Intelligent scheduling method, system and equipment for airport ground service guarantee resources and medium

Abstract

The invention discloses an intelligent scheduling method, system and device for airport ground service guarantee resources and a medium, and is applied to the technical field of artificial intelligence. Based on the standardized guarantee task context data of each flight, performing data braiding on the multi-source heterogeneous data to obtain real-time guarantee task context data of each flight; when any key business event is detected, generating a natural language resource scheduling scheme for a target guarantee task of a target flight influenced by the key business event by adopting a multi-dimensional optimization algorithm through a large language model finely adjusted by airport ground service guarantee vehicle scheduling knowledge based on real-time guarantee task context data of each flight; and resource scheduling is carried out according to the natural language resource scheduling scheme of the target guarantee task, so that a data island is broken, the resource scheduling response speed and the resource utilization rate are improved, scheduling experience precipitation and complex scene intelligent reasoning are realized, and the method is suitable for a dynamic scheduling scene of airport ground service guarantee resources.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, system, equipment and medium for intelligent scheduling of airport ground service support resources. Background Technology

[0002] In current airport ground service operations, the dispatching of vehicles required for flight support tasks (such as baggage handling, water refilling, garbage collection, and shuttle bus transportation) mainly relies on manual scheduling. Dispatchers typically allocate tasks before flight execution based on flight plans, historical experience, and electronic scheduling systems. Some airports have deployed information technology tools such as vehicle positioning systems and flight information systems, enabling the visualization of basic data.

[0003] When flights are delayed, canceled, added unexpectedly, or affected by sudden weather events, the original scheduling becomes invalid. Dispatchers must communicate by phone, manually assess resource availability, and manually adjust task assignments. The entire process lacks systematic dynamic evaluation and optimization capabilities, making it difficult to make globally optimal or suboptimal decisions in a short period of time. Summary of the Invention

[0004] This application provides a method, system, equipment, and medium for intelligent scheduling of airport ground service support resources, aiming to solve problems such as data silos, slow manual response, black boxes in decision-making, and difficulty in accumulating experience in existing technologies. The technical solution provided by this application is as follows: On the one hand, this application provides an intelligent scheduling method for airport ground service support resources, including: Real-time access to flight dynamic data from the flight information system, vehicle status data from the vehicle positioning system, personnel scheduling data from the personnel scheduling system, and meteorological data from the meteorological service system are used as multi-source heterogeneous data. Based on the pre-built standardized support mission context data for each flight, data weaving is performed on multi-source heterogeneous data to obtain the real-time support mission context data for each flight; wherein, the standardized support mission context data includes a minimum set of structured attributes required for a support mission and its resource scheduling. When at least one key business event in a pre-defined set of key business events is detected, a natural language resource scheduling scheme is generated for the target support task of the target flight affected by each key business event, based on the real-time support task context data of each key business event and each flight, through a large language model fine-tuned by airport ground service support vehicle scheduling knowledge, and using a multi-dimensional optimization algorithm. Resource scheduling is carried out according to the natural language resource scheduling scheme corresponding to the target flight's target support mission.

[0005] Optionally, based on pre-built standardized support mission context data for each flight, multi-source heterogeneous data is weaved to obtain real-time support mission context data for each flight, including: Using the standardized support mission context data of each flight as a standardized data structure template, semantic mapping and instantiation of multi-source heterogeneous data are performed to obtain real-time support mission context data containing real-time resource status for each flight.

[0006] Optionally, the set of key business events includes at least: the actual arrival / departure time deviation of a flight is greater than the deviation threshold, vehicle malfunction or change in maintenance status, temporary absence of personnel, sudden weather changes, and mission timeout or abnormal interruption.

[0007] Optionally, based on the real-time support mission context data of each key business event and each flight, a natural language resource scheduling scheme is generated for the target support mission of the target flight affected by each key business event using a multi-dimensional optimization algorithm through a large language model fine-tuned by airport ground service support vehicle scheduling knowledge, including: The context data of each key business event and each flight's real-time support task are input into a large language model. The large language model then performs the following operations: using logical reasoning prompting technology, it identifies the target support tasks for the target flights affected by each key business event; based on scheduling objectives and constraints, it selects a set of candidate resource scheduling schemes for the target flight's target support task; based on the time cost, economic cost, and risk index of each candidate resource scheduling scheme in the candidate resource scheduling scheme set, it calculates a multi-indicator weighted score for each candidate resource scheduling scheme; it selects the candidate resource scheduling scheme whose multi-indicator weighted score meets the selection criteria as the target resource scheduling scheme; and it converts the target resource scheduling scheme into a natural language resource scheduling scheme.

[0008] Optionally, based on the scheduling objectives and constraints, a set of candidate resource scheduling schemes is selected for the target flight's target support task, including: A multi-attribute matching algorithm based on bitmap indexing is used to select a set of candidate resource scheduling schemes that meet the scheduling objectives and constraints for the target flight's target support task.

[0009] Optionally, the target resource scheduling scheme is converted into a natural language resource scheduling scheme, including: Based on pre-configured explanatory prompt templates, the target resource scheduling scheme is converted into a natural language resource scheduling scheme that includes recommended information, multi-dimensional explanations, and comparisons of alternative schemes.

[0010] Optionally, based on the real-time support mission context data of each flight, and using a large language model fine-tuned with airport ground service support vehicle scheduling knowledge, after generating a resource scheduling plan for the target support mission of the target flight affected by the critical business event using a multi-dimensional optimization algorithm, the plan also includes: Visualize the resource scheduling plan; In response to the scheduler's modification of the resource scheduling scheme, a comparison dataset is formed by comparing the modified resource scheduling scheme with the resource scheduling scheme output by the large language model. The comparison dataset is fed back to the large language model to optimize engineering strategies or fine-tune the training of the large language model.

[0011] On the other hand, this application provides an intelligent system for airport ground service support vehicles, including: The data access layer is used to access flight dynamic data from the flight information system, vehicle status data from the vehicle positioning system, personnel scheduling data from the personnel scheduling system, and meteorological data from the meteorological service system in real time as multi-source heterogeneous data. The data weaving layer is used to weave multi-source heterogeneous data to obtain real-time support mission context data for each flight based on pre-built standardized support mission context data for each flight; wherein, the standardized support mission context data includes a minimum set of structured attributes required for a support mission and its resource scheduling. The event detection layer is used to detect key business events in a pre-defined set of key business events in real time. The intelligent decision-making layer is used to generate a natural language resource scheduling scheme for the target support task of the target flight affected by each key business event when the event detection layer detects at least one key business event. Based on the real-time support task context data of each key business event and each flight, it uses a large language model fine-tuned by airport ground service support vehicle scheduling knowledge and adopts a multi-dimensional optimization algorithm. The scheduling execution layer is used to schedule resources according to the natural language resource scheduling scheme corresponding to the target flight's target support task.

[0012] On the other hand, this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned intelligent scheduling method for airport ground service support resources.

[0013] On the other hand, this application also provides a computer-readable storage medium storing computer instructions, which, when executed by a processor, implement the above-mentioned intelligent scheduling method for airport ground service support resources.

[0014] The beneficial effects of this application are as follows: (1) Breaking down data silos and achieving high-quality real-time integration: Data weaving technology solves the problem of integrating multi-source heterogeneous data, providing a unified, reliable, and low-latency data foundation for scheduling decisions, and effectively shortening the data integration delay; (2) Significantly improved dynamic response capability: The event-driven mechanism enables immediate dispatching in the event of an incident, thereby effectively improving the timeliness of dispatching suggestions, greatly reducing flight support delays, and improving flight punctuality. (3) Improved resource matching accuracy and utilization: By generating resource scheduling schemes through multi-dimensional optimization algorithms, resource utilization can be effectively improved and empty running rate can be significantly reduced; (4) The scheduling decision is interpretable and highly reliable: the resource scheduling scheme is described in natural language, which can effectively improve the efficiency of human-machine collaboration.

[0015] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0016] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram outlining the intelligent scheduling method for airport ground service support resources in the embodiments of this application; Figure 2 This is a functional structure diagram of the intelligent scheduling system for airport ground service support resources in the embodiments of this application; Figure 3 This is a schematic diagram of the hardware structure of the electronic device in the embodiments of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and beneficial effects of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] This application provides an intelligent scheduling system for airport ground service support resources. (See attached document.) Figure 1As shown, the airport ground service support resource intelligent scheduling system 100 provided in this application embodiment includes at least: The data access layer 101 is used to access flight dynamic data from the flight information system, vehicle status data from the vehicle positioning system, personnel scheduling data from the personnel scheduling system, and meteorological data from the meteorological service system in real time as multi-source heterogeneous data. Data weaving layer 102 is used to weave multi-source heterogeneous data based on pre-built standardized support mission context data for each flight to obtain real-time support mission context data for each flight; wherein, the standardized support mission context data includes a minimum set of structured attributes required for a support mission and its resource scheduling. Event detection layer 103 is used to detect key business events in a pre-defined set of key business events in real time; The intelligent decision layer 104 is used to generate a natural language resource scheduling scheme for the target support task of the target flight affected by each key business event when the event detection layer 103 detects at least one key business event. Based on the real-time support task context data of each key business event and each flight, it uses a large language model fine-tuned by airport ground service support vehicle scheduling knowledge and adopts a multi-dimensional optimization algorithm. The scheduling execution layer 105 is used to perform resource scheduling according to the natural language resource scheduling scheme corresponding to the target flight's target support task.

[0019] In one possible implementation, the data weaving layer 102 is used to perform semantic mapping and instantiation filling on multi-source heterogeneous data using standardized support mission context data of each flight as a standardized data structure template, so as to obtain real-time support mission context data containing real-time resource status corresponding to each flight.

[0020] In one possible implementation, the set of critical business events includes at least: the actual arrival / departure time deviation of a flight is greater than the deviation threshold, vehicle malfunction or change in maintenance status, temporary absence of personnel, sudden weather changes, and mission timeout or abnormal interruption.

[0021] In one possible implementation, the intelligent decision layer 104 is used to input the real-time support task context data of each key business event and each flight into a large language model, and perform the following operations through the large language model: using logical reasoning prompting technology to identify the target support task of the target flight affected by each key business event; based on scheduling objectives and constraints, to screen a set of candidate resource scheduling schemes for the target support task of the target flight; based on the time cost, economic cost, and risk index of each candidate resource scheduling scheme in the set of candidate resource scheduling schemes, to calculate the multi-index weighted score of each candidate resource scheduling scheme; to select the candidate resource scheduling scheme whose multi-index weighted score meets the screening criteria as the target resource scheduling scheme; and to convert the target resource scheduling scheme into a natural language resource scheduling scheme.

[0022] In one possible implementation, the intelligent decision layer 104 is used to employ a multi-attribute matching algorithm based on bitmap indexing to select a set of candidate resource scheduling schemes that meet the scheduling objectives and constraints for the target flight's target support task.

[0023] In one possible implementation, the intelligent decision layer 104 is used to convert the target resource scheduling scheme into a natural language resource scheduling scheme that includes recommended information, multi-dimensional explanations, and comparisons of alternative schemes, based on a pre-configured explanatory prompt template.

[0024] In one possible implementation, the airport ground service support resource intelligent scheduling system 100 provided in this application embodiment further includes: The visualization layer 106 is used to visualize the resource scheduling scheme; The continuous optimization layer 107 is used to respond to the scheduler's modification operation on the resource scheduling scheme, and to form a comparison dataset with the resource scheduling scheme output by the large language model. The comparison dataset is fed back to the large language model to optimize the engineering strategy or fine-tune the model training.

[0025] Based on the above embodiments, this application provides an intelligent scheduling method for airport ground service support resources, see below. Figure 2 As shown in the embodiments of this application, the general flow of the intelligent scheduling method for airport ground service support resources is as follows: Step 201: Real-time access to flight dynamic data from the flight information system, vehicle status data from the vehicle positioning system, personnel scheduling data from the personnel scheduling system, and meteorological data from the meteorological service system as multi-source heterogeneous data.

[0026] In this embodiment, the data access layer 101 deploys various standardized data adapters, enabling it to interface with flight information systems, vehicle positioning systems, personnel scheduling systems, and meteorological service systems, thereby achieving real-time collection of four types of core multi-source heterogeneous data, specifically including: Flight status data access: Data is retrieved every 10 seconds via the RESTful API interface provided by the flight information system. The data includes flight number, scheduled departure / arrival time, actual departure / arrival time, flight status (awaiting departure / en route / landed / delayed), parking position, etc. At the same time, push notifications from the flight information system are subscribed to ensure real-time acquisition of flight status change data (such as sudden delays or diversions). Vehicle status data access: Data is transmitted in real time via the GPS positioning module of the vehicle positioning system and the vehicle-mounted IoT terminal using the MQTT protocol. The data content includes vehicle ID, license plate number, real-time location coordinates (latitude and longitude), operating status (idle / working / returning / fault), remaining battery / fuel, current task ID, driving speed, etc. The transmission frequency is 5 seconds / time. Personnel scheduling data access: Through a direct database connection to the personnel scheduling system (using JDBC interface), real-time data such as personnel ID, name, job type (driver / ground staff), skills and qualifications (e.g., special vehicle operation certificate), daily schedule, on-duty status, and temporary leave records are obtained. At the same time, real-time reports of personnel temporarily leaving their posts are received. Through real-time data connection, there is no need for timed synchronization, and data can be used immediately by looking up and inserting. Meteorological data access: Meteorological data of the airport and surrounding areas are obtained in real time through the HTTP interface of the meteorological service system, including temperature, precipitation, wind speed, visibility, and meteorological warning level (such as rainstorm warning and gale warning), with a data update frequency of 30 seconds / time.

[0027] After the above-mentioned multi-source heterogeneous data is accessed, it is stored in a temporary data buffer to prepare for subsequent data weaving processing.

[0028] Step 202: Based on the pre-built standardized support mission context data of each flight, perform data weaving on the multi-source heterogeneous data to obtain the real-time support mission context data of each flight; wherein, the standardized support mission context data includes a minimum set of structured attributes required for a support mission and its resource scheduling.

[0029] In this embodiment, the data weaving layer 102 integrates a data weaving engine and a standardized task context database. The data weaving layer 102 enables data weaving processing of multi-source heterogeneous data, specifically including: Step 2021: Pre-build standardized assurance mission context data: For each flight at the airport, standardized support task context data is pre-constructed. Each standardized support task context data is associated with a unique flight identifier (such as flight number + departure and arrival dates), and each standardized support task context data contains a minimum set of structured attributes required for a support task and its resource scheduling, as detailed below: Break down support tasks: Break down standardized support tasks according to the flight support process. Support tasks include, but are not limited to, aircraft water refueling, sewage suction, passenger boarding bridge docking, baggage loading and unloading, food supply, etc. Each support task is assigned a unique task ID. Generate a minimal set of structured attributes: The minimal set of structured attributes should include at least the basic task attributes (task ID, task name, planned execution time window, estimated duration, task priority), resource requirement attributes (required vehicle type, number of vehicles, personnel skill requirements), and site attributes (task execution location, work area restrictions), etc.

[0030] For example, in the standardized support mission context data of flight CA1234 (2025-12-10 Beijing-Shanghai), the minimum structured attribute set corresponding to "water refueling mission" is: mission ID=QS-1234-10, planned execution time window=2025-12-10 08:30-08:45, estimated duration=10 minutes, mission priority=normal, required vehicle type=water refueling vehicle, personnel skill requirement=holding a Class A special vehicle operator's license, mission execution gate=T1-211.

[0031] Step 2022: Generate real-time assurance mission context data based on data weaving: The data weaving engine is invoked to fuse multi-source heterogeneous data based on pre-built standardized support mission context data, as follows: Metadata Extraction and Parsing: The data weaving engine automatically extracts metadata (including data source identifiers, data field semantics, and data update time) from multi-source heterogeneous data, and converts unstructured / semi-structured data (such as weather warning text) into structured data through preset parsing rules; Semantic mapping and association: Based on the ontology model of the airport ground service domain, fields in multi-source heterogeneous data are semantically aligned with the minimum structured attribute set of standardized support task context data to establish data association relationships; for example, "actual landing time of flight CA1234" is associated with the "planned execution time window" attribute of its corresponding support task, and "real-time location of water refueling vehicle C1" is associated with the "task execution position" attribute of "water refueling task". Data quality verification and completion: Real-time quality verification of data during the fusion process. Verification rules include data integrity (e.g., whether key vehicle status fields are missing), timeliness (e.g., whether data transmission delay exceeds 30 seconds), and logical consistency (e.g., whether personnel skills and qualifications match task requirements). Automatic completion (e.g., completing missing vehicle locations based on historical data interpolation) or marking is triggered for abnormal data. Real-time context integration: The validated multi-source heterogeneous data is integrated with the standardized support mission context data of the corresponding flight, and information such as the actual execution time window of the mission, the list of available vehicles, the list of on-duty personnel, and the meteorological impact assessment are updated to form the real-time support mission context data of each flight, which is stored in the real-time database and dynamically updated.

[0032] Furthermore, in this embodiment, to address the insufficient alignment accuracy caused by semantic differences in entities within multi-source heterogeneous data, when semantically aligning fields in multi-source heterogeneous data with the minimum structured attribute set of standardized support task context data based on an airport ground service domain ontology model, dynamic embedding vectors can be used to enhance entity semantic relationships, providing high-precision data support for subsequent data fusion and scheduling decisions. Specifically, the data weaving layer 102 also includes a GNN dynamic embedding module; wherein the GNN dynamic embedding module includes: Heterogeneous Entity Feature Extraction Submodule: Used to extract the structured attribute features and dynamic state features of various entities in multi-source heterogeneous data, forming entity feature vectors; Dynamic graph construction submodule: Based on the airport ground service domain ontology model and real-time data, construct and update a heterogeneous graph. GNN model inference submodule: Deploy graph neural network models adapted to heterogeneous entities (such as relational graph convolutional network RGCN, heterogeneous graph Transformer HGT) and perform dynamic entity embedding computation; The Embedded Vector Application Submodule uses the generated entity dynamic embedding vectors for semantic similarity calculation and cross-system entity matching, outputting high-precision semantic alignment results.

[0033] Based on this, when semantically aligning fields from multi-source heterogeneous data with the minimum structured attribute set of standardized support task context data using an ontology model for airport ground services, the following methods can be used, but are not limited to: Step 1: The heterogeneous entity feature extraction submodule extracts heterogeneous entities and entity features based on the metadata of multi-source heterogeneous data, as detailed below: (1) Classification of heterogeneous entity types: Heterogeneous entities include flight entities, mission entities, vehicle entities, personnel entities, gate entities, and meteorological entities. Each entity is associated with a unique identifier (e.g., the flight entity is uniquely identified by "flight number + take-off and landing date", and the vehicle entity is uniquely identified by "vehicle ID"). (2) Entity feature extraction and quantization: Structured attribute features: Extract static attributes of each entity (such as airline, aircraft type, and flight range for flight entities; vehicle type, rated load, and operational qualifications for vehicle entities; skill certificates and job type for personnel entities), and convert them into fixed-dimensional structured feature vectors (64 dimensions) through one-hot encoding, numerical normalization, and other methods. Dynamic status features: Extract the real-time dynamic attributes of each entity (such as the actual take-off and landing time and delay duration of flight entities; the real-time location, operating status, and remaining battery power of vehicle entities; and the warning level and visibility value of meteorological entities), sample and quantize them into dynamic feature vectors (with a dimension of 32) according to time windows (1 minute / window). Feature fusion: The structured feature vector is concatenated with the dynamic state feature vector to form the initial feature vector of each entity (total dimension is set to 96 dimensions), and is dynamically adjusted as real-time data is updated.

[0034] Step 2: The dynamic graph construction submodule constructs a heterogeneous graph based on the domain ontology model and real-time entity data and updates it in real time, as detailed below: (1) Heterogeneous graph initialization: Using the various heterogeneous entities extracted in step 1 as graph nodes, construct graph edges based on predefined airport ground service domain relationship types (such as "flight-include-task", "task-require-vehicle", "task-execute-personnel", "task-associated-aircraft position", "task-affected-weather") to form an initial heterogeneous graph, where the weight of the edge is initially set to 1.0 (representing the default value of the relationship strength). (2) Dynamic updates of heterogeneous graphs: Node Update: When a new entity is added to the multi-source data (such as a new temporary flight or a newly deployed vehicle), its feature vector is automatically extracted and added as a graph node; when an entity's status is cancelled (such as a flight being completed or a vehicle being scrapped), the corresponding node is removed from the graph. Edge updates: When the relationship between entities changes (such as changes in the vehicle assigned to the task or adjustments to the flight parking position), add or delete the corresponding graph edge; dynamically adjust the edge weight based on the frequency of entity interaction (such as the number of times a vehicle performs a task for a certain type of flight) (weight range 0.1-5.0, the higher the frequency, the greater the weight). Update frequency: Synchronized with the real-time data update frequency of the data weaving layer, the heterogeneous graph is incrementally updated every 30 seconds to ensure that the graph structure is consistent with the actual business state.

[0035] Step 3: The GNN model inference submodule loads the pre-trained heterogeneous graph neural network model, which has been fine-tuned with airport ground service domain data, and performs dynamic embedding calculations on entities in the heterogeneous graph, as follows: Model selection and fine-tuning: Relational Graph Convolutional Network (RGCN) was selected as the core model. This model supports assigning independent weight parameters to different types of relationships and adapts to complex relationships between heterogeneous entities. Based on historical heterogeneous entity data and semantic alignment labels (such as manually labeled "flight number-task ID" correctly associated samples), the RGCN model was fine-tuned to optimize the domain adaptability of the model. The initial learning rate for fine-tuning was set to 1e-4, and 5 iterations were performed. Dynamic embedding computation process: Input layer: The node feature vectors (96-dimensional) of the dynamic heterogeneous graph, along with the edge relationship type and edge weights, are used as model input; Graph convolutional layer propagation: Node features are iteratively updated through 3 layers of RGCN convolutional layers. Each convolutional layer aggregates information based on the features of a node's neighboring nodes and edge relationship types using a weighted summation method.

[0036] in, For nodes exist The feature vector of the layer, Let be the set of edge relations. For nodes In edge relations The following is a collection of neighbors. For edge relations The corresponding weight matrix, For neighboring nodes In the The feature vector of the layer, for The layer's bias vector, For activation functions; For all edge relations Summation, For nodes In edge relations All neighboring nodes Summation; Embedded vector output: After convolutional layer iteration, the node feature vector is mapped to a fixed-dimensional (128-dimensional) dynamic embedded vector through the output layer. This vector integrates the entity's own attributes, dynamic state and the association information of neighboring entities, and can accurately represent the semantic features of the entity. Real-time update mechanism: Every time the heterogeneous graph completes a dynamic update (every 30 seconds), the GNN model is triggered to re-execute the embedding calculation, generating an updated entity dynamic embedding vector to ensure that the embedding vector is synchronized with the entity state.

[0037] Step 4: The embedding vector application submodule uses the dynamically generated embedding vectors from the GNN for the semantic alignment process of the data weaving layer, replacing the traditional rule-based semantic mapping method, as follows: Semantic similarity calculation: For heterogeneous entities across systems (such as the "flight number" entity in the flight information system and the "associated flight" entity in the vehicle dispatching system, and the "task-associated flight" entity in the personnel scheduling system), calculate the cosine similarity of their dynamic embedding vectors. The similarity calculation formula is: sim(v1,v2)=(v1·v2) / (||v1||·||v2||), where v1 and v2 are the dynamic embedding vectors of the two entities, respectively. Entity matching decision: Set a similarity threshold (e.g., 0.85). When the similarity between two entities is higher than the threshold, they are determined to be semantically consistent homologous entities, and a cross-system association is established. When the similarity is in the range of 0.7-0.85, it is marked as a suspected match and manual review is triggered. When the similarity is lower than 0.7, they are determined to be non-homologous entities. Alignment result output and application: The semantic alignment result is fed back to the unified data view building unit of the data weaving layer to integrate multi-source data to form a complete business context; at the same time, the alignment result is stored in the semantic mapping knowledge base to provide a reference for subsequent alignment of similar entities and further improve alignment efficiency.

[0038] The above methods can effectively improve the semantic alignment accuracy of heterogeneous entities. The specific advantages are as follows: 1) Dynamic embedding vectors integrate the dynamic state and association information of entities, solving the problem that static attribute matching cannot cope with changes in entity state; 2) GNN models can automatically learn complex semantic associations between heterogeneous entities, reducing reliance on manual rules and adapting to the dynamic changes in airport ground service business; 3) Improved semantic alignment accuracy makes the unified data view after multi-source data fusion more accurate, providing a high-quality data foundation for subsequent key business event detection and scheduling scheme generation, and indirectly improving the rationality and timeliness of scheduling decisions.

[0039] Step 203: When at least one key business event in the pre-defined set of key business events is detected, based on the real-time support task context data of each key business event and each flight, a natural language resource scheduling scheme is generated for the target support task of the target flight affected by each key business event by using a large language model fine-tuned by airport ground service support vehicle scheduling knowledge and employing a multi-dimensional optimization algorithm.

[0040] In this embodiment, the event detection layer 103 has a built-in key business event set and event monitoring module. The key business event set includes, but is not limited to: Flight status anomalies: The actual arrival / departure time of a flight deviates from the scheduled time by more than 5 minutes; flight diversion; flight cancellation. Abnormal vehicle status events: vehicle fault reporting, vehicle maintenance status change, vehicle battery / fuel level below threshold; Abnormal personnel status events: personnel taking temporary leave, personnel leaving their posts without reporting, personnel skills or qualifications expiring; Meteorological events affecting operations: upgraded weather warning levels (such as orange or higher warnings), visibility below the operational threshold, and strong winds affecting outdoor operations; Task execution exception events: ensure task execution timeout, task interruption, and task resource shortage.

[0041] The event monitoring module detects in real time whether any critical business events have occurred. When at least one critical business event is detected, the scheduling plan generation process is triggered, and the target flight affected by the event and the corresponding target support task are marked (e.g., the early landing of flight CA1234 triggers its "water refueling task" and "baggage loading and unloading task" as target support tasks).

[0042] In this embodiment, the intelligent decision layer 104 is deployed with a large language model (hereinafter referred to as "fine-tuned LLM") fine-tuned by airport ground service support vehicle scheduling knowledge. The fine-tuned LLM, combined with multi-dimensional optimization algorithms, generates a natural language resource scheduling scheme for the target support task of the target flight, specifically including: Model input data preparation: The detected key business events (such as "Flight CA1234 landed 15 minutes ahead of schedule"), the real-time support mission context data of the target flight (including the attributes of the target support mission, the information of currently available vehicles / personnel, and meteorological data), and historical scheduling cases (used to assist decision-making) are used as input data and organized into a structured text format that can be recognized by the large language model. Multi-dimensional optimization algorithm configuration: Preset multi-dimensional optimization goals and quantitative indicators. Optimization goals include optimal time (shortest time for vehicles to reach the task point), optimal cost (lowest fuel consumption during empty runs), and optimal risk (lowest vehicle failure rate and highest personnel qualification matching). The weight of the quantitative indicators corresponding to each optimization goal is dynamically allocated by the fine-tuning LLM according to the current event type (e.g., VIP flights are given priority for optimal time with a weight of 0.8, and ordinary flights are given priority for optimal cost with a weight of 0.6). Scheduling Scheme Reasoning and Generation: Based on the input data, the fine-tuned LLM employs logical reasoning prompting technology to identify the target support tasks for the target flights affected by various key business events. A multi-attribute matching algorithm based on bitmap indexing is used to filter a set of candidate resource scheduling schemes that meet the scheduling objectives and constraints for the target flight's target support tasks. Based on the time cost, economic cost, and risk index of each candidate resource scheduling scheme in the set, a multi-indicator weighted score is calculated for each scheme. The candidate resource scheduling scheme whose multi-indicator weighted score meets the screening criteria is selected as the target resource scheduling scheme, i.e., the optimal structured resource scheduling scheme. (Including the assigned vehicle ID, driver ID, adjusted execution time window, and work route for the target support task), and through the built-in natural language generation module, based on the pre-configured explanatory prompt template, the optimal structured resource scheduling scheme is converted into a natural language expression to obtain a natural language resource scheduling scheme. This natural language resource scheduling scheme includes the core information of the recommended scheme, the rationale for the scheme (such as "It is recommended to assign the water refueling vehicle C1 (license plate Zhejiang BXXXX) to perform the CA1234 water refueling task. The vehicle is currently located at T1-203, which is only 300 meters away from the task station T1-211 and can be reached within 2 minutes, which can meet the time requirement of early landing"), alternative schemes, and reasons for not selecting them. Scheme verification: Perform compliance verification on the natural language resource scheduling scheme (such as whether it complies with airport operation area restrictions and whether resources conflict). After the verification is passed, the scheme is output to the scheduling execution layer 105.

[0043] Step 204: Perform resource scheduling according to the natural language resource scheduling scheme corresponding to the target flight's target support task.

[0044] In this embodiment, the scheduling execution layer 105 includes a scheduling instruction issuance interface, an in-vehicle terminal, a personnel terminal, and a scheduling monitoring module, which can realize the implementation and status tracking of the scheduling plan, specifically including: Dispatch instruction conversion and issuance: The natural language resource scheduling scheme is parsed into structured dispatch instructions, which are then pushed to the vehicle terminal (driver's APP), personnel scheduling system, and vehicle dispatch system via interfaces. The instruction content includes task ID, target flight information, execution time, work location, and assigned resource information. The vehicle terminal notifies the driver via pop-up window and voice prompts, and the personnel scheduling system updates the personnel task status simultaneously. Optionally, 5G uRLLC slicing can also be used for instruction issuance. Execution process tracking: The vehicle positioning system monitors the driving trajectory and arrival status of the assigned vehicles in real time, and the personnel terminal provides real-time feedback on the task execution progress (such as "departed", "arrived at the work point", "task completed"). Dispatchers can view the execution status of the dispatch plan through a visual monitoring interface. Exception handling and scheme adjustment: If a new critical business event is detected during execution (such as a sudden vehicle malfunction), repeat steps 203-204 to regenerate and execute the adjusted natural language resource scheduling scheme to ensure the task is completed smoothly.

[0045] Furthermore, after resource scheduling is carried out according to the natural language resource scheduling scheme corresponding to the target flight's target support task, the accuracy and applicability of the scheduling scheme can be improved through human-machine collaboration. Simultaneously, the model can be continuously evolved through data feedback, specifically including: Step 1: Visualizing the resource scheduling plan; Once the fine-tuned large language model generates a resource scheduling plan (structured plan and natural language plan) for the target flight target support task, the visualization display module in the visualization display layer 106 is automatically triggered, transforming the scheduling plan into an intuitive visualization interface and pushing it to the dispatcher's terminal, as follows: Step 1.1: Visual Content Integration: Extract the core elements of the resource scheduling plan, including target flight information (flight number, gate, takeoff and landing status), target support task list (task ID, task type, adjusted execution time window), assigned resource details (vehicle ID, license plate, real-time location, driving route; personnel ID, qualifications, on-duty status), multi-dimensional optimization indicators (estimated arrival time, empty driving distance, cost estimation, risk level), and natural language interpretation text; Step 1.2: Multi-format visualization presentation: Time series visualization: The execution time and resource consumption cycle of each target support task are displayed through Gantt charts, and the time deviation from the original plan is marked. The time axis can be zoomed for viewing. Spatial visualization: Based on a GIS map engine, it overlays geographic information such as airport terminals, aircraft stands, and roads to mark the current location, planned driving trajectory, and task execution stand of assigned vehicles in real time, and uses different colors to distinguish the vehicle's operating status (idle / dispatch in progress / operating). Structured information display: Detailed parameters of resource scheduling schemes are presented in tabular form, including task-resource matching relationships, quantitative values ​​of optimization indicators, and natural language explanations for the schemes. Clicking on table entries allows you to view corresponding visual charts. Anomaly warning label: Potential risks in the plan (such as resource shortages or insufficient time redundancy) are marked with a red box and a prompt text, and the basis for the risk assessment is displayed.

[0046] Step 1.3: Terminal Push and Access Control: Based on the dispatcher's job responsibilities (such as regional dispatcher, special task dispatcher), a visual dispatching plan within the corresponding permission scope is pushed, which supports simultaneous viewing on multiple terminals and ensures information collaboration within the dispatching team.

[0047] Step 2: The dispatcher modifies the operation response and constructs the comparison dataset; After viewing the visual scheduling plan through the interactive terminal, the dispatcher can perform interactive operations and record data for the plan, as detailed below: (1) The interactive operation types and response logic include: Confirm Execution: If the dispatcher approves the plan, they click the "Confirm Execution" button. The system records the confirmation operation timestamp, marks the plan as "executable," and pushes it to the scheduling execution layer. Manual modification: If the plan has problems such as unreasonable resource matching or time planning deviation, the dispatcher can perform modification operations through the visual interface, including changing the assigned vehicle / personnel, adjusting the task execution order, modifying the time window, etc. The system synchronizes the modified visual content in real time and records the modification traces (modified fields, original content, new content, modification time, operator). Rejection and Regeneration: If the plan has major defects (such as irreconcilable resource conflicts or non-compliance with airport operation standards), the dispatcher clicks the "Reject" button, fills in the rejection reason (such as "the assigned vehicle does not have the corresponding operation qualification" or "the time window conflicts with other tasks"), the system records the rejection information and triggers the process of regenerating the plan from the model.

[0048] (2) The logic for constructing the comparison dataset includes: Data Acquisition: Through the dataset management module in continuous optimization layer 107, the following core data are automatically collected and associated: the original resource scheduling scheme output by the large language model (structured data + natural language text), the resource scheduling scheme modified by the dispatcher (structured data + natural language text, if any), modification trace data, rejection reason text (if any), the corresponding target flight real-time support task context data, and the triggered key business event information. Data structuring: The collected data is standardized, and each data item is assigned a unique association identifier (such as "event ID + flight number + task ID"). Natural language text (rejection reasons, solution explanations) is converted into structured labels (such as "qualification mismatch" and "time conflict" labels) to form a four-dimensional comparison dataset containing "input context - original solution - manually corrected solution - correction label". Data storage and management: The comparison dataset is stored in a structured database and indexed by “event type”, “flight type”, and “task type”. It supports searching by time range, optimization indicators and other conditions, and sets up a data backup mechanism to ensure the integrity and traceability of the dataset.

[0049] Step 3: Compare the dataset feedback with the large language model optimization; The constructed comparison dataset is periodically fed back to the model optimization module in the continuous optimization layer 107. Iterative upgrades of the large language model are achieved through engineering strategy optimization or model fine-tuning training, as detailed below: Data filtering and priority ranking: The model optimization module filters the comparison dataset according to preset rules, giving priority to data samples with "many manual corrections", "typical rejection reasons" and "covering complex business scenarios (such as multiple events overlapping)" to form a high-quality training / optimization dataset and avoid interference from invalid data to the model; Engineering strategy optimization: For frequently occurring correction types in the comparison dataset (such as "vehicle qualification matching error" and "time window calculation deviation"), the model's inference strategy is automatically adjusted, including optimizing the prompt word template (supplementing the guidance logic for qualification verification and time conflict checking), adjusting the weight allocation rules of the multi-dimensional optimization algorithm (such as increasing the weight of qualification matching indicators), and updating the domain knowledge dictionary (supplementing newly added vehicle models / qualification types). Optimization strategy verification: Apply the adjusted engineering strategy to the test dataset to verify the optimization effect of the strategy (such as the percentage reduction in error rate and the percentage increase in solution confirmation rate). If the preset threshold is reached (such as a reduction in error rate of ≥20%), then solidify the strategy and apply it to online model inference.

[0050] Model fine-tuning training: When the size of the comparison dataset reaches a preset threshold (e.g., a cumulative total of 1000 valid samples), the model fine-tuning process is initiated. The low-rank adaptation (LoRA) algorithm is used to perform secondary fine-tuning on the large language model that has been fine-tuned by airport ground service scheduling knowledge. The "input context-human correction scheme" in the comparison dataset is used as the training sample pair, and the model parameters are optimized using the cross-entropy loss function. Fine-tuning process control: Set the number of iterations for fine-tuning training and the learning rate (e.g., initial learning rate of 10). -5 The training process involves 5 iterations and employs an early stopping mechanism to prevent overfitting (training stops when the validation set loss does not decrease for 3 consecutive iterations). Model Updates and Canary Releases: After fine-tuning, the performance of the new model is compared with that of the original model (such as solution accuracy, scheduler confirmation rate, and generation efficiency). If the performance improvement of the new model is not less than a set percentage (such as ≥15%), the online model will be gradually replaced through canary releases, while retaining the rollback mechanism to ensure the stability of the scheduling business.

[0051] Furthermore, in this embodiment, based on the resource scheduling driven mechanism based on critical business events, the types of critical business events can be expanded to include temporary disinfection for epidemic prevention, gate changes, air traffic control, etc., and user-defined DSLs are supported. Simultaneously, a Complex Event Processing (CEP) engine is introduced to recognize multi-event combination patterns (such as thunderstorms + alternate landings). The resource scheduling triggering strategy can switch between three modes: instant rescheduling, batch window rescheduling, and mixed triggering. Specifically, the following modules are added: Event type extension management module: includes a basic event library upgrade unit, a user-defined DSL (domain-specific language) parsing unit, and an event metadata registration unit, supporting event type extension and custom rule management; CEP Engine Module: Deploys a complex event processing engine, including an event stream access unit, a combined pattern definition unit, a pattern matching inference unit, and an event aggregation output unit, to achieve multi-event combination recognition; Trigger strategy configuration module: includes strategy definition unit, strategy switching control unit, scene-strategy mapping unit, and priority arbitration unit, supporting the configuration and dynamic switching of three trigger modes; Event-Policy Linkage Module: Establishes an association mapping between extended events, complex events, and triggering policies to ensure that the corresponding scheduling and reordering policies are accurately matched after an event is triggered.

[0052] In specific implementation, the following methods may be adopted, but are not limited to: Step 1: Add a new critical business event based on the airport's temporary events, as detailed below: (1) Add the following three types of key business events: Epidemic prevention related event: "Temporary disinfection" (Triggering conditions: Airport issues epidemic prevention disinfection notice, key flights require special disinfection; Data source: Airport epidemic prevention management system; Related scenarios: International flight support, sudden epidemic prevention needs). Aircraft stand related events: "Aircraft stand change" (triggering conditions: the actual aircraft stand is inconsistent with the planned aircraft stand, temporary occupation / release of the aircraft stand; data source: flight information system, airport aircraft stand management system; related scenario: re-adaptation of flight support resources). Air traffic control related events: "Airflow control" (Triggering conditions: Air traffic control department issues flow control notice, flight arrival / departure queues are adjusted; Data source: Air traffic control service interface; Related scenarios: Flight take-off and landing times are postponed, and support mission sequence is adjusted). Newly added critical business events are entered into the critical business event set, and event attributes (event ID, event name, trigger threshold, associated resource type, priority) are added. Event detection rules are also updated to ensure that new events can be monitored in real time.

[0053] (2) Add key business events through custom DSL: DSL Syntax Specification Definition: A simplified DSL syntax is designed to adapt to airport ground service operations, allowing users to define custom events using natural language-style statements. Syntax elements include: event name, triggering conditions (data fields, comparison operators, thresholds), data source, associated entities, and event priority; for example: custom event [VIP passenger temporary pick-up], triggering conditions [flight number = CAXXXX AND passenger type = VIP AND pick-up request report = true], data source [passenger service system], associated entities [passenger stairs, shuttle bus], priority [high]. DSL Parsing and Conversion: After the user inputs a custom DSL statement through the visual configuration interface, the DSL parsing unit converts the DSL statement into a structured event rule (including trigger condition expression, data query path, and associated entity ID) that the system can recognize through lexical analysis (identifying keywords, field names, and operators) and syntax analysis (verifying the validity of the statement). If there is a syntax error during the conversion process, an error message is returned and guidance is provided for correction. Custom event registration and activation: After the parsed structured event rules are reviewed by the administrator, they are registered to the event library and automatically associated with the data acquisition interface of the event detection module and the data weaving layer to realize real-time monitoring and triggering of custom events. At the same time, it supports lifecycle management such as editing, disabling, and deleting event rules.

[0054] Step 2: CEP engine deployment and multi-event combination pattern recognition, solving the problem that single event triggering cannot cover complex business scenarios, as detailed below: CEP Engine Selection and Configuration: Select a CEP engine that supports streaming data processing (such as Esper or Flink CEP). Based on the event flow characteristics of airport ground service business, configure the core parameters of the engine: event stream access rate (adapt to the 30-second / time data update frequency of the data weaving layer), time window size (default 1 minute, can be adjusted by user configuration), and pattern matching timeout threshold (default 3 minutes). Multi-event combination pattern definition: Through the pattern definition unit of the CEP engine, typical complex event combination patterns are preset, and users can also customize combination patterns based on extended event types. Typical pattern examples are as follows: Pattern 1: "Thunderstorm + Diversion" (Combination conditions: Meteorological event [Thunderstorm Warning] and flight event [Diversion] occur successively within 10 minutes; Associated entities: Diverted flight, temporary backup gate, shuttle bus); Pattern 2: "Air Traffic Control + Gate Change + Epidemic Prevention and Disinfection" (Combination conditions: Air traffic control event [Air Traffic Control], gate change event [Gate Change], and epidemic prevention event [Temporary Disinfection] exist simultaneously within 30 minutes; Associated entities: Affected flight, changed gate, disinfection vehicle, ground staff). Event Stream Access and Pattern Matching: Event Stream Access: The CEP engine receives single event data (including event ID, occurrence time, associated entities, and data characteristics) pushed by the data weaving layer in real time through the event stream access unit, forming a standardized event stream; Pattern Matching Inference: The engine analyzes the event stream in real time based on preset combination pattern rules, captures the event set within the window through a sliding time window, and determines whether the temporal relationship and attribute association between events meet the combination pattern conditions; for example, if a "thunderstorm warning" event is detected within a 10-minute time window, and a "diversion" event occurs within the following 5 minutes, and both are associated with the same flight, then the "thunderstorm + diversion" combination pattern is determined to be matched; Matching result aggregation output: When a matching combination pattern is detected, the CEP engine aggregates the individual event data, combined event identifier, associated entity set, and occurrence time window of the combined event into structured complex event information, pushes it to the trigger strategy configuration module, and marks the event priority (combined events have higher priority than individual events).

[0055] Step 3: Dynamically switch between three scheduling triggering strategies—instant reordering, batch window reordering, and hybrid triggering—to adapt to the scheduling response requirements of different business scenarios, as detailed below: (1) The implementation logic of the three scheduling triggering strategies is as follows: Instant Rescheduling Mode: Triggering Conditions: Applicable to high-priority events (such as VIP flight support anomalies, major meteorological disasters, and urgent epidemic prevention needs). When such events (single events or combined events) are detected, scheduling rescheduling is triggered immediately. Execution Logic: The system suspends the queue of currently unexecuted scheduling tasks, prioritizes the scope of event impact and real-time support task context data to generate a new scheduling plan, and immediately issues adjustment instructions through the scheduling execution layer, with response delay controlled within 1 minute. Configuration Parameters: A list of high-priority events is preset, and a threshold is set for rescheduling priorities to be higher than the original tasks (such as a priority difference ≥ 2).

[0056] Batch Window Reordering Mode: Triggering Conditions: Applicable to low-priority, high-frequency events (such as minor delays of regular flights or temporary maintenance of a single vehicle). Individual reordering of such events can easily lead to frequent scheduling fluctuations. Execution Logic: Set a fixed batch time window (such as 5 minutes or 10 minutes, user-configurable). Aggregate all triggered events within the window. After the window ends, perform a unified scheduling reordering to generate a batch scheduling plan that covers the impact of all events. It also supports event deduplication and merging within the window (such as merging multiple minor delay events of the same flight into a single delay event). Configuration Parameters: Batch window duration, low-priority event list, event aggregation rules.

[0057] Hybrid Trigger Mode: Trigger Condition: Suitable for scenarios where mixed priority events coexist, balancing immediate response to high-priority events with batch processing of low-priority events; Execution Logic: Preset priority arbitration rules, immediate reordering is executed immediately when a high-priority event is triggered; low-priority events enter a batch time window for aggregation, and if no high-priority event appears in the window, batch reordering is executed after the window ends; if a high-priority event appears in the window, the batch window is immediately interrupted, and while executing immediate reordering, the aggregated low-priority events are included in the scope of this reordering; Configuration Parameters: Priority division standard, batch window duration, event interruption and merging rules.

[0058] (2) Configuration and dynamic switching of triggering strategies: Preset strategies and scenario associations: The system pre-sets mapping relationships between typical business scenarios and triggering strategies (such as "VIP flight support scenario" associated with "instant reordering", "daily ordinary flight support scenario" associated with "batch window reordering", and "peak hour support scenario" associated with "hybrid triggering"). Users can directly select or customize the mapping relationship. Manual / Automatic Switching Mechanism: Administrators can manually switch triggering strategies through the system configuration interface, and the switch takes effect immediately; automatic switching is also supported. The system automatically selects the optimal triggering strategy based on real-time business load (such as the number of concurrent events and the number of scheduled tasks) and event priority distribution (such as automatically switching to real-time rearrangement when there are ≥3 concurrent high-priority events). Strategy execution monitoring: The trigger strategy configuration module records the execution status of each strategy in real time (number of triggers, reordering time, scheduling scheme confirmation rate), and generates strategy execution reports to provide data support for users to optimize strategy configuration.

[0059] Step 4: After the newly added critical business events and complex events are detected, the trigger strategy configuration module determines the reordering mode and triggers the intelligent decision layer to generate a scheduling plan. During the plan generation process, the business requirements corresponding to the extended events are automatically associated (such as the need to dispatch dedicated disinfection vehicles for epidemic prevention and disinfection, and the need to replan vehicle driving routes for changes in machine positions). After the scheduling is executed, the event processing results are fed back to the event database to provide data basis for subsequent event rule optimization.

[0060] The above solutions effectively enhance the adaptability of resource scheduling driven by critical business events, primarily in the following aspects: 1) Expanded event coverage enables precise response to special business scenarios such as epidemic prevention, gate adjustments, and traffic control; 2) The CEP engine enables complex event combination recognition, improving the accuracy of identifying complex scenarios with multiple events overlapping and avoiding scheduling omissions caused by single event triggers; 3) Three triggering strategies adapt to different priorities and business load scenarios, reducing frequent scheduling fluctuations and lowering response latency for high-priority events; 4) User-defined DSL lowers the technical threshold for event expansion, enhances the flexibility and scalability of the mechanism, and ultimately achieves a precise, efficient, and intelligent upgrade of airport ground service resource scheduling.

[0061] The intelligent scheduling method for airport ground service support resources provided in this application embodiment will be further described in detail below. The intelligent scheduling method for airport ground service support resources provided in this application embodiment is based on the core concepts of "business event-driven, multi-source data fusion, AI-assisted decision-making, human-machine collaborative confirmation, and closed-loop continuous optimization," realizing the transformation of vehicle scheduling from "static pre-scheduling" to "dynamic adaptive scheduling." Specifically, as follows: (1) Construct a business perception system with "flight-mission-resource" as the core; Each flight's support task is broken down into standardized task units (such as "Water Refill - Task ID 1001"), and task attributes are bound to them: start time window, duration, required vehicle type, personnel qualification requirements, and priority level (normal / delayed / international / VIP).

[0062] Real-time collection of task-related resource status, including but not limited to: Vehicle: Location, operating status (idle / operating / return trip), fault status, maintenance plan, vehicle type capabilities; Personnel: On-duty status, current location, skill certificates, current task, leave information; External environment: weather warnings, runway closures, and adjustments to airport operation levels.

[0063] All data is accessed through the data weaving platform to form a complete business context view.

[0064] (2) Real-time data fusion and governance based on data fabric; Data Access: Construct a unified data weaving layer to enable real-time access to multi-source data. Metadata extraction: Through automatic scanning and adapter mechanisms, metadata is extracted from various source systems (such as flight information systems, vehicle GPS systems, personnel scheduling systems, and air traffic control meteorological service interfaces), including data source structure, update frequency, field semantics, etc. Semantic mapping: Based on a predefined airport ground service domain ontology model, semantic alignment and unified identification are performed on heterogeneous data entities (such as "flight number", "vehicle ID" and "task type") to establish cross-system associations; Data quality verification: Set real-time quality rules (such as integrity checks, range checks, logical consistency checks, and timeliness checks) to automatically trigger repair or marking of abnormal data; Anomaly detection and repair: Detect data stream anomalies (such as delays, interruptions, and sudden changes in values) through streaming data processing technology, and automatically compensate or issue alarms in conjunction with business rules; Logical model construction: Based on a unified "flight-task-resource" model, the cleaned data is integrated to form a high-quality, real-time queryable data view to support upper-level scheduling analysis.

[0065] (3) Establish an event-driven scheduling and triggering mechanism; Key business events are set as trigger conditions for scheduling optimization, including: changes in actual flight arrival / departure time (deviation > 5 minutes); vehicle malfunction reports or changes in maintenance status; temporary leave or absence of personnel; sudden weather changes affecting outdoor operations; and task execution timeouts or abnormal interruptions.

[0066] Once the above events are detected, the scheduling evaluation process will be initiated immediately.

[0067] (4) Utilize large language models to generate scheduling suggestions; The current business context (including affected flights, task list, available resource pool, historical scheduling records, etc.) is input into a large language model (LLM) fine-tuned with industry knowledge. Its core lies in constructing a multi-stage reasoning and optimization framework centered on the LLM. This framework combines the semantic understanding and reasoning capabilities of the large language model with traditional operations research optimization algorithms, specifically performing the following operations: Step 1: Context Awareness and Impact Assessment (LLM-based Semantic Parsing Module); After transforming multi-source heterogeneous data (such as flight status, vehicle GPS coordinates, and personnel status) into a unified contextual text description through a data weaving layer, LLM, as a high-level semantic parser, identifies the scope of event impact in the following ways: Entity and Relationship Extraction: Leveraging the Named Entity Recognition (NER) capability of LLM, key entities (such as flight number, task ID, vehicle ID) are extracted from the text, and the relationships between entities are understood based on predefined domain graphs (such as "flight-includes-task", "task-requires-vehicle").

[0068] Impact Chain Reasoning: This technique employs Chain-of-Thought Prompting to guide LLM's step-by-step reasoning. For example, a prompt template might be: "Event [Flight CA1234 lands 15 minutes early]. Please analyze step-by-step: 1. What support tasks does this flight have? 2. What resources and personnel are currently assigned to these tasks? 3. What is the current status of these resources? 4. Will this event cause resource or time conflicts?" Based on this, LLM generates a structured impact summary (e.g., "Conclusion: The water truck task for CA1234 was originally handled by Train B, but Train B is still serving CA5678 and is expected to finish in 10 minutes. Therefore, the CA1234 task will face a delay risk of at least 5 minutes").

[0069] Step 2: Feasible solution generation (resource selection based on constraints); In this phase, the LLM works in conjunction with a rule-based resource filter. Based on the impact assessment from the first step, the LLM outputs scheduling objectives (e.g., "find an available vehicle for CA1234 to perform the water refueling task") and constraints (e.g., "the vehicle must be within the T1 terminal area," "must have water refueling capability," "the driver must hold the corresponding qualifications"). An efficient multi-attribute matching algorithm (e.g., fast lookup based on bitmap indexing) is executed in the real-time resource pool to quickly filter out the set of all candidate vehicles {C1, C2, C3...} that theoretically meet the hard constraints.

[0070] Step 3: Calculation and sorting of optimal solutions based on multi-dimensional optimization algorithms; Objective function quantification: Calculate a series of quantifiable metrics for each candidate solution, such as: Time cost: Travel time of vehicle C(i) to the task point (based on GIS path planning algorithm) + estimated task execution time.

[0071] Economic costs: fuel consumption during empty runs, labor costs, etc.

[0072] Risk Index: Calculated based on data such as vehicle battery / fuel level, recent failure rate, and driver proficiency.

[0073] LLM Weighting and Comprehensive Evaluation: LLM is introduced as a dynamic weight allocator. Based on the current context (e.g., "Is this a VIP flight?", "Is the airport about to upgrade its operational level due to weather?"), LLM simulates the decision preferences of a senior dispatcher and generates dynamic weights for each objective. For example, if the prompt is: "This is an international flight with a delayed preceding flight; please prioritize minimizing time and appropriately reduce economic cost considerations," LLM will output weight suggestions, such as Wt=0.9, We=0.1.

[0074] Subsequently, based on the dynamic weights given by LLM, each candidate solution is weighted and scored using the following formula, and the optimal and second-best solutions are obtained by ranking them according to the weighted scores.

[0075] Score=Wt Time cost + We Economic Cost - Wr Risk Index Step 4: Natural Language Generation and Interpretation (NLG module); Algorithm / Process: The LLM receives the optimal solution obtained in the third step, along with its various quantitative indicators and the reasons for weight allocation.

[0076] Through carefully designed explanatory prompt templates, LLM transforms cold numbers into natural language suggestions and credible reasons. For example: "Recommended solution: Dispatch vehicle C (license plate: Zhejiang BXXXX) to perform the CA1234 water purification mission." Reasons generated: "The reasons are as follows: 1. Highest efficiency: Car C has completed its previous task and is located at gate T1-203, only 300 meters from gate T1-211 for CA1234, expected to arrive within 2 minutes, saving 8 minutes of waiting time. 2. Controllable risk: The car has sufficient battery power (85%) and is in good condition. 3. Overall optimal: Although there is a closer car D, it is currently serving a higher-priority VIP flight, and relocating it would create a chain of risks." Multiple options comparison: LLM generates similar explanatory text for the top 2-3 options, and briefly explains why they were not chosen as the first choice (e.g., "Option 2: Dispatch car D. Advantages: Saves another 2 minutes. Disadvantages: May affect VIP flight support, higher risk."), providing dispatchers with a comprehensive decision-making perspective.

[0077] (5) Human-machine collaborative confirmation and feedback learning mechanism; Dispatchers can view the natural language suggestions and visual scheduling diagrams (such as Gantt charts and vehicle trajectories) generated by LLM on the terminal and perform the following operations: directly confirm execution; manually modify (such as changing vehicles or adjusting the order); reject and fill in the reason.

[0078] All human decision-making actions are recorded, and the comparison results between AI suggestions and human corrections are fed back to the large language model for subsequent prompt optimization or fine-tuning training, so as to achieve continuous evolution of the scheduling strategy.

[0079] (6) Automatic issuance and execution tracking of scheduling instructions; The confirmed scheduling plan is automatically broken down into specific task instructions and pushed to: The vehicle terminal or driver's app will notify you of a new task; The vehicle dispatch system updates the task queue; The scheduling system synchronizes personnel / vehicle status.

[0080] Continuously track the status of task execution to form a complete business loop of "planning → execution → feedback → optimization".

[0081] The intelligent scheduling method for airport ground service support resources provided in this application achieves the following technical effects: (1) Breaking down data silos and achieving high-quality real-time integration: Data weaving technology ensures efficient integration and continuous updating of multi-source heterogeneous data, providing a reliable data foundation for intelligent scheduling.

[0082] (2) Upgrade the business model from "static scheduling" to "dynamic scheduling": The event-driven mechanism enables scheduling in the event of an event, which can significantly improve the response speed of flight changes, reduce delays, and improve the on-time rate of flights.

[0083] (3) Improve the accuracy of resource matching and the rationality of scheduling: By comprehensively evaluating multi-dimensional factors such as space, time, capability, and status, resource waste and task conflicts can be avoided, and vehicle utilization can be improved.

[0084] (4) Enhance the interpretability and credibility of scheduling decisions: Output the reasons for the suggestions and the impact analysis in natural language form to help schedulers quickly understand and trust the AI ​​suggestions and improve the efficiency of human-machine collaboration.

[0085] (5) Accumulate scheduling experience and build an evolvable intelligent scheduling capability: Accumulate samples of “excellent scheduling decisions” through feedback mechanisms, gradually form enterprise-level scheduling knowledge assets, and reduce dependence on individual experts.

[0086] (6) Supports intelligent reasoning for complex business scenarios: The large language model has contextual understanding and multi-condition reasoning capabilities, and can handle complex abnormal scenarios such as "flight advance + vehicle failure + personnel leave", providing scientific response strategies.

[0087] (7) Improve flight punctuality and passenger satisfaction: reduce vehicle idling and waiting time, optimize task coordination, improve the efficiency of ground service vehicles, and reduce the risk of delays.

[0088] After introducing the intelligent scheduling system and method for airport ground service support resources provided in the embodiments of this application, the electronic equipment provided in the embodiments of this application will be briefly introduced next.

[0089] See Figure 3As shown, the electronic device 300 provided in this application embodiment includes at least a processor 301, a memory 302, and a computer program stored in the memory 302 and capable of running on the processor 301. When the processor 301 executes the computer program, it implements the above-mentioned intelligent scheduling method for airport ground service support resources provided in this application embodiment.

[0090] In one possible implementation, processor 301 can be a single processing element or a collective term for multiple processing elements. For example, processor 301 can be a central processing unit (CPU), or one or more integrated circuits configured to implement the intelligent scheduling method for airport ground service support resources provided in the embodiments of this application. Specifically, processor 301 can be a general-purpose processor, including but not limited to CPUs, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0091] In one possible implementation, memory 302 may include a readable medium in the form of volatile memory, such as random access memory (RAM) 3021 and / or cache memory 3022, and may further include read-only memory (ROM) 3023; memory 302 may also include a program tool 3025 having a set (at least one) of program modules 3024, including but not limited to: operating subsystem, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0092] In one possible implementation, the electronic device 300 provided in this application embodiment may further include a bus 303 connecting different components (including processor 301 and memory 302). The bus 303 represents one or more types of bus structures, including memory bus, peripheral bus, local area bus, etc.

[0093] In one possible implementation, the electronic device 300 can also communicate with one or more devices that enable a user to interact with the electronic device 300 (e.g., mobile phones, computers, etc.), and / or with external devices 304 such as devices that enable the electronic device 300 to communicate with one or more other electronic devices 300 (e.g., routers, modems, etc.). This communication can be performed via an input / output (I / O) interface 305. Furthermore, the electronic device 300 can also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 306. Figure 3 As shown, network adapter 306 communicates with other modules of electronic device 300 via bus 303. It should be understood that, although... Figure 3 As not shown, other hardware and / or software modules may be used in conjunction with electronic device 300, including but not limited to microcode, device drivers, redundant processors, external disk drive arrays, Redundant Arrays of Independent Disks (RAID) subsystems, tape drives, and data backup storage subsystems.

[0094] It should be noted that, Figure 3 The electronic device 300 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0095] Furthermore, this application embodiment also provides a computer-readable storage medium storing computer instructions. When executed by a processor, these computer instructions implement the intelligent scheduling method for airport ground service support resources provided in this application embodiment. Specifically, the computer instructions can be built into or installed in a processor, enabling the processor to implement the intelligent scheduling method for airport ground service support resources provided in this application embodiment by executing the built-in or installed computer instructions.

[0096] Of course, the intelligent scheduling method for airport ground service support resources provided in the embodiments of this application can also be implemented as a program product, which includes program code. When the program code is executed by a processor, it implements the intelligent scheduling method for airport ground service support resources provided in the embodiments of this application.

[0097] The program product provided in this application embodiment can be any combination of one or more readable media, wherein the readable media can be a readable signal medium or a readable storage medium, and the readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination thereof. Specifically, more specific examples of readable storage media (a non-exhaustive list) include: electrical connections with one or more wires, portable disks, hard disks, RAM, ROM, erasable programmable read-only memory (EPROM), optical fibers, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0098] The program product provided in this application embodiment can be a CD-ROM and include program code, and can also run on electronic devices such as computers and laptops. However, the program product provided in this application embodiment is not limited to this. In this application embodiment, the readable storage medium can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, device, or apparatus.

[0099] It should be noted that although several units or sub-units of the device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0100] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0101] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0102] Obviously, those skilled in the art can make various modifications and variations to the embodiments of this application without departing from the spirit and scope of the embodiments of this application. Therefore, if these modifications and variations to the embodiments of this application fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.

Claims

1. A method for intelligent scheduling of airport ground service support resources, characterized in that, include: Real-time access to flight dynamic data from the flight information system, vehicle status data from the vehicle positioning system, personnel scheduling data from the personnel scheduling system, and meteorological data from the meteorological service system are used as multi-source heterogeneous data. Based on the pre-constructed standardized support mission context data for each flight, the multi-source heterogeneous data is woven to obtain the real-time support mission context data for each flight; wherein, each of the standardized support mission context data includes a minimum set of structured attributes required for a support mission and its resource scheduling. When at least one key business event in a pre-defined set of key business events is detected, a natural language resource scheduling scheme is generated for the target support task of the target flight affected by each key business event, based on the real-time support task context data of each key business event and each flight, through a large language model fine-tuned by airport ground service support vehicle scheduling knowledge, and using a multi-dimensional optimization algorithm. Resource scheduling is performed according to the natural language resource scheduling scheme corresponding to the target flight's target support mission.

2. The intelligent scheduling method for airport ground service support resources as described in claim 1, characterized in that, Based on pre-built standardized support mission context data for each flight, the multi-source heterogeneous data is woven to obtain real-time support mission context data for each flight, including: Using the standardized support mission context data of each flight as a standardized data structure template, semantic mapping and instantiation are performed on the multi-source heterogeneous data to obtain real-time support mission context data containing real-time resource status for each flight.

3. The intelligent scheduling method for airport ground service support resources as described in claim 1, characterized in that, The set of key business events includes at least: actual arrival / departure time deviation of flights exceeding the deviation threshold, vehicle malfunction or change in maintenance status, temporary absence of personnel, sudden weather changes, and mission timeouts or abnormal interruptions.

4. The intelligent scheduling method for airport ground service support resources as described in claim 1, characterized in that, Based on real-time support mission context data for each key business event and each flight, and using a large language model fine-tuned with airport ground service vehicle scheduling knowledge, a multi-dimensional optimization algorithm is employed to generate a natural language resource scheduling scheme for the target support mission of the target flight affected by each key business event. This scheme includes: The context data of each key business event and each flight's real-time support task are input into the large language model. The large language model then performs the following operations: using logical reasoning prompting technology, it identifies the target support task of the target flight affected by each key business event; based on scheduling objectives and constraints, it selects a set of candidate resource scheduling schemes for the target support task of the target flight; based on the time cost, economic cost, and risk index of each candidate resource scheduling scheme in the set of candidate resource scheduling schemes, it calculates a multi-indicator weighted score for each candidate resource scheduling scheme; it selects the candidate resource scheduling scheme whose multi-indicator weighted score meets the selection criteria as the target resource scheduling scheme; and it converts the target resource scheduling scheme into a natural language resource scheduling scheme.

5. The intelligent scheduling method for airport ground service support resources as described in claim 4, characterized in that, Based on the scheduling objectives and constraints, a set of candidate resource scheduling schemes is selected for the target flight's target support task, including: A multi-attribute matching algorithm based on bitmap indexing is used to select a set of candidate resource scheduling schemes that meet the scheduling objectives and constraints for the target support task of the target flight.

6. The intelligent scheduling method for airport ground service support resources as described in claim 4, characterized in that, Converting the target resource scheduling scheme into a natural language resource scheduling scheme includes: Based on a pre-configured explanatory prompt template, the target resource scheduling scheme is converted into a natural language resource scheduling scheme that includes recommended information, multi-dimensional explanations, and comparisons of alternative schemes.

7. The intelligent scheduling method for airport ground service support resources as described in any one of claims 1-6, characterized in that, Based on the real-time support mission context data of each flight, and using a large language model fine-tuned with airport ground service support vehicle scheduling knowledge, a multi-dimensional optimization algorithm is employed to generate a resource scheduling scheme for the target support mission of the target flight affected by the aforementioned critical business event. This also includes: The resource scheduling scheme is then visualized. In response to the scheduler's modification operation on the resource scheduling scheme, a comparison dataset is formed by comparing the modified resource scheduling scheme with the resource scheduling scheme output by the large language model; The comparison dataset is fed back to the large language model to optimize the large language model through engineering strategies or to fine-tune the model training.

8. An intelligent scheduling system for airport ground service support resources, characterized in that, include: The data access layer is used to access flight dynamic data from the flight information system, vehicle status data from the vehicle positioning system, personnel scheduling data from the personnel scheduling system, and meteorological data from the meteorological service system in real time as multi-source heterogeneous data. The data weaving layer is used to weave the multi-source heterogeneous data based on the pre-built standardized support mission context data of each flight to obtain the real-time support mission context data of each flight; wherein, each of the standardized support mission context data includes a minimum set of structured attributes required for a support mission and its resource scheduling. The event detection layer is used to detect key business events in a pre-defined set of key business events in real time. The intelligent decision-making layer is used to generate a natural language resource scheduling scheme for the target support task of the target flight affected by each key business event when the event detection layer detects at least one key business event. Based on the real-time support task context data of each key business event and each flight, it uses a large language model fine-tuned by airport ground service support vehicle scheduling knowledge and adopts a multi-dimensional optimization algorithm. The scheduling execution layer is used to perform resource scheduling according to the natural language resource scheduling scheme corresponding to the target flight's target support task.

9. An electronic device, characterized in that, include: The system includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the intelligent scheduling method for airport ground service support resources as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions, which, when executed by a processor, implement the intelligent scheduling method for airport ground service support resources as described in any one of claims 1-7.

Images