data from different airports into comparable standardized vectors (such as 1024-dimensional feature vectors). In addition, the large model also has a self-attention mechanism that can automatically identify key feature associations such as "aircraft model distribution-parking stand size" and dynamically adjust the corresponding attention weights.

[0110] Multimodal data related to the parking stand allocation problem of each of the first and second airports are obtained. A large model set in the airport ground management system is called and the multimodal data of each of the first and second airports are input into the large model. The large model is used to extract initial features of data of different dimensions in the multimodal data of different airports, and the initial features of each data item are mapped to a unified feature space to obtain feature vectors of each data item corresponding to the first and second airports. Based on the self-attention mechanism in the large model, the attention weights between different modal data items within the same airport are calculated to reflect the relative importance of each modal feature in the task. Based on the attention weights between different items of data from the first airport, the feature vectors of each item of data corresponding to the first airport are fused to obtain the multimodal fusion features of the first airport. Based on the attention weights between different items of data from the second airport, the feature vectors of each item of data corresponding to the second airport are fused to obtain the multimodal fusion features of the second airport. Based on the multimodal fusion features of each of the first and second airports, the similarity of the corresponding multimodal fusion features of the two airports is calculated, that is, the similarity of the parking stand allocation problems of the first and second airports is determined.

[0111] For example, the currently available multimodal data of Airport A and Airport B include but are not limited to: aircraft model adaptation requirements, flight operation data (including flight volume and type), parking space resource configuration information (including parking space number and type), meteorological data, historical scheduling records, etc.; for Airport A and Airport B, the multimodal data of the corresponding airports are respectively input into the multimodal pre-training model based on the Transformer architecture, and the model extracts features from the multimodal data of the corresponding airports to obtain initial features (such as aircraft model adaptation ratio, flight volume, number of parking spaces, etc.), and then maps the initial features of each data to a unified feature space to obtain feature vectors of Airport A and Airport B respectively; through the self-attention mechanism, the large model can dynamically focus on key data (such as flight peak hours and parking space adaptability) to optimize resource allocation. Therefore, through the self-attention mechanism, the multimodal Pre-train the model and calculate the attention weights between different modal data within the same airport. Specifically: the flight density data of airport A may obtain a higher weight because it has a greater impact on the allocation of parking spaces; meteorological data may obtain a lower weight because it has a smaller direct impact on the allocation of parking spaces; based on the attention weights of the data in the corresponding airport, the feature vectors of the modal data of airport A and airport B are fused respectively to obtain the multimodal fusion features corresponding to airports A and B. For example, the multimodal fusion features of airport A may focus more on flight density and aircraft model adaptation, while the multimodal fusion features of airport B may focus more on parking space types and historical scheduling records; based on the multimodal fusion features of airport A and airport B, the similarity between the two is calculated. For the similarity calculation between the two, for example, cosine similarity, Euclidean distance and other methods can be used to measure the similarity of the multimodal fusion features of the two airports.

[0112] In some embodiments, determining the similarity of parking stand allocation problems between the corresponding airports based on the multimodal data of each of the first airport and the second airport also includes: determining the similarity of each data in the multimodal data of each of the first airport and the second airport; and determining the similarity of the parking stand allocation problems based on the weight of each data of the similarity of each data.

[0113] It is understandable that, during the construction of the parking space allocation model at the second airport, corresponding weight information is pre-set for data of different modalities. This application does not impose any particular restrictions on the method for pre-setting weight information for data of different modalities.

[0114] For example, the dimensions corresponding to the currently acquired multimodal data include but are not limited to: aircraft model adaptation requirements, flight operation data, parking space resource configuration information, meteorological data and historical scheduling records, and the corresponding preset weights are: 0.2, 0.3, 0.25, 0.1, and 0.15 respectively; for data of different modes, the similarity of the same modal data corresponding to different airports is calculated respectively, wherein the degree of overlap of the model list and the difference in the proportion of jet bridge parking spaces can be calculated using Jaccard similarity or cosine similarity; the difference in flight volume and flight type proportion can be calculated using Euclidean distance or cosine similarity; the number of parking spaces can be calculated using structural similarity or proportion similarity. The differences in quantity and type can be calculated; the differences in meteorological data can be calculated using Euclidean distance or proportional similarity; the differences in historical scheduling records can be calculated using proportional similarity; the similarities of different modal data between airports A and B can be: aircraft adaptation requirements -0.7, flight operation data -0.928, parking space resource configuration information -0.867, meteorological data -0.708, historical scheduling records -0.84; combined with the preset weights of different modal data, the similarity of the parking space allocation problems of airports A and B can be calculated to be 0.83195, that is, the similarity between airports A and B in the parking space allocation problem is about 83.2%.

[0115] S303: Determine a model migration strategy based on the similarity of the parking slot allocation problem.

[0116] S304: Training the first parking stand allocation model for the first airport according to the model migration strategy and the historical parking stand allocation data of the second airport to obtain a second parking stand allocation model suitable for the second airport.

[0117] When there is a constructed parking stand allocation model for the first airport and a parking stand allocation model corresponding to the second airport is constructed this time, based on the similarity of the parking stand allocation problems of the first and second airports, a model migration strategy is determined for constructing the parking stand allocation model of the second airport based on the parking stand allocation model of the first airport. According to the model migration strategy, combined with the historical parking stand allocation data of the second airport, the first parking stand allocation model of the first airport is retrained to obtain a second parking stand allocation model suitable for the second airport.

[0118] In some embodiments, when the similarity of the parking stand allocation problem is greater than or equal to a first threshold, the model migration strategy is determined to be state alignment and model output layer adjustment; when the model migration strategy is state alignment and adjustment of the model output layer, the parameters of the feature extraction layer and the intermediate hidden layer of the first parking stand allocation model are frozen; the state space applicable to the second airport is mapped to the state space of the first parking stand allocation model, and the historical parking stand allocation data of the second airport is used to train the first parking stand allocation model to adjust the output layer parameters of the first parking stand allocation model to obtain a second parking stand allocation model.

[0119] Among them, the output layer is used to output the parking stand allocation action, the feature extraction layer is used to extract the basic features in the parking stand allocation problem, and the middle hidden layer is used to determine the parking stand allocation action. The historical parking stand allocation data includes: historical parking stand environmental status information and historical parking stand allocation actions of the second airport; the first threshold can be, for example, 0.8.

[0120] It can be understood that the state space is used to reflect the operating status of the airport, and the state space includes the input information required by the first parking stand allocation model. That is, the data information included in the state space is consistent with the parking stand environment state information corresponding to the first airport. Since the second parking stand allocation model is obtained by model migration based on the first parking stand allocation model, and the first parking stand allocation model is designed based on the state space of the first airport, if the historical parking stand allocation data of the second airport is directly input into the first parking stand allocation model, the first parking stand allocation model will not be able to correctly understand or process the historical parking stand allocation data of the second airport due to the possible different definition, structure or dimension of the state space. Therefore, it is necessary to convert the parking stand environment state information of the second airport obtained at that time into the parking stand environment state information of the first airport, that is, to map the state space applicable to the second airport to the state space of the first parking stand allocation model.

[0121] For example, if the similarity of the currently determined parking stand allocation problems of airport A and airport B is 0.81, and 0.81>0.8, it indicates that the parking stand allocation problems of airport A and airport B have a high overlap in the state space. At this time, the model migration strategy is determined to align the states of airport A and airport B and only adjust the output layer of the first parking stand allocation model; when the model migration strategy is state alignment and adjustment of the model output layer, the parameters of the feature extraction layer and the intermediate hidden layer of the first parking stand allocation model are frozen; the state space applicable to the second airport is mapped to the state space of the first parking stand allocation model, and the historical parking stand allocation data of the second airport is used to train the first parking stand allocation model to adjust the output layer parameters of the first parking stand allocation model, and the trained model is determined as the second parking stand allocation model; specifically, when the two airports have similar flight structures and resource allocations, the model fine-tuning method can be directly used to make small adjustments to the output layer parameters of the first parking stand allocation model of airport A to quickly adapt to the data distribution of airport B.

[0122] In some embodiments, when the similarity of the parking stand allocation problem is greater than or equal to a second threshold and less than a first threshold, the model migration strategy is determined to be state alignment and model hierarchical adjustment; when the model migration strategy is state alignment and model hierarchical adjustment, the state space applicable to the second airport is mapped to the state space of the first parking stand allocation model, and the historical parking stand allocation data of the second airport is used to train the first parking stand allocation model in stages to obtain a second parking stand allocation model.

[0123] The second threshold may be 0.6, for example.

[0124] In an embodiment of the present application, model hierarchical adjustment refers to adjusting parameters at different levels of the first parking stand allocation model to adapt to the data set of the second airport; when the similarity of the parking stand allocation problem is greater than or equal to the second threshold and less than the first threshold, the staged training of the first parking stand allocation model specifically includes: a first stage, a second stage, and a third stage; specifically, in the first stage, the parameters of the feature extraction layer and the intermediate hidden layer of the first parking stand allocation model are frozen, and the parameters of the output layer of the first parking stand allocation model are adjusted; in the second stage, the parameters of the feature extraction layer of the first parking stand allocation model are frozen, and the parameters of the intermediate hidden layer and the output layer of the first parking stand allocation model are adjusted; in the third stage, the parameters of each layer of the first parking stand allocation model are adjusted.

[0125] For example, if the similarity of the currently determined parking stand allocation problems of airport A and airport B is 0.7, and 0.6<0.7<0.8, it indicates that the parking stand allocation problems of airport A and airport B partially overlap in the state space, and the two show certain similarities in the key features of parking stand allocation (such as the number of parking stands, flight arrival or departure time distribution, aircraft model, etc.), but there are also obvious differences. At this time, the model migration strategy is determined to be state alignment of airport A and airport B and hierarchical adjustment of the first parking stand allocation model; when the model migration strategy is state alignment and model hierarchical adjustment, the state space applicable to the second airport is mapped to the state space of the first parking stand allocation model, and the historical parking stand allocation data of the second airport is used to adjust the first parking stand allocation model. The model is trained in stages, and the model parameters are gradually adjusted to adapt to the needs of Airport B. Specifically, the staged training includes: in the first stage, the parameters of the feature extraction layer and the intermediate hidden layer of the first parking stand allocation model are frozen, and only the parameters of the output layer of the first parking stand allocation model are adjusted to enable it to make decisions based on the actual needs of the second airport; in the second stage, the parameters of the feature extraction layer of the first parking stand allocation model are frozen, and the parameters of the intermediate hidden layer and the output layer of the first parking stand allocation model are adjusted to better capture the specific characteristics and rules of the second airport; in the third stage, the parameters of all layers are unfrozen, and global adjustments are made. The model is fully trained using the historical parking stand allocation data of the second airport to optimize the model's performance on the parking stand allocation problem of Airport B, thereby obtaining the second parking stand allocation model.

[0126] In some embodiments, when the similarity of the parking stand allocation problem is greater than or equal to the third threshold and less than the second threshold, the model migration strategy is determined to be a new model branch, and the new model branch and the original model branch are layered adjusted; when the model migration strategy is a new model branch, and the new model branch and the original model branch are layered adjusted, a special model branch is added on the basis of the original model branch of the first parking stand allocation model; the first parking stand allocation model after the special model branch is added is trained in stages to obtain a second parking stand allocation model.

[0127] The dedicated model branch is used to process the state space and action space having differences between the first airport and the second airport, and the third threshold may be 0.45, for example.

[0128] It can be understood that the newly added model branch refers to the addition of one or more sub-models (i.e. branches) on the basis of the first parking space allocation model. The branch shares some or all inputs with the original model, but has independent parameters and structure, which is used to process the unique features or rules in the target task. By adding new model branches, the model can learn and adapt to new task requirements while retaining the original knowledge.

[0129] In an embodiment of the present application, when the similarity of the parking stand allocation problem is greater than or equal to the third threshold and less than the second threshold, the phased training of the first parking stand allocation model specifically includes: first-stage training, second-stage training, and third-stage training; in the first stage, the original model branch of the first parking stand allocation model is frozen, and the parameters of the dedicated model branch are adjusted; in the second stage, the feature extraction layer of the original model branch of the first parking stand allocation model is frozen, the parameters of the intermediate hidden layer and the output layer of the original model branch are adjusted, and the parameters of the dedicated model branch are adjusted; in the third stage, the parameters of the original model branch and the dedicated model branch are adjusted.

[0130] For example, if the similarity of the currently determined parking stand allocation problems of airport A and airport B is 0.5, and 0.45<0.7<0.6, it indicates that the parking stand allocation problems of airport A and airport B have little overlap in the state space, and there are obvious differences between the two in the key features of parking stand allocation (such as the number of parking stands, flight arrival or departure time distribution, aircraft model, etc.). At this time, the model migration strategy is determined to be a new model branch, and the new model branch and the original model branch are adjusted in layers; when the model migration strategy is a new model branch, and the new model branch and the original model branch are adjusted in layers, a dedicated model branch is added on the basis of the original model branch of the first parking stand allocation model, and the parameters of the dedicated model branch are initialized. Usually, part of the parameters of the original model branch can be copied or pre-trained weights can be used. Initialization is performed to accelerate the training process; the first parking stand allocation model after adding the dedicated model branch is trained in stages to obtain a second parking stand allocation model; specifically, in the first stage, the original model branch of the first parking stand allocation model is frozen, the parameters of the dedicated model branch are adjusted, and the parameters of the feature extraction layer, the intermediate hidden layer and the output layer of the newly added model branch are adjusted according to the historical parking stand allocation data of the second airport; in the second stage, the feature extraction layer of the original model branch of the first parking stand allocation model is frozen, the parameters of the intermediate hidden layer and the output layer of the original model branch are adjusted, and the parameters of the dedicated model branch are adjusted; in the third stage, all layers of the newly added model branch and the original model branch are globally optimized, and the model is comprehensively trained using the historical parking stand allocation data of the second airport to optimize model performance.

[0131] The parking stand allocation model training method provided in this embodiment obtains multimodal data related to the parking stand allocation problem from a first airport and a second airport, and determines the similarity of the parking stand allocation problems based on the multimodal data from the first and second airports. Based on the similarity of the parking stand allocation problems, a model migration strategy is determined. Then, according to the determined model migration strategy and the historical parking stand allocation data of the second airport, a first parking stand allocation model for the first airport is trained to obtain a second parking stand allocation model suitable for the second airport. While maintaining the personalized adaptability of the allocation model, this method significantly reduces model development and maintenance costs, enabling the allocation model to automatically adapt to changes in airport operations, improving the efficiency of allocation model construction, reducing the complexity of building allocation models between different airports, improving the versatility of the allocation model, and enhancing the efficiency of the allocation model in solving dynamic parking stand allocation problems.

[0132] The parking stand allocation model training method provided in the above embodiment can be used to dynamically allocate parking stands at an airport, thereby improving airport operation efficiency and service quality.

[0133] Based on the above embodiment, the parking stand allocation model is applied to the parking stand allocation method. Figure 4 This is a flow chart of the parking space allocation method provided by this application. The execution subject of this embodiment can be, for example, an airport ground management system equipped with a parking space dynamic control module. Figure 4 As shown, the training method of the parking space allocation model provided in this embodiment includes:

[0134] S401: Acquire parking stand environment status information of a second airport.

[0135] The parking stand environment status information is a dynamic data set used to describe the real-time status of all parking stands in the airport.

[0136] It is understood that the airport ground management system can collect the real-time status of all parking stands at the second airport to obtain the occupancy status of the corresponding parking stands, such as "free," "occupied," and "under maintenance." Furthermore, the airport ground management system can also obtain dynamic flight information at the current airport in real time. This application does not impose any specific restrictions on the data and information that the airport ground management system can obtain.

[0137] For example, the currently acquired parking stand environment status information of Airport B includes: "Stand 1-1: Aircraft A is occupied, estimated departure time 14:30", "Stand 1-2: Idle", "Stand 1-3: Aircraft B is occupied, estimated departure time 15:15, ground service (refueling) is in progress", "Stand 2-1: Under maintenance, jet bridge equipment failure, estimated recovery at 16:00", "Stand 2-2: Idle", "Stand 2-3: Aircraft C is occupied, estimated departure time 16:00, jet bridge boarding is in progress", "Stand 3-1: Idle", "Stand 3-2: Aircraft D is occupied, estimated departure time 17:00, remote stand, shuttle bus is in place", "Stand 3-3: Idle".

[0138] S402: Input the parking stand environment status information of the second airport into the second parking stand allocation model to obtain the parking stand allocation action output by the second parking stand allocation model.

[0139] It can be understood that the parking stand allocation action is a specific operation instruction generated by the parking stand allocation model based on the parking stand environmental status information; the parking stand allocation action output by the second parking stand allocation model accurately arranges the temporal and spatial relationships between aircraft, stands, and ground services to avoid flight arrival and departure conflicts and improve airport ground operation efficiency.

[0140] Exemplarily, the parking stand environmental status information of the second airport is input into the second parking stand allocation model to obtain the parking stand allocation actions output by the second parking stand allocation model. Specifically, the parking stand allocation actions obtained this time include: "14:10: Guide aircraft A to taxi from stand 1-1 to taxiway 1; 14:20: Taxiway 1 is open, aircraft A enters runway 1 to queue up", "15:00: Refueling is completed, notify ground staff to evacuate equipment; 15:05: Guide aircraft B to taxi from stand 1-3 to taxiway 2; 15:10: Taxiway 2 is open, aircraft B enters runway 2 to queue up", "15:45: Boarding is completed, notify the jet bridge to evacuate; 15:50: Guide aircraft C to taxi from stand 2-3 to taxiway 3; 15:55: Taxiway 3 is open, aircraft C enters runway 3 to queue up".

[0141] The parking stand allocation method provided in this embodiment obtains parking stand environmental status information at a second airport, inputs this information into a second parking stand allocation model, and generates a parking stand allocation action output by the second parking stand allocation model. This method can respond to dynamic flight changes in real time, achieving efficient and dynamic allocation of parking stand resources.

[0142] Figure 5 Schematic diagram of the structure of the training device for the parking space allocation model provided by this application. Figure 5As shown, the present application provides a training device for a parking stand allocation model, and the training device 50 for the parking stand allocation model includes:

[0143] The determination module 501 is configured to determine the similarity between the parking space allocation problems of the first airport and the second airport.

[0144] The determining module 501 is further configured to determine a model migration strategy based on the similarity of the parking space allocation problem.

[0145] Processing module 502 is configured to train a first parking stand allocation model for the first airport according to the model migration strategy and historical parking stand allocation data for the second airport to obtain a second parking stand allocation model suitable for the second airport, wherein the first parking stand allocation model is a trained model, the input of the first parking stand allocation model is parking stand environment status information, and the output of the first parking stand allocation model is a parking stand allocation action.

[0146] Optionally, the training device for the parking space allocation model further includes: an acquisition module 503.

[0147] The acquisition module 503 is used to acquire multimodal data related to the parking stand allocation problem at each of the first airport and the second airport, wherein the multimodal data includes multiple items of the following data: airport images, airport layout topology maps, parking stand-related facility data, flight dynamic data, environmental data, and parking stand allocation optimization objectives.

[0148] The determining module 501 is further configured to determine the similarity of the parking stand allocation problem based on the multimodal data of the first airport and the second airport.

[0149] Optionally, the processing module 502 is further used to input the multimodal data of each of the first airport and the second airport into a large model, extract the initial features of each item of data in the multimodal data through the large model, and map the initial features of each item of data to a unified feature space to obtain the feature vectors of each item of data corresponding to the first airport and the second airport, and determine the attention weights between different items of data of each of the first airport and the second airport.

[0150] The determination module 501 is further configured to determine the multimodal fusion features of the first airport and the second airport respectively based on the feature vectors of the respective data items corresponding to the first airport and the second airport, and the attention weights between the different data items of the first airport and the second airport respectively.

[0151] The determining module 501 is further configured to determine the similarity of the parking space allocation problem based on the multimodal fusion features of the first airport and the second airport.

[0152] Optionally, the determining module 501 is further configured to determine the similarity of each item of data in the respective multimodal data of the first airport and the second airport.

[0153] The determining module 501 is further configured to determine the similarity of the parking space allocation problem based on the weight of each data item of the similarity of each data item.

[0154] Optionally, the determination module 501 is further configured to determine that the model migration strategy is state alignment and model output layer adjustment when the similarity of the parking space allocation problem is greater than or equal to a first threshold.

[0155] The processing module 502 is further configured to freeze the parameters of the feature extraction layer and the intermediate hidden layer of the first parking space allocation model when the model migration strategy is state alignment and adjustment of the model output layer.

[0156] The processing module 502 is further configured to map the state space applicable to the second airport to the state space of the first parking stand allocation model, and train the first parking stand allocation model using historical parking stand allocation data of the second airport to adjust output layer parameters of the first parking stand allocation model, thereby obtaining the second parking stand allocation model.

[0157] Optionally, the determination module 501 is further configured to determine that the model migration strategy is state alignment and model layer adjustment when the similarity of the parking space allocation problem is greater than or equal to a second threshold and less than a first threshold.

[0158] The processing module 502 is further configured to, when the model migration strategy is state alignment and model hierarchical adjustment, map the state space applicable to the second airport to the state space of the first parking stand allocation model, and train the first parking stand allocation model in stages using historical parking stand allocation data of the second airport to obtain a second parking stand allocation model;

[0159] In the first stage, parameters of the feature extraction layer and the intermediate hidden layer of the first parking space allocation model are frozen, and parameters of the output layer of the first parking space allocation model are adjusted;

[0160] In the second stage, parameters of a feature extraction layer of the first parking stand allocation model are frozen, and parameters of an intermediate hidden layer and an output layer of the first parking stand allocation model are adjusted;

[0161] In the third stage, the parameters of each layer of the first parking stand allocation model are adjusted.

[0162] Optionally, the determination module 501 is further used to determine that the model migration strategy is to add a new model branch when the similarity of the parking space allocation problem is greater than or equal to a third threshold and less than a second threshold, and to perform hierarchical adjustments to the new model branch and the original model branch.

[0163] The processing module 502 is further configured to, when the model migration strategy is to add a new model branch and to perform hierarchical adjustment on the new model branch and the original model branch, add a dedicated model branch based on the original model branch of the first parking stand allocation model, wherein the dedicated model branch is configured to process the state space and action space that are different between the first airport and the second airport.

[0164] The processing module 502 is further configured to perform phased training on the first parking stand allocation model after adding the dedicated model branch to obtain the second parking stand allocation model;

[0165] In the first stage, the original model branch of the first parking stand allocation model is frozen, and the parameters of the dedicated model branch are adjusted;

[0166] In the second stage, the feature extraction layer of the original model branch of the first parking stand allocation model is frozen, the parameters of the intermediate hidden layer and the output layer of the original model branch are adjusted, and the parameters of the dedicated model branch are adjusted;

[0167] In the third stage, the parameters of the original model branch and the dedicated model branch are adjusted.

[0168] Figure 6 This is a schematic diagram of the structure of the parking space allocation device provided by this application. Figure 6 As shown, the present application provides a parking stand allocation device, the parking stand allocation device 60 includes:

[0169] The acquisition module 601 is configured to acquire parking stand environment status information of a second airport, where the parking stand environment status information includes parking stand information, taxiway information, runway information, and unassigned flight status information.

[0170] Processing module 602 is configured to input the parking stand environment status information of the second airport into a second parking stand allocation model to obtain a parking stand allocation action output by the second parking stand allocation model, wherein the second parking stand allocation model is trained using the parking stand allocation model training method described in the above embodiment.

[0171] Figure 7 This is a schematic diagram of the structure of the electronic device provided in this application. Figure 7 As shown, the present application provides an electronic device, which includes a receiver 701, a transmitter 702, a processor 703 and a memory 704.

[0172] Receiver 701, for receiving instructions and data;

[0173] Transmitter 702, used to send instructions and data;

[0174] Memory 704, for storing computer-executable instructions;

[0175] Processor 703 is configured to execute computer-executable instructions stored in memory 704 to implement the steps of the parking stand allocation model training method or the parking stand allocation method in the above-described embodiments. For details, please refer to the relevant descriptions in the above-described parking stand allocation model training method or parking stand allocation method embodiments.

[0176] Optionally, the memory 704 may be independent or integrated with the processor 703 .

[0177] When the memory 704 is independently provided, the electronic device further includes a bus for connecting the memory 704 and the processor 703 .

[0178] The present application also provides a computer-readable storage medium, which stores computer-executable instructions. When a processor executes the computer-executable instructions, the training of the parking space allocation model or the parking space allocation method as performed by the above-mentioned electronic device is implemented.

[0179] An embodiment of the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the training of the parking stand allocation model or the parking stand allocation method described in any of the aforementioned embodiments.

[0180] Those skilled in the art will appreciate that all or some of the steps, systems, and functional modules / units in the methods, systems, and devices disclosed above may be implemented as software, firmware, hardware, or any combination thereof. In hardware implementations, the division between functional modules / units described above does not necessarily correspond to the division between physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all of the physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on computer-readable media, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is well known to those skilled in the art, the term computer storage media encompasses both volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as is well known to those skilled in the art, communication media typically embodies computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0181] So far, the technical solution of the present application has been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the scope of protection of the present application is obviously not limited to these specific embodiments. The above embodiments are only used to illustrate the technical solution of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, ordinary technicians in this field should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some or all of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solution to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A training method for a parking stand allocation model, characterized in that: include: Determine the similarity of the parking slot allocation problems at the first airport and the second airport; Determining a model migration strategy based on the similarity of the parking slot allocation problem; According to the model migration strategy and the historical parking stand allocation data of the second airport, the first parking stand allocation model of the first airport is trained to obtain a second parking stand allocation model suitable for the second airport, wherein the first parking stand allocation model is a trained model, the input of the first parking stand allocation model is parking stand environment status information, and the output of the first parking stand allocation model is a parking stand allocation action.

2. The method according to claim 1, characterized in that Determining the similarity between the parking space allocation problems of the first airport and the second airport includes: Obtaining multimodal data related to the parking stand allocation problem at each of the first airport and the second airport, wherein the multimodal data includes multiple items of the following data: airport images, airport layout topology maps, parking stand-related facility data, flight dynamic data, environmental data, and parking stand allocation optimization objectives; A similarity of the parking stand allocation problem is determined based on the multimodal data of each of the first airport and the second airport.

3. The method according to claim 2, characterized in that Determining the similarity of the parking stand allocation problem based on the multimodal data of the first airport and the second airport includes: Inputting the multimodal data of each of the first airport and the second airport into a large model, extracting initial features of each item of data in the multimodal data using the large model, mapping the initial features of each item of data into a unified feature space, obtaining feature vectors corresponding to each item of data of the first airport and the second airport, and determining attention weights between different items of data of the first airport and the second airport; Determining multimodal fusion features of the first airport and the second airport based on the feature vectors of the respective items of data corresponding to the first airport and the second airport, and the attention weights between the different items of data of the first airport and the second airport; Determine the similarity of the parking space allocation problem based on the multimodal fusion features of the first airport and the second airport.

4. The method according to claim 2, characterized in that Determining the similarity of the parking stand allocation problem based on the multimodal data of the first airport and the second airport includes: determining a similarity between each item of multimodal data of the first airport and the second airport; The similarity of the parking space allocation problem is determined according to the weight of each data item and the similarity of each data item.

5. The method according to any one of claims 1 to 4, characterized in that The determining of the model migration strategy based on the similarity of the parking slot allocation problem includes: When the similarity of the parking space allocation problem is greater than or equal to a first threshold, determining that the model migration strategy is state alignment and model output layer adjustment; The step of training the first parking stand allocation model for the first airport according to the model migration strategy and the historical parking stand allocation data of the second airport to obtain a second parking stand allocation model suitable for the second airport includes: When the model migration strategy is state alignment and adjustment of the model output layer, freezing the parameters of the feature extraction layer and the intermediate hidden layer of the first parking slot allocation model; The state space applicable to the second airport is mapped to the state space of the first parking stand allocation model. The first parking stand allocation model is trained using historical parking stand allocation data of the second airport to adjust output layer parameters of the first parking stand allocation model to obtain the second parking stand allocation model.

6. The method according to any one of claims 1 to 4, characterized in that The determining of the model migration strategy based on the similarity of the parking slot allocation problem includes: When the similarity of the parking space allocation problem is greater than or equal to a second threshold and less than a first threshold, determining that the model migration strategy is state alignment and model layer adjustment; The step of training the first parking stand allocation model for the first airport according to the model migration strategy and the historical parking stand allocation data of the second airport to obtain a second parking stand allocation model suitable for the second airport includes: When the model migration strategy is state alignment and model hierarchical adjustment, the state space applicable to the second airport is mapped to the state space of the first parking stand allocation model, and the first parking stand allocation model is trained in stages using historical parking stand allocation data of the second airport to obtain the second parking stand allocation model; In the first stage, parameters of the feature extraction layer and the intermediate hidden layer of the first parking space allocation model are frozen, and parameters of the output layer of the first parking space allocation model are adjusted; In the second stage, parameters of a feature extraction layer of the first parking stand allocation model are frozen, and parameters of an intermediate hidden layer and an output layer of the first parking stand allocation model are adjusted; In the third stage, the parameters of each layer of the first parking stand allocation model are adjusted.

7. The method according to any one of claims 1 to 4, characterized in that The determining of the model migration strategy based on the similarity of the parking slot allocation problem includes: When the similarity of the parking space allocation problem is greater than or equal to a third threshold and less than a second threshold, determining that the model migration strategy is to add a new model branch, and hierarchically adjusting the new model branch and the original model branch; The step of training the first parking stand allocation model for the first airport according to the model migration strategy and the historical parking stand allocation data of the second airport to obtain a second parking stand allocation model suitable for the second airport includes: In a case where the model migration strategy is to add a new model branch and perform hierarchical adjustment on the new model branch and the original model branch, a dedicated model branch is added on the basis of the original model branch of the first parking stand allocation model, wherein the dedicated model branch is used to process the state space and action space having differences between the first airport and the second airport; The first parking stand allocation model with the dedicated model branch added is trained in stages to obtain the second parking stand allocation model; In the first stage, the original model branch of the first parking stand allocation model is frozen, and the parameters of the dedicated model branch are adjusted; In the second stage, the feature extraction layer of the original model branch of the first parking stand allocation model is frozen, the parameters of the intermediate hidden layer and the output layer of the original model branch are adjusted, and the parameters of the dedicated model branch are adjusted; In the third stage, the parameters of the original model branch and the dedicated model branch are adjusted.

8. A parking stand allocation method, characterized in that: include: Acquiring parking stand environment status information of the second airport, wherein the parking stand environment status information includes parking stand information, taxiway information, runway information, and unassigned flight status information; Inputting the parking stand environment status information of the second airport into a second parking stand allocation model to obtain a parking stand allocation action output by the second parking stand allocation model, wherein the second parking stand allocation model is trained using the method according to any one of claims 1 to 7.

9. A training device for a parking stand allocation model, characterized in that: include: A determination module, configured to determine the similarity between the parking space allocation problems of the first airport and the second airport; The determining module is further configured to determine a model migration strategy based on the similarity of the parking space allocation problem; and a processing module, configured to train a first parking stand allocation model for the first airport according to the model migration strategy and historical parking stand allocation data for the second airport to obtain a second parking stand allocation model suitable for the second airport, wherein the first parking stand allocation model is a trained model, the input of the first parking stand allocation model is parking stand environment status information, and the output of the first parking stand allocation model is a parking stand allocation action.

10. A parking space allocation device, characterized in that: include: an acquisition module, configured to acquire parking stand environment status information of the second airport, wherein the parking stand environment status information includes parking stand information, taxiway information, runway information, and unassigned flight status information; a processing module, configured to input the parking stand environmental status information of the second airport into a second parking stand allocation model, and obtain a parking stand allocation action output by the second parking stand allocation model, wherein the second parking stand allocation model is trained using the parking stand allocation model training method described in the first aspect and various possible implementations of the first aspect.

11. An electronic device, characterized in that: include: a processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 8.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 8 when executed by a processor.

13. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 8 when the computer program is executed by a processor.

Images