A Deep Learning-Based Integrated Air-Ground Cooperative Scheduling Method
By constructing joint adaptation representation data and performing matrix matching processing based on deep learning methods, the problems of task allocation conflict and air-ground support mismatch in low-altitude scheduling are solved, realizing efficient air-ground integrated collaborative scheduling and improving the executability and on-time execution capability of scheduling schemes.
Patent Information
- Application Number
- CN202611368744.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-04
- Publication Date
- 2026-11-17
AI Technical Summary
Existing low-altitude scheduling methods struggle to achieve unified scheduling of task allocation matching strength, forward reachability, and collaborative constraints when multiple tasks are concentrated, the status of air nodes changes dynamically, and the risk of safe flight corridors fluctuates. This can easily lead to task allocation conflicts and air-to-ground support mismatches.
A deep learning-based approach is adopted to acquire and preprocess data of the tasks to be scheduled, air nodes, and ground nodes, construct joint adaptation representation data, and use the Gumbel-Sinkhorn network for matrix matching processing to generate task air node matching matrix and air-ground node collaboration matrix. Forward reachability analysis and collaboration constraint screening are then performed to generate an integrated air-ground collaborative scheduling scheme.
It improves the integrity and consistency of low-temperature air conditioning data, reduces task allocation conflicts and duplicate node occupation, enhances the executability and on-time execution capability of scheduling schemes, and reduces corridor passage conflicts.
Smart Images

Figure CN123117675A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of low-altitude air traffic control, and in particular to a deep learning-based integrated air-ground collaborative scheduling method. Background Technology
[0002] Existing low-altitude air-conditioning scheduling methods typically assign tasks based on the tasks to be scheduled and the operational status of air nodes. They select execution targets based on task location, execution time, payload requirements, remaining flight time of air nodes, and current occupancy status, and make passage decisions by considering the passage time of safe flight corridors, corridor occupancy status, and risk level. Some methods also incorporate the communication coverage and energy supply capabilities of ground nodes to provide auxiliary support to air nodes, thus forming a basic air-ground collaborative scheduling process.
[0003] However, existing methods often employ single rules or phased screening to handle the relationships between tasks, air nodes, ground nodes, and safe flight corridors, making it difficult to simultaneously express task assignment relationships, corridor access relationships, area access relationships, and ground support relationships. When multiple tasks occur simultaneously, air node statuses change dynamically, and safe flight corridors are occupied or subject to risk fluctuations, problems such as task allocation conflicts, air-to-ground support mismatches, corridor access conflicts, and unexecutable candidate solutions can easily arise. It is also difficult to balance matching strength, forward reachability, and collaborative constraint screening during unified scheduling.
[0004] Therefore, how to provide a deep learning-based air-ground integrated collaborative scheduling method is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention
[0005] One objective of this invention is to propose an air-ground integrated collaborative scheduling method based on deep learning. This invention makes full use of deep learning and reachability analysis technology to achieve air-ground collaborative scheduling, which has the advantages of accurate matching, few conflicts and high executability.
[0006] According to an embodiment of the present invention, an air-ground integrated cooperative scheduling method based on deep learning includes the following steps: Acquire and preprocess data of tasks to be scheduled, air node operation data, ground node operation data, and safe flight corridor data to generate a basic dataset for air-ground collaborative scheduling. Based on the air-ground collaborative scheduling basic dataset, the task requirement status, air node execution status, ground node support status and safe flight corridor basic status are extracted to generate an air-ground collaborative scheduling status dataset. Construct joint adaptation representation data between tasks, air nodes, ground nodes, and safe flight corridors based on the air-ground collaborative scheduling status dataset; The joint adaptation representation data is aggregated for adaptation relationships and the matching strength is quantified to generate a joint matching score. The joint matching score is input into the Gumbel-Sinkhorn network for matrix matching processing to generate the task air node matching matrix and the air-ground node collaborative matching matrix. A hybrid Zonotope is constructed based on the mission air node matching matrix, the air-ground node collaborative matching matrix, the air node execution status, and the basic status of the safe flight corridor. Forward reachability analysis and collaborative constraint screening are then performed to generate candidate scheduling schemes for air-ground collaboration. The candidate air-ground collaborative scheduling schemes are arranged into executable scheduling relationships to generate an integrated air-ground collaborative scheduling scheme.
[0007] Optionally, the air node refers to the flight object participating in the low-altitude mission execution, the ground node refers to the ground facility or ground support object that provides support for the low-altitude mission execution process, the safe flight corridor refers to the range of low-altitude paths that the air node is allowed to pass through during the mission execution process, and the preprocessing includes time alignment, spatial coordinate unification, mission area marking, node affiliation marking, abnormal data removal, missing data completion, and data normalization processing.
[0008] Optionally, the generation of the air-ground coordinated scheduling status dataset specifically includes: The task status elements of the tasks to be scheduled in the air-ground coordinated scheduling basic dataset are arranged, and the task status is marked according to the arrangement results to generate the task requirement status. The system performs node attribution orchestration and extracts execution capabilities from the operational data of air nodes, and determines the current execution availability of air nodes based on task requirements and status, thereby generating the execution status of air nodes. Based on the ground node operation data, the ground support status elements are collected to determine the support availability status of ground nodes for mission areas and air nodes, and the ground node support status is generated. Collect corridor passage status elements from the safe flight corridor data, determine the passage status of the safe flight corridor at the current time and location, and generate the basic status of the safe flight corridor. A unified state index is used to organize the mission requirement status, air node execution status, ground node support status, and safe flight corridor basic status, generating an air-ground collaborative scheduling status dataset.
[0009] Optionally, the construction of the joint adaptation representation data specifically includes: The execution relationship is arranged for the task requirement status and air node execution status in the air-ground collaborative scheduling status dataset to generate execution adaptation data between tasks and air nodes. Arrange the traffic relationships between the execution status of air nodes and the basic status of safe flight corridors to generate traffic adaptation data between air nodes and safe flight corridors; Arrange the regional arrival relationships of the basic status and mission requirements of the safe flight corridor to generate arrival adaptation data between the safe flight corridor and the mission area; The support relationships between the ground node support status and the basic status of the safe flight corridor are arranged to generate support adaptation data between the ground node and the safe flight corridor. The execution adaptation data, passage adaptation data, arrival adaptation data and support adaptation data are collected to generate joint adaptation characterization data.
[0010] Optionally, the generation of the joint matching score specifically includes: Numerical mapping and execution field aggregation are performed on the execution adaptation data of tasks and air nodes in the joint adaptation representation data to generate the execution matching strength between tasks and air nodes. The system assigns status values and aggregates access fields to the access adaptation data between air nodes and safe flight corridors to generate the access matching strength between air nodes and safe flight corridors. The arrival matching data of the safe flight corridor and the mission area are converted into distance and duration and assigned regional status values to generate the arrival matching strength between the safe flight corridor and the mission area. The support adaptation data of ground nodes and safe flight corridors are assigned coverage status values and support margin is quantified to generate the support matching strength between ground nodes and safe flight corridors. Based on the combination relationships of the same mission, the same air node, the same ground node, and the same safe flight corridor, the execution matching strength, passage matching strength, arrival matching strength, and support matching strength are summarized in layers to generate a joint matching score.
[0011] Optionally, the generation of the task air node matching matrix and the air-to-ground node collaborative matching matrix specifically includes: The joint matching scores are matrixed according to mission identifiers, air node identifiers, ground node identifiers and safe flight corridor identifiers to generate mission air node scoring matrix and air-ground node collaborative scoring matrix. The task air node scoring matrix and the air-to-ground node collaborative scoring matrix are input into the Gumbel-Sinkhorn network. The scoring values at each matrix position are perturbed and sampled through the Gumbel perturbation layer to generate the task air node perturbation scoring matrix and the air-to-ground node collaborative perturbation scoring matrix. Input the mission air node disturbance score matrix and the air-to-ground node collaborative disturbance score matrix into the temperature scaling layer, adjust the degree of discretization of the disturbance score values, and generate the mission air node scaling score matrix and the air-to-ground node collaborative scaling score matrix. Input the task air node scaling score matrix and the air-ground node co-scaling score matrix into the Sinkhorn iterative normalization layer, perform row and column alternating normalization on the score matrix, and generate the task air node double random matching matrix and the air-ground node co-scaling double random matching matrix. The matching positions of the mission air node double random matching matrix and the air-to-ground node collaborative double random matching matrix are filtered and the matrix is finalized to generate the mission air node matching matrix and the air-to-ground node collaborative matching matrix.
[0012] Optionally, the generation of the air-ground cooperative candidate scheduling scheme specifically includes: Based on the mission air node matching matrix and the air-to-ground node collaborative matching matrix, the effective matching positions are extracted, and the effective matching positions are combined and associated with the air node execution status and the basic status of the safe flight corridor to generate air-to-ground matching combination data; State range arrangement and matching relationship embedding are performed on the air-ground matching combination data to construct a hybrid Zonotope. The hybrid Zonotope is an reachability set structure composed of central states, continuous generators, discrete generators and constraints. A forward reachability analysis of the hybrid Zonotope is performed along the direction of safe flight corridor to generate a forward reachable set of air nodes; The forward reachable set of air nodes is filtered for safe flight corridor constraints, air node execution constraints, and matching relationship constraints, generating collaborative constraint filtering results. Based on the results of the collaborative constraint screening, combinations of tasks, air nodes, ground nodes, and safe flight corridors that meet the constraints are retained, and candidate fields are collected to generate air-ground collaborative candidate scheduling schemes.
[0013] Optionally, the generation of the integrated air-ground coordinated scheduling scheme specifically includes: Candidate scheduling schemes for air-ground coordination are sorted and arranged according to task priority, estimated arrival time, start time of task execution period and remaining flight time to generate candidate scheme ranking data; The task identifier, air node identifier, ground node identifier, and safe flight corridor identifier in the candidate scheme ranking data are mapped to scheduling objects to generate air-ground collaborative scheduling mapping data; Perform node occupancy verification, corridor access verification, and task time period verification on the air-ground coordinated scheduling mapping data to generate executable scheduling mapping data; Based on the executable scheduling mapping data, the scheduling execution order is arranged to generate an integrated air-ground collaborative scheduling scheme. The integrated air-ground collaborative scheduling scheme includes mission identifier, air node identifier, ground node identifier, safe flight corridor identifier, mission execution time period, mission target location, and scheduling execution order. The scheduling execution order includes mission execution order, air node execution order, ground node support order, and safe flight corridor passage order.
[0014] The beneficial effects of this invention are: This invention performs unified preprocessing on data from tasks to be scheduled, air node operation data, ground node operation data, and safe flight corridor data. It further extracts task requirement status, air node execution status, ground node support status, and basic safe flight corridor status, enabling previously scattered task information, node capability information, and corridor access information to be centrally expressed in a single air-ground collaborative scheduling status dataset. Therefore, the scheduling process no longer relies solely on distance, idle status, or manual rule judgments, but simultaneously reflects task time period, payload requirements, remaining flight time, communication status, ground support capabilities, corridor occupancy, and risk level, improving the completeness and consistency of low-altitude scheduling data representation.
[0015] This invention further constructs joint adaptation representation data among missions, air nodes, ground nodes, and safe flight corridors, and quantifies the matching strength of execution adaptation, passage adaptation, arrival adaptation, and support adaptation to generate a joint matching score. Then, a matrix matching process is performed using a Gumbel-Sinkhorn network, enabling the generation of mission air node assignment relationships and air-ground node collaboration relationships within a unified matching framework. Compared to traditional phased screening methods, this process reduces mission allocation conflicts, duplicate node occupation, and air-ground support mismatch issues, making candidate scheduling schemes more suitable for multi-object collaborative scheduling requirements.
[0016] This invention also constructs a hybrid Zonotope based on the mission air node matching matrix, the air-to-ground node collaborative matching matrix, the air node execution status, and the basic status of the safe flight corridor. Forward reachability analysis and collaborative constraint screening are then performed, enabling joint verification of the continuous movement range, remaining flight time, corridor passage boundaries, passage time, and risk status of air nodes before the scheduling scheme is generated. The final integrated air-to-ground collaborative scheduling scheme is generated through the orchestration of executable scheduling relationships, which improves mission on-time execution capability, reduces corridor passage conflicts, and enhances the executability of the scheduling scheme in actual low-altitude operation scenarios. Attached Figure Description
[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a deep learning-based air-ground integrated collaborative scheduling method proposed in this invention; Figure 2 This is a flowchart of the matrix matching process for an air-ground integrated collaborative scheduling method based on deep learning proposed in this invention. Detailed Implementation
[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.
[0019] refer to Figures 1-2 A deep learning-based air-ground integrated collaborative scheduling method includes the following steps: Acquire and preprocess data of tasks to be scheduled, air node operation data, ground node operation data, and safe flight corridor data to generate a basic dataset for air-ground collaborative scheduling. Based on the air-ground collaborative scheduling basic dataset, the task requirement status, air node execution status, ground node support status and safe flight corridor basic status are extracted to generate an air-ground collaborative scheduling status dataset. Construct joint adaptation representation data between tasks, air nodes, ground nodes, and safe flight corridors based on the air-ground collaborative scheduling status dataset; The joint adaptation representation data is aggregated for adaptation relationships and the matching strength is quantified to generate a joint matching score. The joint matching score is input into the Gumbel-Sinkhorn network for matrix matching processing to generate the task air node matching matrix and the air-ground node collaborative matching matrix. A hybrid Zonotope is constructed based on the mission air node matching matrix, the air-ground node collaborative matching matrix, the air node execution status, and the basic status of the safe flight corridor. Forward reachability analysis and collaborative constraint screening are then performed to generate candidate scheduling schemes for air-ground collaboration. The candidate air-ground collaborative scheduling schemes are arranged into executable scheduling relationships to generate an integrated air-ground collaborative scheduling scheme.
[0020] In this embodiment, an air node refers to a flight object participating in the execution of a low-altitude mission, a ground node refers to a ground facility or ground support object that provides support for the execution of a low-altitude mission, a safe flight corridor refers to the range of low-altitude paths that air nodes are allowed to pass through during the execution of their missions, and preprocessing includes time alignment, spatial coordinate unification, mission area marking, node affiliation marking, abnormal data removal, missing data completion, and data normalization. Time alignment of various data types is performed using 1 second as the uniform time granularity. Data with a time offset of no more than 5 seconds is adjusted to the corresponding time position, while data with a time offset of more than 5 seconds is removed. All spatial coordinates are transformed to a unified coordinate reference system, and coordinate values are retained to 3 decimal places. Spatial regions are then divided according to the rule of a side length of 50, and data falling into the same region are assigned the same task region label, and node affixing is completed based on the unique node identifier. Data with field values below the 2nd percentile or above the 98th percentile are removed as outliers. Linear interpolation is used to fill in data with no more than 2 consecutive missing records, while data with more than 2 consecutive missing records are not included in the filling. Minimum and maximum value normalization methods are used to normalize the data of each numerical field, mapping all values to the 0-1 range.
[0021] In this embodiment, the generation of the air-ground coordinated scheduling status dataset specifically includes: The task status elements of the tasks to be scheduled in the air-ground coordinated scheduling basic dataset are arranged, and the task status is marked according to the arrangement results to generate the task requirement status. Task status elements include task execution requirements and task area status; task execution requirements include task start location, task target location, task execution period, estimated task execution duration, task payload requirements, and task priority; task area status includes task area occupancy status and task area conflict status. Task status markers are arranged in the following order: urgent execution, pending execution, and deferred execution. Tasks whose current time is between 30 seconds before the start time and 30 seconds after the end time of their execution period and whose assigned task quantity in the task area is 0 are configured as pending execution. Tasks whose current time is more than 30 seconds before the start time of their execution period, or whose priority is not urgent and whose remaining schedulable time is not less than 1.2 times the estimated execution time, are configured as deferred execution. Tasks whose current time has reached their execution period and whose execution periods overlap by more than 10 seconds within the same task area are configured as urgent execution. The system performs node attribution orchestration and extracts execution capabilities from the operational data of air nodes, and determines the current execution availability of air nodes based on task requirements and status, thereby generating the execution status of air nodes. Execution capabilities include the current location of the air node, remaining flight time, remaining payload capacity, current mission, current flight speed, and communication connection status; The current availability status is determined as follows: Air nodes that are not executing tasks, whose current location is no more than 0.8 times the achievable distance within the remaining flight time, whose remaining flight time is no less than the time required to reach the mission's starting location, 1.1 times the sum of the estimated mission execution time and the return time, whose remaining payload capacity is no less than 1.05 times the mission payload requirement, and whose cumulative communication interruption time within the last 5 seconds does not exceed 1 second are configured as idle execution nodes. The achievable distance within the remaining flight time is the product of the current flight speed and the remaining flight time. Air nodes that have already executed tasks and whose remaining flight time and payload capacity after completing the current task still meet the requirements of any mission execution, and whose cumulative communication interruption time within the last 5 seconds does not exceed 1 second are configured as occupied execution nodes. Air nodes that still have the ability to reach the mission's starting location, continue flight, carry mission payload, and maintain communication connectivity, but whose remaining flight time, remaining payload capacity, or communication connectivity status does not fully meet the mission execution requirements, are configured as restricted execution nodes. Air nodes that do not have the ability to reach the mission's starting location, continue flight, carry mission payload, or maintain communication connectivity are configured as unexecutable nodes. Based on the ground node operation data, the ground support status elements are collected to determine the support availability status of ground nodes for mission areas and air nodes, and the ground node support status is generated. Ground support status elements include ground node location, service coverage area, number of air nodes that can be supported, communication support capability, communication coverage radius, energy supply capability, and current occupancy status; The availability status determination includes configuring ground nodes as idle support nodes: those whose service coverage area overlaps with the mission area by no less than 60% of the mission area; whose communication distance with air nodes is no greater than 0.9 times the communication coverage radius; whose current occupied number is 0; whose remaining number of air nodes that can be supported is no less than 1; and whose remaining communication support capacity and remaining energy supply capacity are both no less than 20%. Ground nodes with a capacity of no less than 20% are configured as occupied support nodes; ground nodes with a remaining number of air nodes less than 1, remaining communication support capacity less than 20%, or remaining energy supply capacity less than 20% are configured as restricted support nodes, provided that the spatial overlap between the service coverage area and the mission area is no less than 60% and the communication distance between the ground node and the air node is no greater than 0.9 times the communication coverage radius. Ground nodes with a spatial overlap between the service coverage area and the mission area less than 60% or a communication distance between the ground node and the air node greater than 0.9 times the communication coverage radius are configured as unsupportable nodes. Collect corridor passage status elements from the safe flight corridor data, determine the passage status of the safe flight corridor at the current time and location, and generate the basic status of the safe flight corridor. The corridor passage status elements include the safe flight corridor boundary, passage time range, permitted flight altitude, permitted flight speed, corridor occupancy status, corridor risk markings, corridor risk level, and safe flight corridor passage direction; The corridor risk levels are divided into Level 1, Level 2, Level 3 and Level 4. Level 1 is the risk-free passage level, Level 2 is the low-risk passage level, Level 3 is the restricted passage level, and Level 4 is the restricted passage level. The safe flight corridor accessibility status is determined in the following order: impassable, restricted access, congested access, and passable. The following corridors are configured as passable: those currently within the permitted access time, with corridor occupancy below 80% of capacity, risk level not exceeding level 2, and a minimum usable altitude range of 20m and a minimum usable speed range of 5m / s. Congested corridors are configured as follows: those with occupancy exceeding 80% but below capacity, risk level not exceeding level 2, and both usable altitude and speed meet access requirements. Restricted corridors are configured as follows: those with usable altitude range less than 20m, usable speed range less than 5m / s, or risk level higher than level 2 but not marked as restricted access. Impassable corridors are configured as follows: those currently outside the permitted access time, with corridor occupancy at capacity or marked as restricted access. The passage capacity is the number of air nodes that are allowed to pass simultaneously within the same passage time range of the safe flight corridor; A unified state index is used to organize the mission requirement status, air node execution status, ground node support status, and safe flight corridor basic status, generating an air-ground collaborative scheduling status dataset. The unified state index organization includes establishing state index relationships according to unified time and location, mission area marking, node affiliation marking, and safe flight corridor identification.
[0022] In this embodiment, the construction of joint adaptation representation data specifically includes: The execution relationship is orchestrated for the task requirement status and air node execution status in the air-ground collaborative scheduling status dataset to generate execution adaptation data between tasks and air nodes. The execution adaptation data includes the spatial distance between the task and the air node, the estimated arrival time, the task time period adaptation status, the flight time margin, the payload margin, and the communication availability status. The spatial distance between mission air nodes is the three-dimensional spatial distance between the mission's starting position and the current position of the air node. The estimated arrival time is the time obtained by dividing the spatial distance of the mission air node by the current flight speed of the air node; The task time period adaptation status includes time period adaptation status and time period mismatch status. When the expected arrival time is no later than the end time of the task execution time period and the expected completion time is no earlier than the start time of the task execution time period, it is configured as time period adaptation status; otherwise, it is configured as time period mismatch status. The flight time margin is the remaining flight time of the air node after deducting the estimated arrival time, the estimated mission execution time, and the return time in sequence; the return time is the time obtained by dividing the three-dimensional spatial distance between the mission target position and the air node's assigned return position by the current flight speed of the air node; The load margin is the remaining load capacity of the air node after deducting the mission load requirement; The communication availability status includes the communication availability status and the communication unavailability status. When the air node is in the communication connection state and the cumulative communication interruption time in the last 5 seconds does not exceed 1 second, it is configured as the communication availability status; otherwise, it is configured as the communication unavailability status. The system orchestrates the accessibility relationships between the execution status of air nodes and the basic status of safe flight corridors, generating accessibility adaptation data between air nodes and safe flight corridors. The accessibility adaptation data includes corridor entrance reachability status, corridor time adaptation status, flight parameter adaptation status, corridor occupancy adaptation status, and corridor risk adaptation status. The accessibility status of the corridor entrance includes the accessibility status and the accessibility status. The location of the safe flight corridor entrance is the boundary endpoint with the smallest spatial distance from the current location of the air node in the boundary of the safe flight corridor. When there are multiple boundary endpoints with the same distance, the boundary endpoint at the beginning of the passage direction is selected. When the distance from the current location of the air node to the location of the safe flight corridor entrance is not greater than 0.9 times the product of the current flight speed and the remaining flight time of the air node, it is configured as the accessibility status; otherwise, it is configured as the accessibility status. The corridor time-adaptation status includes time-adaptation status and time-misadaptation status. The time required for the air node to travel from its current position to the entrance of the safe flight corridor is the time obtained by dividing the three-dimensional spatial distance between the two by the current flight speed of the air node. The expected time to enter the corridor is the current time plus the required time. When the expected time to enter the corridor is between 5 seconds before the start time and 5 seconds after the end time of the passage time range, it is configured as time-adaptation status; otherwise, it is configured as time-misadaptation status. Flight parameter adaptation status includes parameter adaptation status and parameter incompatibility status. When the current flight altitude of the air node is not lower than the minimum allowable flight altitude by more than 2m and not higher than the maximum allowable flight altitude by more than 2m, and the current flight speed of the air node is not lower than the minimum allowable flight speed by more than 0.5m / s and not higher than the maximum allowable flight speed by more than 0.5m / s, it is configured as parameter adaptation status; otherwise, it is configured as parameter incompatibility status. The corridor occupancy adaptation status includes occupancy adaptation status and occupancy mismatch status. When the number of corridors occupied is less than the passage capacity and the proportion of the number of corridors occupied to the passage capacity is not higher than 90%, it is configured as occupancy adaptation status; otherwise, it is configured as occupancy mismatch status. The corridor risk adaptation status includes risk adaptation status and risk mismatch status. When the corridor risk marker is not restricted and the corridor risk level is not higher than level 2, it is configured as risk adaptation status; otherwise, it is configured as risk mismatch status. Arrange the regional arrival relationships of the basic status and mission requirement status of the safe flight corridor to generate arrival adaptation data between the safe flight corridor and the mission area. The arrival adaptation data includes the distance from the corridor exit to the mission area, the arrival time of the mission area, the coverage status of the mission time period, the occupancy adaptation status of the mission area, and the conflict adaptation status of the mission area. The distance from the corridor exit to the mission area is the three-dimensional spatial distance between the safe flight corridor exit location and the mission target location; the safe flight corridor exit location is the boundary endpoint at the end of the travel direction within the safe flight corridor boundary; The time to reach the mission area is the time obtained by dividing the distance from the corridor exit to the mission area by the travel calculation speed, which is the average of the minimum and maximum allowed flight speeds. The mission time period coverage status includes time period coverage status and time period coverage status. When the estimated departure time from the safe flight corridor plus the mission area arrival time falls within the mission execution time period, it is configured as time period coverage status; otherwise, it is configured as time period coverage status. The estimated departure time from the safe flight corridor is the time obtained by adding the estimated entry time into the corridor to the corridor travel time. The corridor travel time is the time obtained by dividing the safe flight corridor travel length by the travel calculation speed. The safe flight corridor travel length is the length along the centerline of the safe flight corridor from the entrance position to the exit position. The centerline of the safe flight corridor is the travel centerline formed by connecting the equidistant points on both sides of the safe flight corridor boundary. The task area occupancy adaptation status includes occupancy adaptation status and occupancy mismatch status. When the number of tasks allocated to the task area is 0, it is configured as occupancy adaptation status; otherwise, it is configured as occupancy mismatch status. The task area conflict adaptation status includes conflict adaptation status and conflict incompatibility status. When the overlapping duration of task execution times in the same task area is no more than 10 seconds, it is configured as conflict adaptation status; otherwise, it is configured as conflict incompatibility status. The support relationship between the ground node support status and the basic status of the safe flight corridor is arranged to generate support adaptation data between the ground node and the safe flight corridor. The execution adaptation data, passage adaptation data, arrival adaptation data and support adaptation data are collected to generate joint adaptation characterization data. Supporting adaptation data includes corridor coverage status, communication coverage status, available quantity margin, communication support margin, and energy replenishment margin. The corridor coverage status includes coverage adaptation status and coverage mismatch status. When the service coverage area of the ground node and the boundary of the safe flight corridor have spatial overlap, and the overlap length is not less than 30% of the passage length of the safe flight corridor, it is configured as coverage adaptation status; otherwise, it is configured as coverage mismatch status. The communication coverage status includes a communication coverage adapted status and a communication coverage unsuitable status. When the minimum distance from the ground node location to the boundary of the safe flight corridor is no greater than 0.9 times the communication coverage radius, and the minimum distance from the ground node location to the centerline of the safe flight corridor is no greater than 0.8 times the communication coverage radius, it is configured as a communication coverage adapted status; otherwise, it is configured as a communication coverage unsuitable status. The available capacity margin is the remaining number of available air nodes after deducting the currently occupied number; The communication support margin is the remaining communication support capacity after deducting the communication occupancy ratio corresponding to corridor occupancy from the communication support capacity. The communication occupancy ratio corresponding to corridor occupancy is the ratio of the number of corridors occupied to the number of passageways multiplied by 50%. The energy supply margin is the remaining energy supply capacity after deducting the energy supply occupancy ratio corresponding to corridor occupancy. The energy supply occupancy ratio corresponding to corridor occupancy is the ratio of the number of corridors occupied to the number of passageways multiplied by 40%.
[0023] In this embodiment, the generation of the joint matching score specifically includes: Numerical mapping and execution field aggregation are performed on the execution adaptation data of tasks and air nodes in the joint adaptation representation data to generate the execution matching strength between tasks and air nodes. Numerical mapping includes normalizing the spatial distance of the mission air node and the estimated arrival time, taking the difference between the normalized distance and 1, configuring the time period adaptation state and the communication availability state as 1, configuring the time period mismatch state and the communication unavailability state as 0, configuring the negative values in the flight time margin and the payload margin as 0, and then normalizing them. The matching strength is obtained by summing the spatial distance of the mission's air node after numerical mapping, the estimated arrival time, the mission time period adaptation status, the flight time margin, the payload margin, and the communication availability status, weighted equally according to the number of execution adaptation fields. The system assigns status values and aggregates access fields to the access adaptation data between air nodes and safe flight corridors to generate the access matching strength between air nodes and safe flight corridors. The status assignment includes setting the entry reachable status, time-adaptive status, parameter-adaptive status, occupancy-adaptive status, and risk-adaptive status to 1, and setting the entry unreachable status, time-incompatible status, parameter-incompatible status, occupancy-incompatible status, and risk-incompatible status to 0. The access matching strength is obtained by summing the access matching strength of the corridor entrance reachability status, corridor time adaptation status, flight parameter adaptation status, corridor occupancy adaptation status and corridor risk adaptation status after the status is assigned, according to the weight of the number of access matching fields. The arrival matching data of the safe flight corridor and the mission area are converted into distance and duration and assigned regional status values to generate the arrival matching strength between the safe flight corridor and the mission area. The distance-time conversion includes normalizing the distance from the corridor exit to the mission area and the time to reach the mission area, and then taking the difference between these values and 1. The area status assignment includes setting the time period coverage status, occupancy adaptation status, and conflict adaptation status to 1, and setting the time period non-coverage status, occupancy mismatch status, and conflict mismatch status to 0. The arrival matching strength is obtained by summing the numerical values of the distance from the corridor exit to the task area, the arrival time of the task area, the coverage status of the task time period, the occupancy adaptation status of the task area, and the conflict adaptation status of the task area, weighted equally according to the number of arrival adaptation fields. The support adaptation data of ground nodes and safe flight corridors are assigned coverage status values and support margin is quantified to generate the support matching strength between ground nodes and safe flight corridors. The coverage status assignment includes setting the coverage adaptation status and communication coverage adaptation status to 1, and setting the coverage mismatch status and communication coverage mismatch status to 0. Support margin normalization includes configuring negative values in the support quantity margin, communication support margin, and energy supply margin to 0 before normalization; The support matching strength is obtained by summing the numericalized corridor coverage status, communication coverage status, support quantity margin, communication support margin, and energy supply margin after equally distributing the weights according to the number of support adaptation fields. Based on the combination relationships of the same mission, the same air node, the same ground node, and the same safe flight corridor, the execution matching strength, passage matching strength, arrival matching strength, and support matching strength are summarized in layers to generate a joint matching score. The hierarchical aggregation includes summing the passage matching strength and arrival matching strength with equal weight to generate the corridor task matching strength; and summing the execution matching strength, corridor task matching strength and support matching strength with equal weight to generate the joint matching score.
[0024] In this embodiment, the generation of the mission air node matching matrix and the air-to-ground node collaborative matching matrix specifically includes: The joint matching scores are matrixed according to mission identifiers, air node identifiers, ground node identifiers and safe flight corridor identifiers to generate mission air node scoring matrix and air-ground node collaborative scoring matrix. The mission air node scoring matrix uses the mission identifier as the row index and the air node identifier as the column index, while the air-to-ground node collaborative scoring matrix uses the air node identifier as the row index and the ground node identifier as the column index. In each scoring matrix, the maximum value of the joint matching score under the same row index and the same column index is configured as the score value of the corresponding matrix position, and the safe flight corridor identifier associated with the maximum value is recorded synchronously. When the number of row indices and the number of column indices of any scoring matrix are inconsistent, virtual rows or virtual columns are added to make the number of rows and columns of the scoring matrix consistent, and the score values in the virtual rows or virtual columns are configured to 0. The task air node scoring matrix and the air-to-ground node collaborative scoring matrix are input into the Gumbel-Sinkhorn network. The scoring values at each matrix position are perturbed and sampled through the Gumbel perturbation layer to generate the task air node perturbation scoring matrix and the air-to-ground node collaborative perturbation scoring matrix. The perturbation sampling process includes generating a random number between 0.0001 and 0.9999 for each matrix position, taking the natural logarithm of the random number and then taking the opposite number to generate the first perturbation intermediate value, taking the natural logarithm of the first perturbation intermediate value and then taking the opposite number to generate the Gumbel perturbation value, and adding the Gumbel perturbation value to the score value of the corresponding matrix position to obtain the perturbation score value. Input the mission air node disturbance score matrix and the air-to-ground node collaborative disturbance score matrix into the temperature scaling layer, adjust the degree of discretization of the disturbance score values, and generate the mission air node scaling score matrix and the air-to-ground node collaborative scaling score matrix. Discretization adjustment includes dividing each disturbance score value in the mission air node disturbance score matrix and the air-to-ground node cooperative disturbance score matrix by the temperature parameter to generate a scaled score value for the corresponding matrix position, and writing the scaled score value into the corresponding position according to the original matrix row index and column index. The temperature parameter is configured to 0.5. Input the task air node scaling score matrix and the air-ground node co-scaling score matrix into the Sinkhorn iterative normalization layer, perform row and column alternating normalization on the score matrix, and generate the task air node double random matching matrix and the air-ground node co-scaling double random matching matrix. The alternating row and column normalization process includes first performing an exponential transformation on each scaled score value in the same row and normalizing it according to the sum of the values in the row, then normalizing each normalized value in the same column according to the sum of the values in the column, and repeating the row normalization and column normalization process until the difference between the sum of the values in each row and 1 is no greater than 0.01 and the difference between the sum of the values in each column and 1 is no greater than 0.01, or the number of alternating row and column normalization processes reaches 20. Each row in the task air node double random matching matrix is used to represent the assignment matching value of the same task to each air node, and each column is used to represent the occupation matching value of the same air node selected by each task. Each row in the air-ground node collaborative double random matching matrix represents the collaborative matching value of the same air node to each ground node, and each column represents the support matching value of the same ground node selected by each air node. The matching positions of the mission air node double random matching matrix and the air-ground node collaborative double random matching matrix are filtered and the matrix is finalized to generate the mission air node matching matrix and the air-ground node collaborative matching matrix. The matching position filtering includes selecting the air node position with the largest value for each task identifier in the task air node double random matching matrix, and selecting the ground node position with the largest value for each air node identifier in the air-ground node collaborative double random matching matrix; when there are multiple column positions with the same largest value under the same row index during any matching position filtering process, the column index identifiers are sorted from smallest to largest, and the column position with the sorting number 1 is selected as the matching position of that row index; The matrix shaping process includes setting the selected matrix positions to 1 and the unselected matrix positions to 0. After the matrix shaping process is completed, the virtual rows or columns in the task air node matching matrix and the virtual rows or columns in the air-to-ground node collaborative matching matrix are deleted, while the matrix positions corresponding to the real task identifier, real air node identifier and real ground node identifier are retained. The task air node matching matrix is used to represent the assignment relationship between the task and the air node, and the air-ground node collaboration matching matrix is used to represent the collaboration relationship between the air node and the ground node.
[0025] In this embodiment, the generation of air-ground cooperative candidate scheduling schemes specifically includes: Based on the mission air node matching matrix and the air-to-ground node collaborative matching matrix, the effective matching positions are extracted, and the effective matching positions are combined and associated with the air node execution status and the basic status of the safe flight corridor to generate air-to-ground matching combination data; Valid matching locations include the task and the air node assignment location with a value of 1 in the task air node matching matrix, and the air node and ground node collaborative location with a value of 1 in the air-ground node collaborative matching matrix. The combined association includes retrieving the safe flight corridor identifier associated with the corresponding joint matching score according to the task identifier, air node identifier and ground node identifier in the effective matching location, and writing the air node execution status and the safe flight corridor basic status into the same air-ground matching combined data. State range arrangement and matching relationship embedding are performed on the air-ground matching combination data to construct a hybrid Zonotope. The hybrid Zonotope is an reachability set structure composed of central states, continuous generators, discrete generators and constraints, which is used to characterize the continuous movement range of air nodes in the safe flight corridor and the discrete collaborative matching state between missions, air nodes and ground nodes. The central state consists of the current position of the air node, the current flight speed, and the position of the centerline of the safe flight corridor. The position of the centerline of the safe flight corridor is obtained by discretizing the centerline of the safe flight corridor at 5m intervals. The continuous generator consists of a position offset range, a velocity offset range, an altitude offset range, and a time offset range. The position offset range is a spatial offset range extending from the centerline of the safe flight corridor to both sides of the safe flight corridor, and is sampled along a direction perpendicular to the centerline of the safe flight corridor to both sides, with a sampling interval of 1m. The sampling points between the centerline position and the two sides are used as the position offset range. The velocity offset range is the velocity range between the minimum and maximum permissible flight speeds. The altitude offset range is the altitude range between the minimum and maximum permissible flight altitudes. The time offset range is the time range between the start and end times of the passage time range, and is discretized at a 1s granularity. The start and end times of the passage time range are both used as the endpoints of the time offset range. Discrete generators include matching discrete generators, corridor occupancy discrete generators, corridor risk discrete generators, and communication status discrete generators. Matching discrete generators are formed by the matrix values of the task air node matching matrix and the air-to-ground node collaborative matching matrix. A matrix value of 1 indicates a valid match, and a matrix value of 0 indicates an invalid match. Corridor occupancy discrete generators are formed by the corridor occupancy status. If the number of occupied corridors is less than the passage capacity, an available occupancy status is written; otherwise, an unavailable occupancy status is written. Corridor risk discrete generators are formed by corridor risk markers. If the corridor risk marker indicates restricted passage, a risk unavailable status is written; otherwise, a risk available status is written. Communication status discrete generators are formed by the communication connection status. If an air node is in a communication connection status, a communication available status is written; otherwise, a communication unavailable status is written. The constraints include the reachable space range corresponding to the safe flight corridor boundary, the reachable time range corresponding to the passage time range, the reachable altitude range corresponding to the allowable flight altitude, the reachable speed range corresponding to the allowable flight speed, and the available flight time range corresponding to the remaining flight time. A forward reachability analysis of the hybrid Zonotope is performed along the direction of safe flight corridor to generate a forward reachable set of air nodes; Forward reachability analysis involves using a unified time position as the advancement index, progressively advancing the central state along the centerline of the safe flight corridor, and synchronously propagating the position offset range, velocity offset range, altitude offset range, and time offset range in the continuous generator to the next advancement position. The advancement time interval is the time difference between adjacent unified time positions. When the time difference between adjacent unified time positions is inconsistent, the actual time difference between the current unified time position and the next unified time position is configured as the current advancement time interval. After each advancement, the current position of the air node is updated according to the current flight speed and the advancement time interval, and the remaining flight time is deducted according to the advancement time interval. The forward reachability set of the air node includes the reachable space range, reachable altitude range, reachable time range, and remaining flight time range under each advancement position. Forward reachability analysis stops when the advancement position reaches the exit position of the safe flight corridor, the remaining flight time is less than 0 seconds, the advancement time exceeds the end time of the passage time range, or the number of advancements reaches 200. The forward reachable set of air nodes is filtered for safe flight corridor constraints, air node execution constraints, and matching relationship constraints, generating collaborative constraint filtering results. The safe flight corridor constraint screening includes eliminating reachable states that exceed the safe flight corridor boundary, exceed the passage time range, do not meet the allowable flight altitude, do not meet the allowable flight speed, have a corridor risk mark of restricted passage, or have a corridor risk level higher than 2. The air node execution constraint screening includes eliminating reachable states that are less than the time required for return, are not in a communication connection state, or cannot reach the mission target position. When the three-dimensional spatial distance between the final propulsion position in the forward reachable set of the air node and the mission target position is greater than 0.9 times the product of the air node's current flight speed and remaining flight time, the corresponding reachable state will be configured as unable to reach the mission target position. Matching relationship constraint filtering includes eliminating reachable states that do not have a valid matching location, whose mission identifier is inconsistent with the mission identifier in the valid matching location, whose air node identifier is inconsistent with the air node identifier in the valid matching location, whose ground node identifier is inconsistent with the ground node identifier in the valid matching location, or whose safe flight corridor identifier is inconsistent with the safe flight corridor identifier associated with the corresponding joint matching score. Based on the results of the collaborative constraint screening, retain the combination of tasks, air nodes, ground nodes and safe flight corridors that meet the constraints, and collect candidate fields to generate air-ground collaborative candidate scheduling schemes. The candidate scheduling schemes for air-ground coordination include the task identifier, air node identifier, ground node identifier, safe flight corridor identifier, task execution period, task target location, task priority, estimated arrival time, remaining flight time, and the screening results of coordination constraints.
[0026] In this embodiment, the generation of the integrated air-ground coordinated scheduling scheme specifically includes: Candidate scheduling schemes for air-ground coordination are sorted and arranged according to task priority, estimated arrival time, start time of task execution period and remaining flight time to generate candidate scheme ranking data; The candidate scheme ranking data is arranged in order of task priority from high to low, estimated arrival time from small to large, task execution period start time from early to late, and remaining flight time from large to small. A candidate ranking number is assigned to each air-ground collaborative candidate scheduling scheme according to the order of the ranking. The task identifier, air node identifier, ground node identifier, and safe flight corridor identifier in the candidate scheme ranking data are mapped to scheduling objects to generate air-ground collaborative scheduling mapping data; The scheduling object mapping includes establishing the task air node assignment relationship between the task identifier and the air node identifier in the same candidate scheme ranking data, establishing the air-ground node coordination relationship between the air node identifier and the ground node identifier, establishing the corridor passage relationship between the air node identifier and the safe flight corridor identifier, and associating the task execution time period, task target location, estimated arrival time and candidate ranking number with the same air-ground coordinated scheduling mapping data. Perform node occupancy verification, corridor access verification, and task time period verification on the air-ground coordinated scheduling mapping data to generate executable scheduling mapping data; Node occupancy verification includes retaining the first-ranked air-to-ground collaborative scheduling mapping data corresponding to the same air node identifier within the same task execution period, and removing duplicate air-to-ground collaborative scheduling mapping data that are ranked later. Corridor access verification includes retaining air-ground collaborative scheduling mapping data where the number of corridor occupancy is less than the passage capacity, the corridor risk marker is not restricted to passage, and the corridor risk level is not higher than level 2. The task time period verification includes generating the estimated arrival time in the task area based on the current time and the estimated arrival time, and retaining the air-ground collaborative scheduling mapping data that falls within the task execution time period; Based on the executable scheduling mapping data, the scheduling execution order is arranged to generate an integrated air-ground collaborative scheduling scheme. The integrated air-ground collaborative scheduling scheme includes mission identifier, air node identifier, ground node identifier, safe flight corridor identifier, mission execution time period, mission target location, and scheduling execution order. The scheduling execution order includes mission execution order, air node execution order, ground node support order, and safe flight corridor passage order. The task execution order is generated from the candidate sorting number associated with the executable scheduling mapping data in ascending order; the air node execution order is arranged from the earliest to the latest start time of the task execution period under the same air node identifier; the ground node support order is arranged from the earliest to the latest start time of the task execution period under the same ground node identifier; and the safe flight corridor passage order is arranged from the earliest to the latest start time of the task execution period under the same safe flight corridor identifier.
[0027] Example 1: To verify the feasibility of this invention in practice, it was applied to a low-altitude emergency patrol area in a city. Between 14:00 and 14:20, this area simultaneously had multiple tasks awaiting dispatch, multiple airborne nodes, multiple ground nodes, and multiple safe flight corridors. The situation presented problems such as concentrated task execution periods, significant differences in remaining flight time among airborne nodes, some safe flight corridors approaching their capacity limits, and uneven support range of ground nodes. These issues easily led to conflicts in task assignment, corridor access, and mismatches in air-to-ground support relationships.
[0028] In this scenario, operational data was collected for 6 scheduled tasks, 5 air nodes, 3 ground nodes, and 4 safe flight corridors. The scheduled tasks were distributed between 14:02 and 14:18, with estimated execution times ranging from 120 to 420 seconds, payload requirements from 1.5 kg to 4.0 kg, and priority values from 1 to 5. The air nodes had remaining flight time ranging from 680 to 1850 seconds, current flight speeds from 8 to 16 m / s, and remaining payload capacities from 2.0 kg to 6.0 kg. Air node 03 was currently executing a task, while air node 05 experienced a 0.8-second communication interruption within the last 5 seconds. The ground nodes could support 2, 3, and 1 air nodes, with communication coverage radii of 800 m, 1200 m, and 600 m, respectively. The safe flight corridors had a capacity of 2, 3, 2, and 1, with risk levels of 1, 2, 3, and 1, respectively.
[0029] In application, the data of tasks to be scheduled, air node operation data, ground node operation data, and safe flight corridor data are first processed into a unified air-ground collaborative scheduling basic dataset, and the task requirement status, air node execution status, ground node support status, and safe flight corridor basic status are extracted. Subsequently, joint adaptation representation data among tasks, air nodes, ground nodes, and safe flight corridors are constructed, and the matching strength of execution adaptation, passage adaptation, arrival adaptation, and support adaptation is quantified to form a joint matching score.
[0030] During the scheduling generation process, the joint matching score is input into the Gumbel-Sinkhorn network to generate the task air node matching matrix and the air-ground node collaborative matching matrix. Then, a hybrid Zonotope is constructed by combining the air node execution status and the basic status of the safe flight corridor. Forward reachability analysis and collaborative constraint screening are performed on the candidate combinations to eliminate combinations with excessively high corridor risk levels, insufficient remaining flight time, node occupation conflicts, or unmet task time periods. Finally, an integrated air-ground collaborative scheduling scheme is formed, which includes task identifier, air node identifier, ground node identifier, safe flight corridor identifier, task execution time period, task target location, and scheduling execution order.
[0031] In the above embodiment scenario, the continuous air-ground coordinated scheduling process from 14:00 to 14:20 was statistically analyzed, and the traditional distance-first scheduling method was used as a comparison method to obtain the following performance comparison data.
[0032] Table 1. Performance Comparison of the Invention and Traditional Air-Ground Cooperative Scheduling Methods
[0033] As can be clearly seen from Table 1, the method of the present invention is superior to the traditional method in many indicators.
[0034] The on-time execution rate of the method of this invention is increased from 86.4% of the traditional method to 90.8%, an increase of 4.4 percentage points. The reason for the improvement is that this invention does not dispatch tasks based solely on distance or node idle status, but combines task execution time, estimated arrival time, flight time margin and safe flight corridor status for joint matching, reducing the situation where tasks are assigned to unsuitable air nodes.
[0035] The average response time of the air node decreased from 38.6s to 34.2s, a reduction of 4.4s. This is because the spatial distance and estimated arrival time between the task and the air node have been pre-quantified in the joint matching score. The Gumbel-Sinkhorn network prioritizes forming a more suitable task air node assignment relationship during matrix matching, thus reducing invalid waiting and reallocation after scheduling.
[0036] The number of times a node is repeatedly occupied is reduced from 9 to 6, and the number of times a corridor is conflicted is reduced from 7 to 5. This shows that the present invention adds node occupancy verification and corridor access verification before generating the final solution, which can eliminate mapping data that repeatedly occupies the same air node or occupies a high-risk corridor within the same time period.
[0037] The air-to-ground support matching efficiency improved from 88.1% to 91.6%, an increase of 3.5 percentage points. The candidate solution executable retention rate improved from 82.7% to 86.9%, an increase of 4.2 percentage points. This is mainly because the present invention incorporates ground node support capacity, safe flight corridor carrying capacity, and forward reachability set into the screening process, making the final retained solutions closer to the actual executable state. The average remaining flight time margin improved from 146.5s to 158.2s, an increase of 11.7s, demonstrating that the present invention takes into account both the return flight time and remaining flight time constraints during scheduling.
[0038] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A deep learning-based air-ground integrated collaborative scheduling method, characterized in that, Includes the following steps: Acquire and preprocess data of tasks to be scheduled, air node operation data, ground node operation data, and safe flight corridor data to generate a basic dataset for air-ground collaborative scheduling. Based on the air-ground collaborative scheduling basic dataset, the task requirement status, air node execution status, ground node support status and safe flight corridor basic status are extracted to generate an air-ground collaborative scheduling status dataset. Construct joint adaptation representation data between tasks, air nodes, ground nodes, and safe flight corridors based on the air-ground collaborative scheduling status dataset; The joint adaptation representation data is aggregated for adaptation relationships and the matching strength is quantified to generate a joint matching score. The joint matching score is input into the Gumbel-Sinkhorn network for matrix matching processing to generate the task air node matching matrix and the air-ground node collaborative matching matrix. A hybrid Zonotope is constructed based on the mission air node matching matrix, the air-ground node collaborative matching matrix, the air node execution status, and the basic status of the safe flight corridor. Forward reachability analysis and collaborative constraint screening are then performed to generate candidate scheduling schemes for air-ground collaboration. The candidate air-ground collaborative scheduling schemes are arranged into executable scheduling relationships to generate an integrated air-ground collaborative scheduling scheme.
2. The air-ground integrated collaborative scheduling method based on deep learning according to claim 1, characterized in that, The airborne nodes refer to the flight objects participating in the execution of low-altitude missions, the ground nodes refer to the ground facilities or ground support objects that provide support for the execution of low-altitude missions, the safe flight corridor refers to the range of low-altitude paths that airborne nodes are allowed to pass through during the execution of missions, and the preprocessing includes time alignment, spatial coordinate unification, mission area marking, node affiliation marking, abnormal data removal, missing data completion, and data normalization processing.
3. The air-ground integrated collaborative scheduling method based on deep learning according to claim 1, characterized in that, The generation of the air-ground coordinated scheduling status dataset specifically includes: The task status elements of the tasks to be scheduled in the air-ground coordinated scheduling basic dataset are arranged, and the task status is marked according to the arrangement results to generate the task requirement status. The system performs node attribution orchestration and extracts execution capabilities from the operational data of air nodes, and determines the current execution availability of air nodes based on task requirements and status, thereby generating the execution status of air nodes. Based on the ground node operation data, the ground support status elements are collected to determine the support availability status of ground nodes for mission areas and air nodes, and the ground node support status is generated. Collect corridor passage status elements from the safe flight corridor data, determine the passage status of the safe flight corridor at the current time and location, and generate the basic status of the safe flight corridor. A unified state index is used to organize the mission requirement status, air node execution status, ground node support status, and safe flight corridor basic status, generating an air-ground collaborative scheduling status dataset.
4. The air-ground integrated collaborative scheduling method based on deep learning according to claim 1, characterized in that, The construction of the joint adaptation representation data specifically includes: The execution relationship is arranged for the task requirement status and air node execution status in the air-ground collaborative scheduling status dataset to generate execution adaptation data between tasks and air nodes. Arrange the traffic relationships between the execution status of air nodes and the basic status of safe flight corridors to generate traffic adaptation data between air nodes and safe flight corridors; Arrange the regional arrival relationships of the basic status and mission requirements of the safe flight corridor to generate arrival adaptation data between the safe flight corridor and the mission area; The support relationships between the ground node support status and the basic status of the safe flight corridor are arranged to generate support adaptation data between the ground node and the safe flight corridor. The execution adaptation data, passage adaptation data, arrival adaptation data and support adaptation data are collected to generate joint adaptation characterization data.
5. The air-ground integrated collaborative scheduling method based on deep learning according to claim 1, characterized in that, The generation of the joint matching score specifically includes: Numerical mapping and execution field aggregation are performed on the execution adaptation data of tasks and air nodes in the joint adaptation representation data to generate the execution matching strength between tasks and air nodes. The system assigns status values and aggregates access fields to the access adaptation data between air nodes and safe flight corridors to generate the access matching strength between air nodes and safe flight corridors. The arrival matching data of the safe flight corridor and the mission area are converted into distance and duration and assigned regional status values to generate the arrival matching strength between the safe flight corridor and the mission area. The support adaptation data of ground nodes and safe flight corridors are assigned coverage status values and support margin is quantified to generate the support matching strength between ground nodes and safe flight corridors. Based on the combination relationships of the same mission, the same air node, the same ground node, and the same safe flight corridor, the execution matching strength, passage matching strength, arrival matching strength, and support matching strength are summarized in layers to generate a joint matching score.
6. The air-ground integrated collaborative scheduling method based on deep learning according to claim 1, characterized in that, The generation of the task air node matching matrix and the air-to-ground node collaborative matching matrix specifically includes: The joint matching scores are matrixed according to mission identifiers, air node identifiers, ground node identifiers and safe flight corridor identifiers to generate mission air node scoring matrix and air-ground node collaborative scoring matrix. The task air node scoring matrix and the air-to-ground node collaborative scoring matrix are input into the Gumbel-Sinkhorn network. The scoring values at each matrix position are perturbed and sampled through the Gumbel perturbation layer to generate the task air node perturbation scoring matrix and the air-to-ground node collaborative perturbation scoring matrix. Input the mission air node disturbance score matrix and the air-to-ground node collaborative disturbance score matrix into the temperature scaling layer, adjust the degree of discretization of the disturbance score values, and generate the mission air node scaling score matrix and the air-to-ground node collaborative scaling score matrix. Input the task air node scaling score matrix and the air-ground node co-scaling score matrix into the Sinkhorn iterative normalization layer, perform row and column alternating normalization on the score matrix, and generate the task air node double random matching matrix and the air-ground node co-scaling double random matching matrix. The matching positions of the mission's air node double random matching matrix and the air-to-ground node collaborative double random matching matrix are filtered and the matrix is finalized to generate the mission's air node matching matrix and the air-to-ground node collaborative matching matrix.
7. The air-ground integrated collaborative scheduling method based on deep learning according to claim 1, characterized in that, The generation of the air-ground collaborative candidate scheduling scheme specifically includes: Based on the mission air node matching matrix and the air-to-ground node collaborative matching matrix, the effective matching positions are extracted, and the effective matching positions are combined and associated with the air node execution status and the basic status of the safe flight corridor to generate air-to-ground matching combination data; State range arrangement and matching relationship embedding are performed on the air-ground matching combination data to construct a hybrid Zonotope. The hybrid Zonotope is an reachability set structure composed of central states, continuous generators, discrete generators and constraints. A forward reachability analysis of the hybrid Zonotope is performed along the direction of safe flight corridor to generate a forward reachable set of air nodes; The forward reachable set of air nodes is filtered for safe flight corridor constraints, air node execution constraints, and matching relationship constraints, generating collaborative constraint filtering results. Based on the results of the collaborative constraint screening, combinations of tasks, air nodes, ground nodes, and safe flight corridors that meet the constraints are retained, and candidate fields are collected to generate air-ground collaborative candidate scheduling schemes.
8. The air-ground integrated collaborative scheduling method based on deep learning according to claim 1, characterized in that, The generation of the integrated air-ground coordinated scheduling scheme specifically includes: Candidate scheduling schemes for air-ground coordination are sorted and arranged according to task priority, estimated arrival time, start time of task execution period and remaining flight time to generate candidate scheme ranking data; The task identifier, air node identifier, ground node identifier, and safe flight corridor identifier in the candidate scheme ranking data are mapped to scheduling objects to generate air-ground collaborative scheduling mapping data; Perform node occupancy verification, corridor access verification, and task time period verification on the air-ground coordinated scheduling mapping data to generate executable scheduling mapping data; Based on the executable scheduling mapping data, the scheduling execution order is arranged to generate an integrated air-ground collaborative scheduling scheme. The integrated air-ground collaborative scheduling scheme includes mission identifier, air node identifier, ground node identifier, safe flight corridor identifier, mission execution time period, mission target location, and scheduling execution order. The scheduling execution order includes mission execution order, air node execution order, ground node support order, and safe flight corridor passage order.
