Unmanned aerial vehicle group electric energy guarantee resource optimal distribution method

Abstract

The invention relates to an unmanned aerial vehicle group electric energy guarantee resource optimization distribution method comprising the following steps: S1, collecting unmanned aerial vehicle states and environment data, and building an initial model; s2, using STGNN to predict future supply demands, and integrating a telecommunication traffic prediction thought to process spatial-temporal dynamics; s3, distributing an electric energy guarantee vehicle according to the alliance formation game and the innovation utility function; s4, monitoring and adjusting distribution in real time; and S5, evaluating system performance and performing iterative improvement. According to the method, a bee colony resource allocation behavior is simulated, and dynamic and predictive resource optimization is realized by using a space-time diagram neural network STGNN prediction model; and edge utility calculation is introduced, so that the stability and efficiency of distribution are ensured.

Description

Technical Field

[0001] This invention belongs to the field of unmanned aerial vehicle (UAV) technology, and in particular relates to a method for optimizing the allocation of power supply resources for UAV swarms. Background Technology

[0002] In urban disaster monitoring scenarios, drone swarms need to continuously monitor and collect data in disaster areas, but power supply is limited. Power supply vehicles, acting as mobile charging stations, need to be efficiently allocated. In existing technology, Chinese patent CN112441160B discloses a mobile base maintenance method for AGV drones with drone maintenance functions. The AGV interacts with the drones and the central control center via a wireless communication device. After receiving task instructions from the central control center, the AGV provides navigation and positioning for the drones via GPS and docks with them, allowing the drones to park on a parking platform or charge on the platform. This method facilitates the use of drones as temporary parking bases for power replenishment when their batteries are low, or for AGV recovery when the batteries are low. However, it does not provide a technical solution for situations where multiple drones need charging at a certain time, requiring coordinated management of the power supply vehicle. Summary of the Invention

[0003] This application addresses the problem of optimizing the allocation of power supply resources for unmanned aerial vehicle (UAV) swarms. It employs an optimization algorithm to rationally allocate limited power supply vehicle resources, balancing the resupply needs of multiple UAVs, improving resource utilization, and extending the overall mission duration of the UAV swarm. This application simulates swarm resource allocation behavior and uses a spatiotemporal graph neural network (STGNN) prediction model to achieve dynamic and predictive resource optimization. The method incorporates edge utility calculation to ensure the stability and efficiency of the allocation.

[0004] To achieve the above objectives, the method for optimizing the allocation of power supply resources for unmanned aerial vehicle (UAV) swarms disclosed in this application includes the following steps: S1 collects drone status and environmental data to build an initial model; S2 uses STGNN to predict future supply needs and incorporates the idea of ​​telecom traffic prediction to handle spatiotemporal dynamics. S3 allocates power to the power supply vehicle according to the utility function; S4 monitors and adjusts allocations in real time; S5 evaluates system performance and iteratively improves it.

[0005] Further, step S1 includes: The current position, remaining battery power, and mission load of each drone are obtained through sensors; a drone state vector is defined. ,in For position coordinates, Remaining battery power For task load; Record the available power and location of each support vehicle, and define resource vectors. ,in Available power. Current position; The disaster area is divided into spatiotemporal grids using a gridding method, and the initial demand matrix is ​​calculated: in For the number of drones, It is a drone With support vehicle The distance.

[0006] Further, step S2 includes: The UAV state vector and resource vector Transform into a spacetime diagram ,in The nodes include drones and support vehicles. The edge weights are derived from the distance and the battery level. Using pre-trained STGNN to extract features and predict demand growth rate: in To adjust the coefficients, a load-power ratio was added to simulate a resource hunger signal in the bee colony. Remaining battery power For task load, It is the first Forecasted demand for drones; Calculate overall forecasting demand: in The weights are derived based on task priority. It is a demand forecasting matrix; It is the task priority weight.

[0007] Further, step S3 includes: Based on the predicted demand matrix Grouped drones form an alliance The utility function is calculated as follows: in To improve allocation efficiency, Nash equilibrium is incorporated to ensure stability. Remaining battery power Available power; It is the first One alliance; It is the effect of alliances; Minimize the cost function: It is the alliance's forecast of demand, It is a resource vector. It is the total number of alliances; The cost function is solved using a genetic algorithm to generate the allocation matrix. ,in Indicates drone By support vehicle supply.

[0008] Further, step S4 includes: Collect actual power consumption and update status. ,in, To consume the difference, It is the UAV state vector; Calculate the deviation , It is a demand forecasting matrix. It is the actual demand matrix, if Rerun the optimization in step S3. For threshold; The adjustment results are sent back to step S2 to update the prediction model.

[0009] Further, step S5 includes: Calculate resource utilization rate: This is the actual amount of electricity used. Available power; Task extension time in, For the absence of an optimized baseline, That is the final amount of electricity. For task load, It extends the time; Simulate 500 rounds using the data from steps S1-S4, and calculate the average utility improvement. It is the optimized utility. It is the initial utility; if Adjust parameters Retrain STGNN.

[0010] Further, step S2 includes: Mapping the states of drones and support vehicles into a dynamic spatiotemporal graph ,in: This represents a set of nodes, including drones and support vehicles; This represents a set of edges, the existence of which is based on spatial proximity and task relevance. Represents the edge weight matrix; The edge weight is defined as follows: in: It is a node and Spatial distance between them; It is a node and The current battery level; This is the maximum battery capacity. It is the degree of task coupling, which measures the degree of node coupling. and The intensity of collaboration on tasks is calculated based on the task allocation matrix; These are weighting coefficients that adjust the importance of spatial distance, power differences, and task coupling, respectively. Calculate the projected demand growth rate: in: It is a node The projected demand growth rate for the next time point; It is a predicted value based on spatiotemporal features extracted from a spatiotemporal graph neural network; It is the load / power ratio, serving as a basic starvation signal; It is an ecological competition factor, measuring nodes Pressure in resource competition; It is the collaboration gain factor, which measures the node's performance. Gain an advantage by leveraging resources acquired through collaboration; These are adjustment coefficients that control the effects of hunger signals, competition factors, and cooperation factors, respectively. Ecological competition factors The calculation formula is: in: It is a node The set of neighboring nodes, defined according to the graph structure; It is the number of neighboring nodes; Neighboring nodes The current demand growth rate; Neighboring nodes The current battery level; It is a node and Spatial distance between them; This is a reference distance; Collaborative gain factor The calculation formula is: in: It is a node and Task coupling degree; Neighboring nodes The power / load ratio reflects the degree of resource surplus. Neighboring nodes The growth rate of demand reflects the degree of resource scarcity. It represents the highest demand growth rate among all nodes; Generate the overall forecast demand matrix: in: It is a dynamic correction factor, based on the node. The historical prediction error adjustment is calculated as follows: It is a node The real demand growth rate; It is the error adjustment factor; It is a positive number that avoids a denominator of zero.

[0011] Furthermore, step S1 also includes: Calculate the combined edge weights: in: It is a drone With support vehicle The Euclidean distance; It is a drone The remaining battery power; It is a support vehicle The remaining capacity; It is a moment pheromone value; These are weight parameters; It is to avoid positive numbers with a denominator of zero; The pheromone update rules are as follows: in: It refers to the pheromone evaporation rate, simulating the forgetting effect; It is an increase in pheromones. It is a constant. For drones The costs in the current allocation path, including distance or energy consumption, are only updated after a successful allocation.

[0012] Furthermore, step S2 also includes: Using STGNN to output drones At any moment Basic demand growth rate: Increased hunger signals due to stress: in: It is a drone The workload; It is a drone The remaining battery power; It is the Sigmoid function. Control the steepness, For threshold; It is an adjustment factor; Calculate the weighted task priority: in As task priority weight, For area or task The aggregation requirement.

[0013] Furthermore, step S3 also includes: Calculate the coalition utility function: in: It is a drone The current energy; It is a support vehicle Available supply amount; It is the allocation efficiency coefficient; It is a drone To the support vehicle The distance; It is the distance attenuation parameter; It is the diversity entropy within the alliance. For drones The proportion of task types should be considered to encourage diversity in task types within the alliance. It is a diversity weight; The objective function is: in: It is an alliance From the center to the support vehicle The distance; It is about forecasting demand; It represents the energy distribution imbalance between the alliance and the support workshop, simulating the balance pressure in the competition for ecological resources. It is an unbalanced penalty weight. Attached Figure Description

[0014] Figure 1 This is a flowchart of the method for optimizing the allocation of power supply resources for unmanned aerial vehicle (UAV) swarms provided in the embodiments of this application.

[0015] Figure 2 This is a framework diagram of the unmanned aerial vehicle (UAV) swarm power supply resource optimization and allocation system provided in the embodiments of this application. Detailed Implementation

[0016] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0017] refer to Figure 1 The overall process of this application includes the following steps: S1 collects drone status and environmental data to build an initial model; S2 uses STGNN to predict future supply needs and incorporates the idea of ​​telecom traffic prediction to handle spatiotemporal dynamics. S3 allocates power to the power supply vehicle according to the utility function; S4 monitors and adjusts allocations in real time; S5 evaluates system performance and iteratively improves it.

[0018] In one embodiment, step S1 includes: Step S11: Collect UAV status data. Obtain the current position, remaining battery power, and mission load of each UAV using sensors. Define the UAV state vector. ,in For position coordinates, Remaining electricity, in kWh. The workload is measured in task points.

[0019] Step S12: Collect power supply vehicle resource data. Record the available power and location of each power supply vehicle, and define a resource vector. ,in Available power. This is the current location.

[0020] Step S13: Construct an environmental model. Divide the disaster area into spatiotemporal grids using a gridding method and calculate the initial demand matrix. ,in For the number of drones, This is the Euclidean distance function.

[0021] This step establishes a data foundation and avoids blindly allocating resources.

[0022] In one embodiment, step S2 includes: Step S21: Input spatiotemporal data. and Transform into a spacetime diagram ,in The nodes include drones and support vehicles. The edge weights are determined based on distance and battery power.

[0023] Step S22: Extract STGNN features. Apply a pre-trained STGNN to extract features and predict the demand growth rate. in To adjust the coefficients, a load-power ratio was added to simulate a resource hunger signal in the bee colony. It is the first Forecasted demand for drones.

[0024] Step S23: Generate the forecast demand matrix. Calculate the overall forecast demand: in It is a task priority weight, determined based on task priority. It is a demand forecasting matrix.

[0025] This step anticipates changes in demand in advance, reduces the risk of sudden power shortages, and extends the time for drones to perform tasks through dynamic allocation.

[0026] In one embodiment, step S3 uses a utility function to allocate resources and balance the needs of multiple drones.

[0027] Step S31: Based on the predicted demand matrix Grouped drones form an alliance utility function in To improve allocation efficiency, Nash equilibrium is incorporated to ensure stability. Available power. This represents the remaining battery power.

[0028] Step S32: Minimize the cost function: It is the first One alliance; It is the effect of alliances; This is the alliance's forecast of demand.

[0029] The genetic algorithm is used to solve this problem, with the constraint being to ensure vehicle capacity. .

[0030] Step S33: Generate the allocation matrix ,in Indicates drone By support vehicle supply.

[0031] This step allocates resources to ensure that each drone receives a fair supply of power.

[0032] In one embodiment, step S4 includes: Step S41: Collect actual power consumption and update status. ,in This is the difference in consumption.

[0033] Step S42 Deviation Calculation and Adjustment: Calculate the deviation. ,like Rerun the optimization in step S3, threshold .

[0034] Calculate the adjustment amount and adjustment path: Step S43 Feedback Loop. The adjustment results are fed back to step S2 to update the prediction model.

[0035] It is an update status; It is a demand deviation; It is the adjustment amount; It is the conservation coefficient.

[0036] This step handles real-time changes and extends the overall mission execution time of the drone.

[0037] In one embodiment, step S5 evaluates system performance and iteratively improves it.

[0038] Step S51 Calculate resource utilization rate: Calculation task extension time in For the absence of an optimized baseline, It is utilization rate; It means extending the time.

[0039] Simulate 500 rounds using the data from steps S1-S4, and calculate the average utility improvement. .

[0040] Iterative optimization. If... Adjust parameters as follows Retrain STGNN.

[0041] In one embodiment, step S2 treats individual drones as "swarm members," whose resupply needs are constrained by the power supply vehicle, thus creating competition. Simultaneously, mission collaboration requirements drive localized cooperation among drones, influencing the distribution of resource demands. Specific steps include: Step S21 maps the states of the drone and the support vehicle to a dynamic spatiotemporal graph. ,in: This represents a set of nodes, including drones and support vehicles; This represents a set of edges, the existence of which is based on spatial proximity and task relevance. This represents the edge weight matrix, where the weights are calculated from multiple factors, including spatial distance, remaining battery power, and task coupling.

[0042] Calculate edge weights: in: It is a node and Spatial distance between them; It is a node and The current battery level; This is the maximum battery level, used for normalization. It is the degree of task coupling, which measures the degree of node coupling. and The intensity of collaboration on tasks is calculated based on the task allocation matrix and ranges from [0, 1]. These are weighting coefficients that adjust the importance of spatial distance, power differences, and task coupling, respectively.

[0043] By dynamically updating edge weights, the graph structure reflects the competition and cooperation relationships of the drone swarm in real time during mission execution, providing accurate input for subsequent demand forecasting.

[0044] Step S22 predicts the demand growth rate, using the following formula: in: It is a node The projected demand growth rate for the next time point; It is a predicted value based on spatiotemporal features extracted from a spatiotemporal graph neural network; It is the load / power ratio, serving as a basic starvation signal; It is an ecological competition factor, measuring nodes Pressure in resource competition; It is the collaboration gain factor, which measures the node's performance. Gain an advantage by leveraging resources acquired through collaboration; These are adjustment coefficients that control the effects of hunger signals, competition factors, and cooperation factors, respectively.

[0045] Ecological competition factors Quantization Node The formula for calculating the resource competition pressure faced within a localized area is: in: It is a node The set of neighboring nodes, defined according to the graph structure; It is the number of neighboring nodes; Neighboring nodes The current demand growth rate; Neighboring nodes The current battery level; It is a node and Spatial distance between them; This is a reference distance used to normalize spatial influences; it is recommended to use the average flight radius of the drone.

[0046] The above formula simulates the impact of spatial distance on competitive pressure using an exponential decay function; nodes that are closer together contribute more to the competitive factor.

[0047] Collaborative gain factor Quantization Node The potential resource support obtained from neighboring nodes through task collaboration is calculated using the following formula: in: It is a node and Task coupling degree; Neighboring nodes The power / load ratio reflects the degree of resource surplus. Neighboring nodes The normalized value of the demand growth rate reflects the degree of resource scarcity. It represents the highest demand growth rate among all nodes.

[0048] The above formula simulates the mitigating effect of collaboration on the growth rate of demand by weighting the degree of task coupling and the degree of resource surplus, thereby realizing resource sharing.

[0049] Step S23 generates a weighted demand matrix: After predicting the demand growth rate of each node, the matrix is ​​combined with task priority weights. Generate the overall forecast demand matrix: in: It is a dynamic correction factor, based on the node. The historical prediction error adjustment is calculated as follows: It is a node The real demand growth rate; It is the error adjustment factor; It should be a small positive number to avoid the denominator being zero.

[0050] The dynamic correction factor dynamically adjusts the predicted values ​​based on historical errors, thereby improving the model's adaptability.

[0051] Transforming the resupply demand prediction problem of drone swarms into a dynamic ecosystem problem significantly improves the model's ability to represent complex interactions. Ecological competition and cooperation gain factors reflect the competitive pressure and cooperative benefits among individual drones. Dynamic correction factors enhance the model's prediction accuracy.

[0052] In one embodiment, step S1 further includes: Pheromones update mechanism: each edge Not only based on spatial distance In addition to the difference in battery / capacity, it also dynamically updates the pheromone value. This indicates the "attractiveness" or "priority" of historical interactions.

[0053] The comprehensive definition of edge weights is as follows: in: It is a drone With support vehicle The Euclidean distance; It is a drone The remaining battery power; It is a support vehicle The remaining capacity; It is a moment pheromone value; These are weighting parameters, which are optimized through simulation. It should be a small positive number to avoid the denominator being zero.

[0054] The pheromone update rule in the improved ant colony algorithm is as follows: in: It refers to the pheromone evaporation rate, simulating the forgetting effect; It is an increase in pheromones. It is a constant. For drones The costs in the current allocation path, including distance or energy consumption, are only updated after a successful allocation.

[0055] In one embodiment, step S2 treats the low battery or high load state of the drone as a "stress signal" in STGNN and enhances the sensitivity to emergency needs through a nonlinear amplification function.

[0056] STGNN Output Drones At any moment Basic demand growth rate: Increased hunger signals due to stress: in: It is a drone The workload; It is a drone The remaining battery power; It is the Sigmoid function, used for non-linear amplification of "starvation signals". Control the steepness, For threshold; It is an adjustment factor.

[0057] Task priority is calculated as follows: in As task priority weight, For area or task The aggregation requirement.

[0058] In one embodiment, step S3 treats the drone and the support vehicle as "predators" and "resource providers" in the ecosystem, respectively, and introduces the "division of labor and cooperation" concept from the bee colony algorithm to dynamically adjust the alliance. The formation and protection vehicle Matching.

[0059] Coalition utility function: in: It is a drone The current energy; It is a support vehicle Available supply amount; It is the allocation efficiency coefficient; It is a drone To the support vehicle The distance; It is the distance attenuation parameter; It is the diversity entropy within the alliance. For drones The proportion of task types should be considered to encourage diversity in task types within the alliance and improve ecosystem adaptability. It is the diversity weight.

[0060] Objective function: in: It is an alliance From the center to the support vehicle The distance; It is about forecasting demand; It represents the energy distribution imbalance between the alliance and the support workshop, simulating the balance pressure in the competition for ecological resources. It is an unbalanced penalty weight.

[0061] The global supply constraint is: ; The unique allocation constraint is: each drone At most one support vehicle will be assigned. .

[0062] In one embodiment, step S4 uses a hybrid swarm intelligence optimization method, which combines the genetic algorithm with the ant colony algorithm (ACO) to simulate biological population evolution and resource exploration.

[0063] Chromosome encoding: Each chromosome represents an allocation matrix. The flat design Indicates drone Assigned to support vehicle .

[0064] Fitness function: in Penalties for violating constraints, such as exceeding supply limits or redistributing resources.

[0065] Ant colony exploration: based on pheromones Guide the allocation path search, prioritizing support vehicle-drone pairs with high historical utility; Genetic evolution: Preserving crossover and mutation operations in GA, population size Algebra Cross rate Variation rate .

[0066] By drawing on the pheromone update mechanism from ant colony foraging behavior, the dynamic interactive modeling capability of spatiotemporal graphs is enhanced, breaking through the limitations of traditional STGNN which relies solely on static features. It simulates the balance between population diversity and resource competition in an ecosystem, making the allocation scheme more robust and fair. It integrates ant colony algorithms and genetic algorithms to simulate the multi-level co-evolutionary process in nature, making it more adaptable to non-convex, high-dimensional, and dynamic environments compared to single optimization algorithms.

[0067] refer to Figure 2 The unmanned aerial vehicle (UAV) swarm power supply resource optimization and allocation system provided in this application includes the following modules: Initial Model Building Module: Collect UAV status and environmental data to build an initial model; Future supply demand forecasting module: Uses STGNN to forecast future supply demand, and incorporates the idea of ​​telecom traffic forecasting to handle spatiotemporal dynamics; Power Supply Vehicle Allocation Module: Allocates power supply vehicles based on alliance formation game and utility function; Real-time monitoring module: Monitors and adjusts allocation in real time; System performance evaluation module: Evaluate system performance and iteratively improve it.

[0068] To verify the effectiveness of the proposed method for optimizing the allocation of power resources for unmanned aerial vehicle (UAV) swarms, we will evaluate its performance in terms of resource utilization, mission extension time, and system stability.

[0069] The experimental data comes from the following two sources to ensure its authenticity and representativeness: The publicly available urban disaster monitoring dataset "UAV-Disaster2020" was used. This dataset contains data on the location, power consumption, and mission load of drones performing missions in urban disaster areas. It covers the monitoring records of 100 drones over 72 hours, with the mission area being a 10 km × 10 km urban grid.

[0070] To simulate larger-scale and more diverse scenarios, synthetic data was generated based on the parameter distribution of real-world datasets, including 500 drones and 50 power supply vehicles. The mission area was expanded to 20 km × 20 km, spanning 168 hours. The synthetic data was generated using the Monte Carlo method, and the power consumption followed a normal distribution. kWh / h, the task load follows a Poisson distribution.

[0071] The disaster area was divided into a 100×100 spatiotemporal grid. The initial battery capacity of the support vehicle was 50 kWh, with a maximum single resupply capacity of 10 kWh. The initial battery capacity of the drones was 2 kWh. Task priority weights were determined. Randomly assigned within the range of [0.5, 1.5].

[0072] Data preprocessing includes removing outliers, such as records with negative battery levels, standardizing location coordinates, and ensuring the consistency of input data.

[0073] The experimental hardware consisted of a server equipped with an NVIDIA RTX 3090 GPU (24GB VRAM), an Intel Xeon Silver 4210 CPU, and 128GB of RAM. Python 3.8 was used as the development language. The STGNN model was implemented using PyTorch 1.9.0, and the genetic algorithm and the hybrid ecological intelligent optimization algorithm HESO were implemented using NumPy and SciPy libraries. The simulation environment was MATLAB R2021a.

[0074] To comprehensively evaluate the performance of the proposed method, it is compared with the following three benchmark methods: Greedy-NN nearest neighbor greedy algorithm: Drones are preferentially assigned to the nearest support vehicle that has remaining power.

[0075] GA Genetic Algorithm: Optimizes resource allocation based on traditional genetic algorithms, considering only distance and power constraints.

[0076] ACO ant colony optimization algorithm: optimizes the allocation path based on the pheromone mechanism, but does not include ecological fitness and cooperation factors.

[0077] The experiment uses the following key indicators to quantify system performance: Resource utilization rate reflects the efficiency of electricity usage by the power supply vehicle; The task extension time, compared to the task time of the baseline method; Fairness of allocation is measured by the standard deviation and mean of the final electricity distribution to determine the balance of the allocation. The calculation time, in seconds, is used to evaluate the real-time performance of the algorithm.

[0078] The experiments were run for 500 rounds on both real and synthetic datasets, and the average performance metrics were recorded. The results show that our application significantly outperforms the benchmark method in terms of resource utilization, task extension time, and allocation fairness. In particular, on the real dataset, resource utilization is improved by about 20%, and task time is extended by about 15%. Although the computation time is slightly higher than other methods, it is still within an acceptable range and suitable for periodic optimization scenarios.

[0079] To analyze the contribution of each innovative module to performance, an ablation experiment was designed, gradually removing key components. The results showed that ecological competition and cooperation factors contributed the most to resource utilization and task time. After removal, The performance decreased by 6 percentage points; the pheromone feedback mechanism and the diversity entropy / imbalance penalty also significantly affected the performance, verifying the necessity of each innovative module.

[0080] This application introduces ecological competition and cooperation factors, pheromone feedback, and diversity entropy, which significantly improves prediction accuracy and allocation efficiency, especially in resource-scarce scenarios.

[0081] Experimental results show that the method proposed in this application has achieved significant results in the optimal allocation of power supply resources for UAV swarms, increasing resource utilization to 0.89, extending mission duration by 9.3 hours, and improving allocation fairness to 0.88, which is superior to benchmark methods such as Greedy-NN, GA, and ACO.

[0082] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for optimizing the allocation of power supply resources for unmanned aerial vehicle (UAV) swarms, characterized in that, Includes the following steps: S1 collects drone status and environmental data to build an initial model; S2 uses STGNN to predict future supply needs and incorporates the idea of ​​telecom traffic prediction to handle spatiotemporal dynamics. S3 allocates power to the power supply vehicle according to the utility function; S4 monitors and adjusts allocations in real time; S5 evaluates system performance and iteratively improves it.

2. The method for optimizing the allocation of power supply resources for unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, Step S1 includes: The current location, remaining battery power, and mission load of each drone are obtained through sensors; a drone state vector is defined. Record the available battery power and location of each support vehicle, and define a resource vector; The disaster area was divided into spatiotemporal grids using a gridding method, and the initial demand matrix was calculated.

3. The method for optimizing the allocation of power supply resources for unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, Step S2 includes: The state vector and resource vector of the UAV are transformed into a spatiotemporal graph, which includes nodes and edges. The nodes include the UAV and the support vehicle, and the edge weights are obtained based on the distance and battery power. Pre-trained STGNN is used to extract features and predict demand growth rate. Calculate the overall forecast demand.

4. The method for optimizing the allocation of power supply resources for unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, Step S3 includes: Based on the predicted demand matrix, group the drones into alliances and calculate the utility function; The goal is to minimize the cost function; A genetic algorithm is used to solve for the cost function and generate the allocation matrix.

5. The method for optimizing the allocation of power supply resources for unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, Step S4 includes: Collect actual power consumption and update status; Calculate the deviation; if the deviation is greater than the preset threshold, rerun the optimization step S3. The adjustment results are sent back to step S2 to update the prediction model.

6. The method for optimizing the allocation of power supply resources for unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, Step S2 includes: Map the states of the drone and the support vehicle into a dynamic spatiotemporal diagram; Calculate the projected demand growth rate: in: It is a node The projected demand growth rate for the next time point; It is a predicted value based on spatiotemporal features extracted from a spatiotemporal graph neural network; It is the load / power ratio, serving as a basic starvation signal; It is an ecological competition factor, measuring nodes Pressure in resource competition; It is the collaboration gain factor, which measures the node's performance. Gain an advantage by leveraging resources acquired through collaboration; These are adjustment coefficients that control the effects of hunger signals, competition factors, and cooperation factors, respectively. Ecological competition factors The calculation formula is: in: It is a node The set of neighboring nodes, defined according to the graph structure; It is the number of neighboring nodes; Neighboring nodes The current demand growth rate; Neighboring nodes The current battery level; It is a node and Spatial distance between them; This is a reference distance; Collaborative gain factor The calculation formula is: in: It is a node and Task coupling degree; Neighboring nodes The power / load ratio reflects the degree of resource surplus. Neighboring nodes The growth rate of demand reflects the degree of resource scarcity. It represents the highest demand growth rate among all nodes; Generate the overall forecast demand matrix: in: It is a dynamic correction factor, based on the node. Historical forecast error adjustment.

7. The method for optimizing the allocation of power supply resources for unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, Step S1 also includes: Calculate the combined edge weights: in: It is a drone With support vehicle The Euclidean distance; It is a drone The remaining battery power; It is a support vehicle The remaining capacity; It is a moment pheromone value; These are weight parameters; It is to avoid positive numbers with a denominator of zero; The pheromone update rules are as follows: in: It refers to the pheromone evaporation rate, simulating the forgetting effect; It is an increase in pheromones. It is a constant. For drones The costs in the current allocation path, including distance or energy consumption, are only updated after a successful allocation.

8. The method for optimizing the allocation of power supply resources for unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, Step S2 also includes: Using STGNN to output drones At any moment Basic demand growth rate: Increased hunger signals due to stress: in: It is a drone The workload; It is a drone The remaining battery power; It is the Sigmoid function. Control the steepness, For the threshold; It is an adjustment factor; Calculate the weighted task priority.

9. The method for optimizing the allocation of power supply resources for unmanned aerial vehicle (UAV) swarms according to claim 1, characterized in that, Step S3 also includes: Calculate the coalition utility function: in: It is a drone The current energy; It is a support vehicle Available supply amount; It is the allocation efficiency coefficient; It is a drone To the support vehicle The distance; It is the distance attenuation parameter; It is the diversity entropy within the alliance. For drones The proportion of task types should be considered to encourage diversity in task types within the alliance. It is a diversity weight; The objective function is: in: It is an alliance From the center to the support vehicle The distance; It is about forecasting demand; It represents the energy distribution imbalance between the alliance and the support workshop, simulating the balance pressure in the competition for ecological resources. It is an unbalanced penalty weight.

Images