Low-altitude airspace grid resource dynamic scheduling method and system

Abstract

The application belongs to the technical field of airspace resource scheduling, and particularly relates to a low-altitude airspace grid resource dynamic scheduling method and system. The method comprises the following steps: discretizing the airspace into a three-dimensional grid, discretizing the time axis into time slots, generating a space-time resource block set with dynamic capacity constraints, and constructing a space-time reachable graph; receiving a data packet containing a main scheme and mutually exclusive alternative schemes, each scheme representing a resource block sequence and being associated with an original weight; checking connectivity based on the space-time reachable graph, and reasonably filtering the original weight according to the historical weight of the operation terminal to obtain a filtered weight; constructing an integer programming model with the goal of maximizing the weighted total efficiency, using Lagrangian relaxation decomposition iteration to solve, triggering heuristic backtracking when the delay or iteration threshold is met, and outputting an allocation scheme; calculating a grid congestion index, and dynamically adjusting the pre-check threshold of the candidate scheme in the next period based on the historical occupation frequency.

Description

Technical Field

[0001] This invention belongs to the field of airspace resource scheduling technology, specifically relating to a method and system for dynamic scheduling of low-altitude airspace grid resources. Background Technology

[0002] With the large-scale deployment of drone logistics and urban air traffic, low-altitude airspace scheduling systems are facing computational bottlenecks caused by high-concurrency flight plan data packets. Existing technologies have revealed the following technical shortcomings in engineering implementation that are directly related to the computational performance of optimization solvers:

[0003] First, traditional route-level models do not discretize and couple the aircraft maneuver envelope, dynamic safety interval, and three-dimensional spatiotemporal grid, resulting in a large deviation between the constraint matrix of the upper-level integer programming model and the actual physical feasible region.

[0004] Secondly, the operation terminal can only submit fixed route data. Once the local grid capacity is saturated, the system needs to initiate global conflict replanning, which leads to a sharp expansion of the solver search space in large-scale concurrent scenarios.

[0005] Third, the mission weight parameters in the flight plan data package lack reasonableness verification based on historical operational data, and abnormally high weight data will distort the gradient of the optimization objective function;

[0006] Fourth, the existing scheduling mechanism cannot identify the long-term, high-frequency occupation of hot grids by a single operator, which leads to a continuous shrinking of the feasible domain space in subsequent scheduling cycles.

[0007] This invention aims to provide a dynamic scheduling method and system for three-dimensional airspace grid resources based on combined constraints and feedback control. By constructing a spatiotemporal grid constraint matrix, performing solver-oriented input filtering, employing Lagrange relaxation decomposition and heuristic backoff collaborative solving, and introducing a constraint boundary dynamic adjustment mechanism based on occupancy frequency, it solves technical problems such as solver input distortion, iterative oscillation, time delay runaway, and feasible region degradation in existing technologies. It achieves efficient, stable, and verifiable allocation of low-altitude airspace resources and ensures the robustness of the solution queue of the cloud-edge collaborative computing platform.

[0008] Therefore, it is very important to design a dynamic scheduling method and system for low-altitude airspace grid resources that can achieve efficient, stable, and verifiable allocation of low-altitude airspace resources and ensure the robustness of the solution queue of the cloud-edge collaborative computing platform. Summary of the Invention

[0009] This invention aims to overcome the technical problems of input distortion, iteration oscillation, time delay out-of-control, and feasible region degradation in existing solvers. It provides a method and system for dynamic scheduling of low-altitude airspace grid resources that can achieve efficient, stable, and verifiable allocation of low-altitude airspace resources and ensure the robustness of the solver queue of the cloud-edge collaborative computing platform.

[0010] To achieve the above-mentioned objectives, the present invention adopts the following technical solution:

[0011] The method for dynamic scheduling of low-altitude airspace grid resources includes the following steps:

[0012] S1 discretizes the target spatial domain into a three-dimensional mesh, discretizes the time axis into time slots, generates a set of spatiotemporal resource blocks carrying dynamic capacity constraints, and constructs a spatiotemporal reachability graph;

[0013] S2, Receive flight plan data packet, the data packet contains a main scheme and at least one alternative scheme mutually exclusive with the main scheme, wherein each scheme represents a sequence of spatiotemporal resource blocks occupied by the aircraft, and each scheme is associated with an original weight value characterizing the priority of the scheme; Perform connectivity verification on each scheme based on the spatiotemporal reachability map, and perform rationality filtering on the original weight value based on the historical weight data of the operation terminal to obtain the filtered weight;

[0014] S3, with the goal of maximizing the weighted total efficiency, constructs an integer programming model constrained by the dynamic capacity constraint and the scheme mutual exclusion constraint, uses Lagrange relaxation decomposition for iterative solution, and triggers a heuristic backoff mechanism when the upper bound of the delay or the threshold of the number of iterations is met, and outputs the resource allocation scheme;

[0015] S4 calculates the grid congestion index and dynamically adjusts the pre-detection threshold for candidate schemes to enter the optimization model in the next scheduling cycle based on the historical frequency of resource block occupancy by the operation terminal.

[0016] Preferably, in step S1, the dynamic capacity constraint is determined based on the space reuse coefficient, the aircraft's nominal cruise speed, the time slot step size, the safety interval, the weather attenuation factor, and the terrain constraint factor; wherein, the safety interval is calculated based on the aircraft's relative speed, braking time, and safety margin.

[0017] Preferably, step S2 further includes the following process:

[0018] Before performing rationality filtering, capacity pre-check is performed on each resource block contained in the main scheme in the received flight plan data packet. If the current allocated load of any resource block reaches the preset ratio threshold of the corresponding dynamic capacity constraint, the main scheme is marked as low feasibility and a switch to the alternative scheme corresponding to the main scheme in the data packet is triggered.

[0019] The rationality filtering specifically includes the following process:

[0020] Maintain a sliding window of historically submitted weight values ​​for the operating terminal, calculate the mean and standard deviation of the weight values ​​within the window, and use the larger of the standard deviation and the preset minimum disturbance scale as the effective standard deviation; when the deviation between the current original weight value and the mean exceeds a preset threshold multiple multiplied by the effective standard deviation, the original weight value is decayed according to the exponential decay factor, the decay degree increases monotonically with the excess amount, and remains continuous at the boundary point.

[0021] Preferably, in the Lagrange relaxation decomposition process, the capacity constraint is relaxed into the objective function by introducing multipliers, and the integer programming model is decomposed into independent subproblems of each operating terminal, with no coupling variables between the subproblems; the multipliers are updated using the subgradient method, and the update step size is determined by multiplying the ratio of the current dual gap to the square of the L2 norm of the subgradient vector by a scaling factor; the dual gap refers to the difference between the objective value of the current dual problem and the objective value of the current optimal feasible solution.

[0022] Preferably, in step S3, the heuristic fallback mechanism includes the following process:

[0023] When the number of iterations reaches the preset maximum value, if the duality gap is still greater than the preset threshold or the real-time solution takes longer than the upper limit of the delay, a greedy allocation is performed according to the efficiency density of each scheme, and a neighborhood search is performed on the unallocated tasks by time offset or spatial offset to the adjacent grid to ensure that a feasible solution is output within the specified delay; the efficiency density is the filtered weight of each scheme divided by the number of resource blocks occupied by the scheme.

[0024] Preferably, the process of dynamically adjusting the pre-detection threshold in step S4 is as follows:

[0025] Based on the congestion index of the resource block and the historical occupancy frequency factor of the operating terminal for the resource block, the weighted product of the congestion index and the historical occupancy frequency factor is calculated as the equivalent occupancy coefficient, wherein the weight coefficients of the congestion index and the frequency factor are preset values, and the historical occupancy frequency factor is the proportion of the terminal's historical occupancy times to the total occupancy times of the resource block.

[0026] In the pre-screening phase of the next scheduling cycle, a candidate solution is allowed to enter the formal optimization model only if the current allocated load of the resource block plus the equivalent occupancy coefficient does not exceed the capacity constraint.

[0027] As a preferred option, the following process is also included:

[0028] Based on the resource blocks in the resource allocation scheme that are in a critical saturation state, a core constraint set is formed. The corresponding Lagrange multipliers are extracted as shadow prices, a quadratic programming model is constructed to generate resource occupancy control certificates, and the certificates are sent to the aircraft flight control terminal for resource occupancy legality verification during the execution phase. The critical saturation state refers to the ratio of the currently allocated load of a resource block to the dynamic capacity constraint of the resource block being no less than a preset threshold.

[0029] Preferably, non-negative slack variables are introduced into the quadratic programming model; the objective function of the quadratic programming model includes a squared deviation term between the voucher and the reference value and a penalty term for the slack variables; for any subset of the core constraint set, the constraints ensure that the marginal cost of the subset does not exceed the sum of all vouchers associated with the resource blocks in the subset plus the slack variables.

[0030] The reference value is the sum of the Lagrange multipliers corresponding to each resource block occupied by the scheme after filtering the weight of each scheme in the resource allocation scheme. If the value is less than 0, it is taken as 0; otherwise, the value itself is taken.

[0031] As a preferred option, the following process is also included:

[0032] Construct a data packet association graph with operating terminals as nodes and the relationships between operating terminals as edges; each edge connects two operating terminal nodes, and the weight of the edge is calculated in the following way:

[0033] Calculate the ratio of the intersection to the union of the resource block sets occupied by the two connected terminals; calculate the correlation coefficient between the difference sequences of the weight values ​​after filtering in adjacent scheduling cycles of the two terminals; and perform a weighted summation of the ratio of the intersection to the union and the correlation coefficient to obtain the weight of the edge.

[0034] The community detection algorithm divides the community into communities. When the average edge weight in a community exceeds a preset threshold, the community is determined to be a high-similarity collaborative input pattern. A queue downgrade mark is generated for the data packets corresponding to the operating terminals in the community. These data packets are scheduled to a low-priority queue, and the number of computing threads occupied by these data packets is limited.

[0035] This invention also provides a dynamic scheduling system for low-altitude airspace grid resources, including:

[0036] The mesh modeling module is used to discretize the target spatial domain into a three-dimensional mesh, generating a set of spatiotemporal resource blocks carrying dynamic capacity constraints; The data filtering module is used to receive flight plan data packets containing the main plan and alternative plans, and to perform rationality filtering on the weight values ​​of each plan based on the historical weight data of the operation terminal. An optimized solution engine is used to construct an integer programming model constrained by the dynamic capacity constraint and the scheme mutual exclusion constraint. Iterative solution is performed using Lagrange relaxation decomposition, and a heuristic backoff mechanism is enabled when the backoff condition is triggered to output a resource allocation scheme. The load feedback module is used to calculate the grid congestion index and dynamically adjust the pre-detection threshold for the next scheduling cycle based on the historical occupancy frequency of resource blocks by the operation terminal. The voucher generation module is used to generate resource occupancy control vouchers based on the resource allocation scheme. The queue isolation module is used to identify cooperative input patterns through packet association graphs, adjust the priority of the solver queue, and limit the number of computation threads.

[0037] Compared with the prior art, the beneficial effects of this invention are: (1) This invention adopts a continuous unified formula through a filtering mechanism, which effectively suppresses abnormal fluctuations in the weight of the objective function, eliminates numerical instability caused by discontinuities in the original piecewise formula, reduces the number of oscillations in multiplier updates by 68%, and significantly improves the convergence speed of dual gaps; (2) This invention coordinates Lagrange decomposition and neighborhood backoff, taking into account both global optimal search and time delay upper bound control, and the solution time is stabilized at 1.76s under extreme conditions, meeting the UTM second-level response requirements; (3) This invention actively manages the constraint boundary relaxation through a normalized frequency feedback mechanism, avoids the continuous tightening of hot spot grid capacity hard constraints, reduces the constraint degradation rate of 2 hours of continuous operation to 11.7%, and ensures the stability of the condition number for long-term operation of the solver; (4) This invention protects against zero denominators in the association graph, identifies and links with queue degradation markers, blocks the interference of abnormal inputs in the solution iteration process, increases the task allocation success rate to 94.2%, and improves the computational robustness of the system. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the overall architecture of the low-altitude airspace grid resource dynamic scheduling system of the present invention; Figure 2 This is a flowchart illustrating a method for dynamic scheduling of low-altitude airspace grid resources in this invention. Figure 3 This is a schematic diagram of a process for solving the Lagrange relaxation decomposition in this invention. Detailed Implementation

[0039] To more clearly illustrate the embodiments of the present invention, specific implementation methods will be described below with reference to the accompanying drawings. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without any creative effort.

[0040] like Figure 2As shown, this invention provides a method for dynamic scheduling of low-altitude airspace grid resources, specifically including the following steps:

[0041] Step 1: Construct a low-altitude spatiotemporal grid model for optimization solutions:

[0042] 1-1: Divide the target airspace by horizontal steps , Vertical step size Divided into a three-dimensional mesh set The timeline is arranged by step. Discretized into time slot sequences Define spatiotemporal resource blocks. To standardize subscripts index. Based on the minimum safe crossing time of the aircraft, the value range is: .

[0043] 1-2: For each resource block Binding dynamic capacity constraints (Right now , This serves as a hard constraint boundary for the optimization model. The capacity represents the maximum number of aircraft that can be accommodated within a single time slot along a single spatial flow direction, calculated as follows: ;

[0044] in, The spatial reuse factor represents the number of independent flight directions that can be supported simultaneously within the grid, typically ranging from 1 to 4. The nominal cruising speed for airspace; For dynamic safety intervals, the typical calculation method is as follows: ;

[0045] in, For the relative speed of the aircraft, Braking time (given by the aircraft performance envelope). For safety margin; This is the meteorological attenuation factor (issued in real time by the meteorological data platform). This is a terrain / no-fly zone structural constraint factor.

[0046] 1-3: Constructing a directed spacetime reachable graph .side The necessary and sufficient condition for its existence is that the aircraft can, under the constraints of climb rate, turning radius, and velocity envelope, proceed from... Legal motor vehicle to ,in and These are two spatiotemporal resource block nodes in the spatiotemporal reachability graph. This graph serves as the topological search space for solving subsequent subproblems.

[0047] Step 2: Structuring and Solver-Oriented Input Filtering of Combined Data Packets

[0048] 2-1: Receiving Operation Terminal Flight plan data package . The set of resource blocks for the main solution; The weights of the original objective function generated by the terminal based on the task type code, load sensor, and battery status; It is a set of mutually exclusive alternative solutions.

[0049] 2-2: Perform multidimensional verification and filtering. First, in First, verify the path continuity; if it fails, return a refactoring instruction. Second, perform capacity boundary pre-check; if... Allocated amount of any resource block If it fails, it is marked as low feasibility and a switch to an alternative solution is triggered. Furthermore, the maximum number of concurrent data packets is generated based on the device registration code and historical performance records. Intercept and split submissions. Finally, maintain the historical weighted sliding window samples of the operator, including the window length. Calculate the mean with standard deviation To avoid division by zero issues when using new terminals or when historical weights remain constant for a long time, we first define the effective standard deviation: ;

[0050] in, This is the minimum disturbance scale preset for the system. Then... Perform the following uniform continuous filtering: ;

[0051] in, , used to adapt to anomaly detection boundaries; This is the gradient stabilization decay coefficient. This formula guarantees that: when the deviation... hour, , , ;When the deviation When the attenuation factor decreases monotonically with the excess, and is continuous at the boundary point, it avoids the model becoming uncomputable due to the standard deviation being zero.

[0052] Step 3: Constraint optimization solution engine based on Lagrange relaxation decomposition:

[0053] Step 3: Based on the structured scheme and filtering weights obtained in Step 2, construct a 0-1 integer programming model, solve it iteratively using Lagrange relaxation decomposition, and ensure the upper bound of computational delay through a built-in heuristic backoff mechanism, outputting the final allocation scheme. With dual variables The solution process for Lagrange relaxation decomposition is as follows: Figure 3 As shown:

[0054] 3-1: To ensure that the primary solution and alternative solutions are represented in the same optimization model in a closed-form expression, define... For the terminal The set of optional scheme indexes, where Indicates the main scheme, Indicates mutually exclusive alternatives; For the first The set of resource blocks occupied by each scheme. These are the filtered or recalculated scheme weights. Let these be the decision variables for choosing this option. Construct a 0-1 integer programming model: ; ; ; ;

[0055] 3-2: Introducing multipliers Relaxing capacity constraints decomposes the original problem into independent sub-problems based on the operating terminals: ; ;

[0056] When the alternative solutions are an enumerated set, the subproblem selects the feasible solution with the maximum net profit; when the alternative solutions need to be generated online, the system... China-Israel border rights Perform A* search to generate candidate paths. Update multipliers using subgradient method: ;

[0057] Where the subgradient vector The The components are: ;

[0058] The step size is: ;

[0059] in, The upper bound of the original maximization problem in the nth iteration is taken as the value of the lower Lagrange dual function in this iteration; The lower bound of the original problem in the nth iteration is taken as the objective function value corresponding to the best feasible solution that the system has searched up to this iteration, satisfying all capacity constraints and mutually exclusive constraints; the difference between the two is the current dual gap; the denominator is... Let L2 norm be the square of the subgradient vector; if If this is true, it means that the capacity constraint has not been violated and the multiplier update can be stopped. This is the step size scaling factor, determined by convergence experiments.

[0060] 3-3: If duality gap ( ), determine if the iteration has converged and output the current optimal feasible allocation scheme; if the number of iterations reaches ( (After setting the upper bound of edge node computing power delay) the dual gap is still If the real-time solution takes longer than the upper bound of the scheduling cycle delay, a rollback is triggered: based on performance density. Greedy allocation; apply neighborhood perturbation, including time offset, to unassigned tasks. Or spatial offset to adjacent grid If the new solution meets the capacity constraints and improves overall efficiency, then the replacement will be accepted, ensuring that the system outputs a feasible solution within the specified time delay.

[0061] Step 4: Resource occupancy control certificate generation:

[0062] 4-1: Extract the final set of winning bids The utilization rate is at a critical saturation state. The resource blocks constitute the core constraint set. .

[0063] 4-2: Constructing a quadratic programming model to generate control voucher vectors Among them Represents a subset of resource blocks and the set of winning bidders. Distinguishing between them. Considering the upper limit of the voucher. There may be a numerical mismatch between the threshold and the core constraint threshold. (The weight of the candidate solution ultimately selected for terminal k). To ensure that the voucher generation module can be solved under all feasible allocation results, a non-negative slack variable is introduced. And set a large penalty coefficient. The quadratic programming model is as follows: ; ; ; ;

[0064] in, Indicates terminal The index of the selected scheme in the final allocation scheme, reference value. for: ;

[0065] The protection threshold is: ;

[0066] Its physical meaning is: core constraint subset The marginal computational cost discounted at the optimal solution is used for computational cost conservation verification during terminal validation. If This indicates that the subset of core constraints satisfies strict credential coverage; if If a missing document is detected, the system records the missing document and triggers manual review or lowers the access level of the corresponding resource block in the next cycle. When the core constraint set is large, only single-element subset constraints can be considered to reduce computational complexity. The QP model is solved using the standard interior-point method. The solution obtained... The slack variable digest, after being hashed and signed with SHA-256, is issued as a resource occupancy control credential to the aircraft flight control terminal for verification of the legality of resource occupancy and verification of command tamper-proofing during the execution phase.

[0067] Step 5: Multidimensional load feedback and constraint boundary adjustment for solution space optimization:

[0068] 5-1: Calculate the resource block congestion index: ;

[0069] in, For the allocated load, For the past The conflict rate within a time slot, which is the ratio of the number of conflict alarms to the total number of allocations, is set to a value of [value missing]. This is used to eliminate the risk of numerical explosion; .

[0070] 5-2: Introducing a normalized frequency factor This means the operating terminal within the historical window. Occupying resource blocks The number of times it occurs within the resource blocks of this window The ratio of the total number of times the device is occupied, taking the value of The equivalent occupancy factor for the next pre-inspection cycle is calculated as follows: ;

[0071] in, >0 is the system's preset baseline equivalent occupancy coefficient, which represents the equivalent resource occupancy of a single operating terminal on a single resource block under nominal operating conditions without congestion or historical occupancy preferences. It is jointly determined by the grid volume of the resource block, the time slot length, and the nominal type of the aircraft. , These are the congestion amplification factor and the frequency amplification factor, respectively. , This coefficient does not relax the hard capacity constraint, but is used as a virtual occupancy penalty term for candidate solutions in the next cycle of pre-detection; a candidate solution is only allowed to enter the formal optimization model when the following equation holds: ;

[0072] Therefore, the more congested the resource block is and the higher the historical frequency of the operation terminal's occupation, the easier it is for its candidate path to be filtered in the pre-inspection stage. This helps to suppress a single entity from occupying hot resource blocks for a long time, release the feasible region to alternative paths, and ensure the stability of the constraint matrix condition number for the long-term operation of the solver.

[0073] 5-3: Unsuccessful bids enter a scrolling window. The system is based on For the cost of heuristics Then search for alternative paths again.

[0074] Step S6: A stable packet association and isolation mechanism for the solution queue:

[0075] 6-1: Synchronously generate time-series hashes when submitting data packets: The allocation results are verified after being publicized to prevent the input data from being tampered with and interfering with the solution queue.

[0076] 6-2: Constructing a packet association graph The node is the operational terminal. To avoid the denominator being zero, the intersection ratio is defined first. Historical relevance , This refers to the set of resource blocks occupied by each of the two connected terminals.

[0077] when hour, ;

[0078] when hour, ;

[0079] Historical relevance is defined as: ; in, and These are the weight fluctuation sequences of operating terminals i and j within the historical sliding window W, respectively, which are the first-order difference sequences of the weight values ​​submitted in each time slot.

[0080] Edge weight is defined as: ;

[0081] in, Typical value , ; The Pearson correlation coefficient is the historical weighted fluctuation series. Ensure edge weights are non-negative. Set edge weights to zero when two endpoints are not spatially related. Run the Louvain algorithm (community detection algorithm) to divide communities. If the average edge weight within a community... ( If a data packet within a community is identified as a high-similarity collaborative input pattern, the system automatically generates a solution queue downgrade flag, lowers the solution priority of that data packet to the lowest queue, and limits the number of edge computing threads it occupies, thereby preventing abnormal input patterns from interfering with the solver iteration process.

[0082] Specifically, the mapping between system data flow and control flow in this invention is shown in Table 1 below: Table 1. Mapping of System Data Flow and Control Flow

[0083] This invention employs a continuous unified formula through the filtering mechanism in step 2, effectively suppressing abnormal fluctuations in the objective function weights and eliminating numerical instability caused by discontinuities in the original piecewise formula. This reduces the number of multiplier update oscillations by 68% and significantly improves the convergence speed of the dual gap. Step 3 combines Lagrange decomposition and neighborhood backoff, balancing global optimal search with delay upper bound control. Under extreme conditions, the solution time remains stable at 1.76 seconds, meeting the UTM second-level response requirement. Step 5 utilizes a normalized frequency feedback mechanism to actively manage constraint boundary relaxation, preventing the continuous tightening of hard constraints on hotspot grid capacity. After 2 hours of continuous operation, the constraint degradation rate drops to 11.7%, ensuring the stability of the condition number for long-term solver operation. Step 6 includes protection against zero denominators in the association graph, linking identification with queue degradation marking to block interference from abnormal collaborative inputs on the solution iteration process. This increases the task allocation success rate to 94.2%, improving the system's computational robustness.

[0084] The performance differences between this invention and traditional technologies are shown in Table 2 below: Table 2. Comparison of performance differences between the method of the present invention and existing technologies.

[0085] Note: The simulation environment is... The mission involves 250 mixed sorties within the city's core area, with 15% anomaly weighted noise injected. Edge nodes are configured with Intel Xeon D-2146NT, and the solver is based on a secondary development of Gurobi 10.0.

[0086] In addition, such as Figure 1 As shown, the present invention also provides a dynamic scheduling system for low-altitude airspace grid resources, including:

[0087] Spatiotemporal grid constraint modeling module (corresponding to step 1): Discretizes the spatial domain and time axis, generating a set of spatiotemporal resource blocks carrying dynamic capacity boundaries and a spatiotemporal reachability graph, which serve as the constraint matrix input for the optimization solver. This module aims to address the deficiency of "distorted physical constraint modeling" in the background technology.

[0088] Flight plan data packet filtering and structuring module (corresponding to step 2): Receives combined flight plan data packets, performs graph connectivity verification and objective function weight filtering based on historical sequences, and outputs a robust objective function coefficient vector. This module aims to address the deficiency in the background technology where "input parameter noise interferes with the convergence of the objective function."

[0089] Constraint optimization decomposition solution engine (corresponding to step 3): Constructs a 0-1 integer programming model, uses Lagrange relaxation decomposition for iterative solution, and incorporates a neighborhood perturbation backoff mechanism to ensure an upper bound on computational latency. This module aims to address the shortcomings of the background technology, such as "combinatorial explosion and latency runaway caused by the single-path data packet mechanism."

[0090] Control certificate generation module (corresponding to step 4): Based on the final allocation scheme and the shadow price of the core constraint set, solve the quadratic programming model to generate resource occupancy control certificates, and send them to the aircraft flight control terminal for runtime verification.

[0091] Load feedback and constraint adjustment module (corresponding to step 5): Calculates the grid congestion index, dynamically adjusts the pre-detection threshold for the next cycle based on historical occupancy frequency, and actively relaxes hotspot constraint boundaries. This module aims to address the deficiency in the background technology of "lack of solver-guided dynamic management of constraint boundaries".

[0092] Solving queue isolation module (corresponding to step 6): Constructing a data packet association graph, identifying highly similar collaborative input patterns through community discovery, and triggering priority rearrangement of the solving input queue and isolation of edge computing threads.

[0093] Based on the technical solution of this invention, the following case scenario illustrates the implementation process of this invention in practical applications. The specific application implementation scheme is as follows:

[0094] This example uses a core area of ​​a city. Taking low-altitude airspace as an example, the complete execution flow of the method of the present invention is illustrated as follows:

[0095] 1. System initialization and parameter configuration:

[0096] Grid step size set to , , The capacity parameter is set to the space reuse factor. Meteorological attenuation factor Topographic constraint factors The safety interval is calculated as follows: ;

[0097] Take typical value , , ,have to .thus: ;

[0098] The filter parameters are set to , , Solver parameters are set to , , Feedback parameters are set to , , , Conflict rate window The queue isolation parameter is set to , , .

[0099] 2. Scheduling cycle execution process:

[0100] (1) Grid modeling: The system loads real-time meteorological data ( ) and no-fly zone polygon ( ), combined Calculate the capacity of each resource block Generate a set of spatiotemporal resource blocks A total of 12,000, constructed The graph has approximately 85,000 edges.

[0101] (2) Data packet reception and filtering: Receive flight plan data packets from terminals A, B, and C. Terminal A's original weight. Historical window average Standard deviation ,because Therefore .calculate: ;

[0102] After filtering: ;

[0103] The filtered weight vector is input into the solver, effectively suppressing the distortion of the objective function gradient caused by abnormally high weights.

[0104] (3) Lagrange iteration solution: initialization The first iteration subproblem A* searches for allocation schemes and calculates the capacity violation. The multipliers are updated using subgradients, and the dual gap decreases to [value] in the 15th iteration. Iterative convergence, outputting the winning set. In the abnormal branch, if convergence is not achieved by the 100th iteration, a rollback of step 3-3 in the preceding method steps will be triggered, according to... Greedily allocate 70% of the tasks, and execute the remaining 30%. Time-shifted neighborhood search, outputting a feasible solution within 2.1s.

[0105] (4) Voucher generation: extraction utilization rate The 12 core resource blocks constitute .by Represents a subset of resource blocks, and calculates... Threshold, use interior point method to solve QP model, generate voucher vectors After being signed with SHA-256, the data is transmitted to the flight control terminal of the winning aircraft via the 5G-A private network.

[0106] (5) Feedback adjustment: Calculate hotspot grid Congestion index. (Known) , ,but: ;

[0107] Terminal B normalized frequency Then the equivalent occupancy factor for the pre-inspection in the next cycle is: ;

[0108] When the solver constructs the constraint matrix in the next cycle, if the candidate path of terminal B involves Then according to Pre-inspection was conducted; due to The candidate path is subject to a stronger virtual occupancy penalty compared to infrequent occupants, thus prioritizing their switching to alternative paths rather than relaxing the hard capacity of hot resource blocks.

[0109] (6) Queue isolation: Constructing an association graph, it was found that the path overlap of terminals D, E, and F was 0.75 and the correlation coefficient of weight fluctuation was 0.88. .but: ;

[0110] Community average marginal weight The system generates a queue degradation flag, schedules D / E / F packets to a low-priority queue, and limits the number of threads that can use them. This ensures the availability of resources for the main queue solver.

[0111] 3. Verification of technical effectiveness:

[0112] After 10 consecutive scheduling cycles, the solver's average execution time was 1.76 seconds, the number of multiplier oscillations decreased by 68% compared to the unfiltered baseline, and the number of constraint matrix conditions remained at [value missing]. The healthy interval indicates that no solution failures have occurred due to feasible region shrinkage.

[0113] The above description is merely a detailed explanation of preferred embodiments and principles of the present invention. For those skilled in the art, there may be changes in specific implementation methods based on the ideas provided by the present invention, and these changes should also be considered within the scope of protection of the present invention.

Claims

1. A method for dynamic scheduling of low-altitude airspace grid resources, characterized in that, The steps include the following: S1 discretizes the target spatial domain into a three-dimensional mesh, discretizes the time axis into time slots, generates a set of spatiotemporal resource blocks carrying dynamic capacity constraints, and constructs a spatiotemporal reachability graph; S2, Receive flight plan data packet, the data packet contains a main scheme and at least one alternative scheme mutually exclusive with the main scheme, wherein each scheme represents a sequence of spatiotemporal resource blocks occupied by the aircraft, and each scheme is associated with an original weight value characterizing the priority of the scheme; Perform connectivity verification on each scheme based on the spatiotemporal reachability map, and perform rationality filtering on the original weight value based on the historical weight data of the operation terminal to obtain the filtered weight; S3, with the goal of maximizing the weighted total efficiency, constructs an integer programming model constrained by the dynamic capacity constraint and the scheme mutual exclusion constraint, uses Lagrange relaxation decomposition for iterative solution, and triggers a heuristic backoff mechanism when the upper bound of the delay or the threshold of the number of iterations is met, and outputs the resource allocation scheme; S4 calculates the grid congestion index and dynamically adjusts the pre-detection threshold for candidate schemes to enter the optimization model in the next scheduling cycle based on the historical frequency of resource block occupancy by the operation terminal.

2. The method for dynamic scheduling of low-altitude airspace grid resources according to claim 1, characterized in that, In step S1, the dynamic capacity constraint is determined based on the space reuse coefficient, the aircraft's nominal cruise speed, the time slot step size, the safety interval, the weather attenuation factor, and the terrain constraint factor; wherein, the safety interval is calculated based on the aircraft's relative speed, braking time, and safety margin.

3. The method for dynamic scheduling of low-altitude airspace grid resources according to claim 2, characterized in that, Step S2 also includes the following process: Before performing rationality filtering, capacity pre-check is performed on each resource block contained in the main scheme in the received flight plan data packet. If the current allocated load of any resource block reaches the preset ratio threshold of the corresponding dynamic capacity constraint, the main scheme is marked as low feasibility and a switch to the alternative scheme corresponding to the main scheme in the data packet is triggered. The rationality filtering specifically includes the following process: Maintain a sliding window of historically submitted weight values ​​for the operating terminal, calculate the mean and standard deviation of the weight values ​​within the window, and use the larger of the standard deviation and the preset minimum disturbance scale as the effective standard deviation; when the deviation between the current original weight value and the mean exceeds a preset threshold multiple multiplied by the effective standard deviation, the original weight value is decayed according to the exponential decay factor, the decay degree increases monotonically with the excess amount, and remains continuous at the boundary point.

4. The method for dynamic scheduling of low-altitude airspace grid resources according to claim 3, characterized in that, In the Lagrange relaxation decomposition process, the capacity constraint is relaxed to the objective function by introducing multipliers, and the integer programming model is decomposed into independent subproblems of each operating terminal. There are no coupling variables between the subproblems. The multipliers are updated using the subgradient method, and the update step size is determined by multiplying the ratio of the current dual gap to the square of the L2 norm of the subgradient vector by the scaling factor. The dual gap refers to the difference between the objective value of the current dual problem and the objective value of the current optimal feasible solution.

5. The method for dynamic scheduling of low-altitude airspace grid resources according to claim 4, characterized in that, In step S3, the heuristic fallback mechanism includes the following process: When the number of iterations reaches the preset maximum value, if the dual gap is still greater than the preset threshold or the real-time solution takes longer than the upper limit of the delay, a greedy allocation is performed according to the efficiency density of each scheme, and a neighborhood search is performed on the unallocated tasks by time offset or spatial offset to the adjacent grid to ensure that a feasible solution is output within the specified delay. The efficiency density is the filtered weight of each scheme divided by the number of resource blocks occupied by that scheme.

6. The method for dynamic scheduling of low-altitude airspace grid resources according to claim 5, characterized in that, In step S4, the process of dynamically adjusting the pre-detection threshold is as follows: Based on the congestion index of the resource block and the historical occupancy frequency factor of the operating terminal for the resource block, the weighted product of the congestion index and the historical occupancy frequency factor is calculated as the equivalent occupancy coefficient, wherein the weight coefficients of the congestion index and the frequency factor are preset values, and the historical occupancy frequency factor is the proportion of the terminal's historical occupancy times to the total occupancy times of the resource block. In the pre-screening phase of the next scheduling cycle, a candidate solution is allowed to enter the formal optimization model only if the current allocated load of the resource block plus the equivalent occupancy coefficient does not exceed the capacity constraint.

7. The method for dynamic scheduling of low-altitude airspace grid resources according to claim 6, characterized in that, It also includes the following processes: Based on the resource blocks in the resource allocation scheme that are in a critical saturation state, a core constraint set is formed. The corresponding Lagrange multipliers are extracted as shadow prices, a quadratic programming model is constructed to generate resource occupancy control certificates, and the certificates are sent to the aircraft flight control terminal for resource occupancy legality verification during the execution phase. The critical saturation state refers to the ratio of the currently allocated load of a resource block to the dynamic capacity constraint of the resource block being no less than a preset threshold.

8. The method for dynamic scheduling of low-altitude airspace grid resources according to claim 7, characterized in that, The quadratic programming model introduces non-negative slack variables; the objective function of the quadratic programming model includes a squared deviation term between the voucher and the reference value and a penalty term for the slack variables; for any subset of the core constraint set, the constraints ensure that the marginal cost of the subset does not exceed the sum of all vouchers associated with the resource blocks in the subset plus the slack variables. The reference value is the sum of the Lagrange multipliers corresponding to each resource block occupied by the scheme after filtering the weight of each scheme in the resource allocation scheme. If the value is less than 0, it is taken as 0; otherwise, the value itself is taken.

9. The method for dynamic scheduling of low-altitude airspace grid resources according to claim 8, characterized in that, It also includes the following processes: Construct a data packet association graph with operating terminals as nodes and the relationships between operating terminals as edges; each edge connects two operating terminal nodes, and the weight of the edge is calculated in the following way: Calculate the ratio of the intersection to the union of the resource block sets occupied by the two connected terminals; calculate the correlation coefficient between the difference sequences of the weight values ​​after filtering in adjacent scheduling cycles of the two terminals; and perform a weighted summation of the ratio of the intersection to the union and the correlation coefficient to obtain the weight of the edge. The community detection algorithm divides the community into communities. When the average edge weight in a community exceeds a preset threshold, the community is determined to be a high-similarity collaborative input pattern. A queue downgrade mark is generated for the data packets corresponding to the operating terminals in the community. These data packets are scheduled to a low-priority queue, and the number of computing threads occupied by these data packets is limited.

10. A low-altitude airspace grid resource dynamic scheduling system, used to implement the low-altitude airspace grid resource dynamic scheduling method according to any one of claims 1-9, characterized in that, The low-altitude airspace grid resource dynamic scheduling system includes: The mesh modeling module is used to discretize the target spatial domain into a three-dimensional mesh, generating a set of spatiotemporal resource blocks carrying dynamic capacity constraints; The data filtering module is used to receive flight plan data packets containing the main plan and alternative plans, and to perform rationality filtering on the weight values ​​of each plan based on the historical weight data of the operation terminal. An optimized solution engine is used to construct an integer programming model constrained by the dynamic capacity constraint and the scheme mutual exclusion constraint. Iterative solution is performed using Lagrange relaxation decomposition, and a heuristic backoff mechanism is enabled when the backoff condition is triggered to output a resource allocation scheme. The load feedback module is used to calculate the grid congestion index and dynamically adjust the pre-detection threshold for the next scheduling cycle based on the historical occupancy frequency of resource blocks by the operation terminal. The voucher generation module is used to generate resource occupancy control vouchers based on the resource allocation scheme. The queue isolation module is used to identify cooperative input patterns through packet association graphs, adjust the priority of the solver queue, and limit the number of computation threads.

Images