A dynamic task priority-based unmanned aerial vehicle-unmanned vehicle closed-loop collaborative scheduling method

Abstract

The application discloses a kind of unmanned aerial vehicle-unmanned vehicle closed-loop collaborative scheduling methods based on dynamic task priority, the method first obtains task point and the resource state information of unmanned aerial vehicle, unmanned vehicle, then constructs task and resource double dynamic priority model, and the optimal matching of task and unmanned aerial vehicle is realized by hungarian algorithm;Task point space is clustered using weighted K-means clustering, and the flight path in unmanned aerial vehicle cluster and the driving path between unmanned vehicle clusters are planned by combining ant colony algorithm;A real-time state monitoring mechanism for unmanned aerial vehicle is established, a multi-level supply trigger threshold and a dynamic return decision strategy are designed, and a closed-loop scheduling process of "task execution-state monitoring-supply decision-task redistribution" is formed;Finally, the scheduling strategy is continuously optimized based on MADDPG multi-agent deep reinforcement learning.

Description

Technical Field

[0001] This invention relates to a closed-loop collaborative scheduling method for unmanned aerial vehicles (UAVs) and unmanned vehicles based on dynamic task priority. Background Technology

[0002] With the rapid development of unmanned systems technology, collaborative operations between drones and unmanned vehicles (UAVs) have become an important technical means in fields such as emergency rescue, regional reconnaissance, and target search. Drones have the advantages of wide field of view and high maneuverability, but their endurance is generally short due to limitations in onboard energy capacity. While UAVs have long endurance, they are limited by restricted reconnaissance perspectives and insufficient mobility in complex terrain. By constructing a heterogeneous "drone-unmanned vehicle" system through air-ground collaboration, complementary capabilities can be achieved, improving overall operational efficiency.

[0003] In existing technologies, various solutions have been proposed for vehicle-machine collaborative scheduling. For example, methods include introducing vehicles as mobile platforms for drones to achieve multi-area reconnaissance, determining the combination of drones and unmanned vehicles participating in the task through collaborative optimization models, and multi-drone and multi-vehicle collaborative scheduling methods for inspection scenarios. These solutions, to some extent, address the limitations of single-platform capabilities and achieve basic task allocation and path planning. However, existing technologies still have the following shortcomings:

[0004] First, the task priority handling mechanism is imperfect. Existing methods mostly take the shortest path or the fastest time as the optimization goal and adopt a static allocation strategy. They fail to dynamically adjust the priority according to the urgency of the task, environmental changes and the passage of time, which may cause important tasks to be delayed.

[0005] Secondly, the closed-loop management of endurance and resupply is lacking. Existing solutions mostly use fixed thresholds to trigger resupply (such as returning to base when the battery level is below 20%), which lacks a decision-making mechanism that intelligently weighs the mission value and remaining capacity, resulting in low resource utilization or mission interruption risks.

[0006] Secondly, the scheduling architecture lacks openness. Most existing systems are open-loop or semi-open-loop structures, with task allocation, path planning, and status monitoring being isolated from each other, making it difficult to dynamically reconfigure based on real-time execution results and resource status.

[0007] Finally, the multi-objective collaborative optimization capability is weak. Existing algorithms are unable to simultaneously take into account multiple objectives such as task completion rate, priority satisfaction and system energy consumption, and lack continuous self-learning capability.

[0008] Therefore, how to achieve efficient collaborative scheduling of drones and unmanned vehicles in dynamic task environments, and how to build a closed-loop scheduling architecture that takes into account dynamic changes in task priorities, drone endurance constraints, and overall system performance, has become a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0009] This invention provides a closed-loop collaborative scheduling method for UAVs and unmanned vehicles based on dynamic task priority to solve the problems existing in the prior art.

[0010] The technical solutions adopted in this invention are as follows:

[0011] A closed-loop cooperative scheduling method for UAVs and unmanned vehicles based on dynamic task priority includes the following steps:

[0012] S1: Obtain a set of task points to be processed. Each task point includes spatial coordinates, initial priority attribute, generation time, maximum allowable waiting time, and real-time environmental information. Obtain resource status information of UAVs and unmanned vehicles. The resource status information of UAVs includes real-time location, current battery level, and remaining task payload. The resource status information of unmanned vehicles includes real-time location, current battery level, and remaining payload.

[0013] S2: Construct a task dynamic priority model based on initial priority attributes, generation time, maximum allowable waiting time, and real-time environmental information; construct a resource dynamic priority model based on the UAV's real-time location, current battery level, and task payload margin, wherein the resource priority determined by the resource dynamic priority model is positively correlated with the UAV's current battery level; and assign tasks to corresponding UAVs according to the matching relationship between the task dynamic priority model and the resource dynamic priority model.

[0014] S3: Based on the initial priority attribute, perform spatial clustering on each task point, and determine the cluster center as the take-off and landing point of the UAV and the parking point of the unmanned vehicle; with the optimization objective of minimizing the correlation between the initial priority attribute and the execution order, plan the flight path of the UAV in each cluster and the driving path of the unmanned vehicle between each cluster.

[0015] S4: Establish a real-time monitoring mechanism for UAV status and set multi-level resupply trigger thresholds. The monitoring mechanism includes monitoring the current battery level and remaining mission payload of the UAV, and determining the remaining executable capacity based on the current battery level; determining the return decision based on the comparison result between the mission priority output by the mission dynamic priority model and the remaining executable capacity; adjusting the UAV to the pick-up position based on the real-time position of the UAV; forming a closed-loop scheduling process of mission execution, status monitoring, resupply decision and mission reallocation.

[0016] S5: Based on a multi-agent cooperative optimization algorithm, the scheduling strategy is continuously optimized with task performance and resource consumption as optimization objectives.

[0017] Furthermore, the task dynamic priority model calculates the dynamic priority of task points according to the following formula:

[0018] ,

[0019] in, For the initial priority attribute, For the current moment, For the generation time, The maximum allowed waiting time, For time weighting coefficients, Environmental factors;

[0020] Environmental factors are determined according to the following formula:

[0021] ,

[0022] in, For real-time ambient temperature, Real-time concentration of harmful gases This is a temperature reference threshold. This is the concentration reference threshold. , These are the weighting coefficients.

[0023] Furthermore, the resource dynamic priority model calculates the resource priority of the UAV according to the following formula:

[0024] ,

[0025] in, For resource priority, This is the current battery level. Full charge This represents the mission payload margin. For maximum load, Distance to the nearest mission point For the maximum acceptable distance, , , These are the weighting coefficients. A positive coefficient ensures that resource priority is positively correlated with current power consumption.

[0026] Furthermore, spatial clustering is performed on each task point based on the initial priority attribute, and the cluster centers are updated according to the following formula:

[0027] ,

[0028] in, For the first Cluster centers, For the first One cluster, This is the initial priority attribute. The coordinates of the task point;

[0029] Calculate the spatial radius of each cluster ,when When the drone's range exceeds half of the stated range, increase the number of clusters and re-cluster the data.

[0030] Furthermore, with the optimization objective of minimizing the correlation between the initial priority attribute and the execution order, it is determined according to the following formula:

[0031] ,

[0032] in, To optimize the target value, For the number of clusters, This is the initial priority attribute. For the task point In its respective number The order of execution within each cluster.

[0033] Furthermore, the dynamic return-to-home decision only takes effect when the drone's battery level is between 40% and 25% of its full charge, or when the mission payload margin is less than 2 units. The return-to-home decision is determined according to the following formula:

[0034] ,

[0035] in, Priority of the current task point Remaining battery power Energy consumption per unit time A set of tasks to be executed. This refers to the flight time required for the drone to return from its current location to the nearest cluster center. This is an adjustment factor that triggers a return trip when the above conditions are met.

[0036] Multi-level supply trigger thresholds are executed in descending order of priority, including:

[0037] The second battery threshold is an insurmountable safety baseline. When the drone's current battery level is less than 25% of its full charge, the drone will immediately abort its mission and return to base regardless of the dynamic return-to-home decision.

[0038] The first battery threshold is the trigger condition for dynamic return-to-home decision-making. When the current battery level of the drone is lower than 40% of the full battery level but not lower than 25%, the trade-off calculation for dynamic return-to-home decision-making is initiated.

[0039] The load threshold is the trigger condition for dynamic return-to-home decision-making. When the UAV's mission load margin is less than 2 units, the trade-off calculation for dynamic return-to-home decision-making is initiated.

[0040] Furthermore, adjusting the autonomous vehicle to the pick-up location based on its real-time location includes:

[0041] The drone calculates the estimated arrival time based on the return-to-home decision and sends the information to the unmanned vehicle.

[0042] The driverless vehicle moves to the pick-up location based on the estimated arrival time and the driverless vehicle's real-time location;

[0043] After landing, the drone performs battery replacement or payload loading.

[0044] Furthermore, the multi-agent collaborative optimization algorithm adopts the MADDPG algorithm framework, and the state space includes the remaining time window of each task point, the current battery level of the UAV, and the remaining amount of the unmanned vehicle's onboard supplies.

[0045] The action space includes task allocation, path fine-tuning, and return-to-base decisions;

[0046] The reward function is determined based on the weighted difference between task completion rate, priority satisfaction, and system energy consumption.

[0047] Furthermore, the set of tasks to be processed is located in emergency rescue scenarios, regional reconnaissance scenarios, or target search and strike scenarios;

[0048] The mission payload margin includes the amount of fire extinguishing munitions, the endurance of reconnaissance equipment, or the weapon payload.

[0049] The present invention has the following beneficial effects:

[0050] (1) By constructing a dynamic task priority model, the initial priority is dynamically adjusted based on the task generation time, the maximum allowable waiting time, and real-time environmental information (temperature, concentration of harmful gases, etc.), which solves the problem of delayed emergency task processing caused by static priority allocation in the existing technology. Combined with the weighted K-means clustering algorithm, high-priority task points are spatially closer to the cluster center, shortening the response time of the UAV and facilitating the priority handling of high-risk areas in emergency rescue and other scenarios.

[0051] (2) By constructing a dynamic resource priority model, the dynamic matching of tasks and execution units is achieved by comprehensively considering the current battery power of the UAV, the remaining mission payload, and the distance to the mission point. In particular, the return decision is determined based on the comparison between "current mission value and remaining executable time" and "the value of the pending mission and return time", which replaces the traditional fixed threshold return strategy. This can reduce unnecessary mission interruptions and improve the continuous execution capability of high-priority tasks while ensuring the safe return of the UAV.

[0052] (3) By establishing a closed-loop scheduling process of "task execution - status monitoring - resupply decision - task reallocation", and using unmanned vehicles as mobile resupply platforms, the unmanned vehicle pick-up position is dynamically adjusted based on the expected arrival time of the UAV, forming a closed-loop management of the UAV's endurance and task execution requirements. This architecture effectively alleviates the constraints on the operating range caused by the limited endurance of UAVs (usually 20-40 minutes), extends the effective operating time of UAVs, and reduces resupply waiting time.

[0053] (4) By minimizing the correlation between task priority and execution order as the optimization objective, and using a multi-agent collaborative optimization algorithm to continuously optimize the scheduling strategy, a trade-off is achieved between task completion rate, priority satisfaction, and system energy consumption. Experimental results show that compared with the fixed rule scheduling method, this scheme improves the completion rate of high-priority tasks, average response time, and system energy consumption, thereby enhancing the overall scheduling efficiency.

[0054] (5) By adjusting task priorities through real-time environmental information (temperature, concentration of harmful gases), combined with multi-level supply trigger thresholds and dynamic rescheduling mechanisms, the system can adapt to changes in the task environment (such as fire spread and dangerous area diffusion), thereby improving its robustness and autonomous decision-making ability in complex dynamic environments. Attached Figure Description

[0055] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0056] The invention will now be further described with reference to the accompanying drawings.

[0057] The UAV-Vehicle Closed-Loop Cooperative Scheduling Method based on Dynamic Task Priority of this invention can be widely applied to dynamic task scenarios requiring air-ground collaborative operations, such as emergency rescue, regional reconnaissance, and target search and strike. In actual implementation, the method achieves efficient cooperative scheduling of UAVs and Vehicles by sequentially executing the following steps: task information and resource status acquisition, dual dynamic priority model construction and task allocation, task point spatial clustering and path planning, UAV status monitoring and closed-loop scheduling, and online optimization of scheduling strategy. The specific implementation process of this method is described in detail below.

[0058] Combination Figure 1First, the system acquires the set of tasks to be processed and the resource status information of drones and unmanned vehicles. The basic information of the tasks to be processed includes spatial coordinates, initial priority attribute, generation time, maximum allowable waiting time, and real-time environmental information. The spatial coordinates of the task points are represented by three-dimensional coordinates (x, y, z) to indicate their specific location in the work area. The initial priority attribute is set to a value from 1 to 10 according to the importance and urgency of the task. The generation time is the timestamp when the task is detected by the scheduling system and included in the task set. The maximum allowable waiting time is preset according to the task type. The real-time environmental information includes environmental sensing data such as real-time ambient temperature and harmful gas concentration around the task points.

[0059] Simultaneously acquired drone resource status information includes the drone's real-time location, current battery level, and remaining mission payload. Current battery level, expressed in watt-hours, represents the drone's remaining energy. Remaining mission payload, depending on the actual mission type, can refer to the amount of fire extinguishing ammunition, the remaining range of reconnaissance equipment, or the amount of weapon payload. Unmanned vehicle (UAV) resource status information includes the UAV's real-time location, current battery level, and remaining payload. Remaining payload includes the number of spare batteries available for drone resupply and the quantity of various mission payload consumables. All mission point information and resource status information are synchronized to the scheduling system in real time, providing a data foundation for subsequent scheduling decisions.

[0060] A dynamic task priority model is constructed based on the acquired task point information, and a dynamic resource priority model is constructed based on the resource status information of the UAV. The dynamic allocation of tasks and UAVs is completed according to the matching relationship between the two models.

[0061] In the task dynamic priority model, task points At any moment The dynamic priority is calculated according to the following formula:

[0062] ,

[0063] in, For the task point The initial priority attribute, For the current moment, For the task point The generation time, For the task point Maximum allowed waiting time; This is the time weighting coefficient, with a value ranging from 0.1 to 0.3; It is an environmental factor that changes dynamically with the real-time environmental information around the task location.

[0064] Environmental factors are determined according to the following formula:

[0065] ,

[0066] in, For the task point At any moment Real-time ambient temperature, For the task point At any moment Real-time concentration of harmful gases This is a temperature reference threshold. This is the concentration reference threshold. , These are the weighting coefficients for environmental factors, with values ​​ranging from 0.1 to 0.3. They can be adjusted according to the environmental monitoring priorities of the actual task scenario.

[0067] In the resource dynamic priority model, drones At any moment Resource priority is calculated according to the following formula:

[0068] ,

[0069] in, For drones At any moment Current battery level, For drones Full charge capacity, For drones At any moment Task load margin. For drones Maximum load capacity, For drones At any moment Euclidean distance to the nearest mission point This represents the maximum acceptable distance between the drone and the mission point. , , The weighting coefficients are for power consumption, mission load, and distance, respectively, and satisfy the following conditions: , The coefficients are positive, making resource priority positively correlated with the drone's current battery level. The specific values ​​of each weight coefficient can be adjusted according to actual mission requirements.

[0070] After calculating the dynamic priorities of the tasks and resources, a matching degree matrix between the tasks and the UAVs is constructed, and the elements in the matrix are... Representation task With drones The matching degree is used to formalize the task allocation problem into a 0-1 integer linear programming problem, with the optimization objective being to maximize the total matching degree, and the constraints being:

[0071] , , ,

[0072] The Hungarian algorithm is used to solve this integer linear programming problem to obtain the optimal allocation scheme between tasks and drones. When a new task is added to the set to be processed, or when the resource status of the drone, such as battery power or mission payload margin, changes significantly, the dynamic priority of tasks and resources is recalculated, and the matching algorithm is executed again to achieve dynamic reallocation of tasks and drones.

[0073] After the initial allocation of tasks and drones is completed, spatial clustering is performed on all task points to be processed based on the initial priority attributes of the task points, and the flight paths of drones in each cluster and the driving paths of unmanned vehicles between each cluster are planned respectively.

[0074] Spatial clustering uses the weighted K-means clustering algorithm, and the cluster centers are updated according to the following formula:

[0075] ,

[0076] in For the first Cluster centers of each cluster. For the first The set of task points corresponding to each cluster. For the task point The initial priority attribute, For the task point The coordinate vectors are used to determine the initial priority of task points during the clustering process, which is then used as the cluster weights to make high-priority task points closer to the cluster center in terms of spatial distribution.

[0077] After updating the cluster centers, calculate the spatial radius of each cluster. , For the task point The Euclidean distance to the cluster center.

[0078] If the spatial radius of a certain cluster If the range of the drone exceeds half of its flight range, the number of clusters is increased and the weighted K-means clustering operation is re-executed. The steps of updating the cluster centers and verifying the cluster radii are repeated until the spatial radii of all clusters meet the flight range constraint. Finally, the cluster centers obtained will be determined as the take-off and landing points of the drones and the docking points of the unmanned vehicles.

[0079] Path planning aims to minimize the correlation between the initial priority attribute of task points and their execution order. This optimization objective is determined by the following formula:

[0080] ,

[0081] in, To optimize the target value, For the number of clusters, For the task point In its respective number The order of execution within each cluster.

[0082] For the flight path planning of UAVs within each cluster, the ant colony algorithm is used to solve the problem. First, the pheromone matrix is ​​initialized. With heuristic information , For the task point With mission points The Euclidean distance between them is used to determine the start and end points of the UAV's flight path, allowing the UAV to traverse all task points within its cluster.

[0083] In the ant colony algorithm, each ant selects the next task point to be visited based on the state transition probability. After completing one path traversal, the pheromone matrix is ​​iteratively updated according to the pheromone update rule. During the pheromone update process, a volatile factor is introduced to avoid excessive accumulation of pheromones. Finally, the optimization target value F is used as the path evaluation index, and the optimal flight path of the UAV in each cluster and the corresponding task point execution order are output.

[0084] For the planning of the driving path of the unmanned vehicle between the clusters, the ant colony algorithm is also adopted. With the goal of minimizing the total driving distance of the unmanned vehicle, the optimal driving path for the unmanned vehicle to visit all cluster centers in sequence is planned, so that the unmanned vehicle can efficiently provide supply support to the drones in each cluster area.

[0085] A real-time monitoring mechanism for UAV status is established, with multi-level resupply trigger thresholds set. Return-to-home decisions are determined based on dynamic task priorities and the UAV's remaining executable capacity, forming a closed-loop scheduling process in conjunction with the dynamic response of unmanned vehicles (UAVs). The scheduling system receives status data reported by UAVs at a frequency of 1Hz, enabling real-time monitoring of UAV status. Monitoring includes the UAV's current battery level and remaining payload. The remaining executable capacity is calculated based on the current battery level and the UAV's energy consumption per unit time, representing the time the UAV can continue performing tasks with the current battery level. Simultaneously, the scheduling system monitors the real-time location, current battery level, and remaining payload of the UAVs to ensure that the UAVs' resupply capacity matches the scheduling requirements.

[0086] The multi-level replenishment trigger thresholds include a power threshold and a payload threshold. These thresholds are executed in descending order of priority. The second power threshold is an insurmountable safety baseline: when the drone's current power level is below 25% of its full charge, the drone immediately aborts its current task and performs an emergency return-to-home operation, regardless of the dynamic return-to-home decision. The first power threshold is when the drone's current power level is below 40% but not below 25% of its full charge. This threshold is the trigger condition for the dynamic return-to-home decision; the dynamic return-to-home decision-making process is initiated only when the power level falls within this range. The payload threshold is when the drone's mission payload margin is below 2 units. This threshold also serves as a trigger condition for the dynamic return-to-home decision; the dynamic return-to-home decision-making process is initiated only when the payload meets this condition. The drone's return-to-home decision is made according to the formula...

[0087] ,in, The dynamic priority of the drone's current task point. This refers to the remaining battery power of the drone. This refers to the energy consumption of the drone per unit time. A set of tasks to be executed. The highest dynamic priority in the set of tasks to be executed. This refers to the flight time required for the drone to return from its current location to the nearest cluster center. This is an adjustment coefficient, with a value range of 0.8 to 1.2.

[0088] This dynamic decision-making only takes effect when the battery level is between 40% and 25% or the payload is less than 2 units. When the formula is met, a return-to-home operation is triggered. If the formula is not met, the drone continues to perform the current task until the battery level drops below the 25% hard threshold.

[0089] Once the drone triggers the return-to-home decision, it immediately calculates the estimated arrival time from its current location to the nearest cluster center and sends this estimated arrival time along with its own location information to the unmanned vehicle. After receiving the information, the unmanned vehicle plans the optimal driving path based on its own real-time location and the drone's estimated arrival time and moves to the corresponding rendezvous point. This rendezvous point is a cluster center that meets the terrain access requirements of the unmanned vehicle. After the drone arrives at the rendezvous point, it makes a precise landing. Subsequently, the unmanned vehicle performs battery replacement and payload loading operations for the drone, completing the energy and payload replenishment for the drone.

[0090] Throughout the mission execution process, the scheduling system continuously monitors the status information of UAVs and unmanned vehicles. When a UAV triggers a resupply threshold or a return-to-home decision, the scheduling system immediately dynamically reallocates the unfinished task to other available UAVs within the work area, forming a closed-loop scheduling process of "task execution - status monitoring - resupply decision - task reallocation" to ensure the continuity and efficiency of the overall mission execution.

[0091] The scheduling strategy is continuously optimized based on a multi-agent cooperative optimization algorithm, enabling the scheduling system to adapt to dynamically changing task environments and achieve an optimal trade-off between task efficiency and resource consumption. The multi-agent cooperative optimization algorithm adopts the MADDPG algorithm framework. The constructed state space includes the remaining time window for each task point, the current battery level of the UAV, and the remaining payload of the unmanned vehicle. It also incorporates the spatial coordinates and dynamic priorities of each task point, the real-time location, payload capacity, and current task execution status of each UAV, the real-time location and current battery level of each unmanned vehicle, as well as environmental information such as the current time and real-time weather, comprehensively representing the task and resource status within the operational area.

[0092] The constructed action space includes the executable actions of UAV agents and unmanned vehicle agents. The actions of UAV agents include task selection, path fine-tuning, and return-to-home decision. Task selection involves selecting the index of the next task point to be executed, path fine-tuning involves fine-tuning the current flight direction by angle Δθ, and return-to-home decision is represented by a binary variable to indicate whether to trigger return-to-home resupply. The actions of unmanned vehicle agents include movement direction selection and docking decision. Movement direction involves selecting the index of the next cluster center to be docked, and docking decision is represented by a binary variable to indicate whether to stay at the current position.

[0093] The algorithm's reward function is determined based on the weighted difference between task completion rate, priority satisfaction, and system energy consumption. The core is the calculation of the comprehensive reward, which is:

[0094] ,

[0095] in, , , , , .

[0096] The system uses the weighted reward for task completion as a positive incentive and the energy consumption of the drone and the total execution time of the task as negative penalties to achieve collaborative optimization of multiple objectives.

[0097] During algorithm training, each drone and each unmanned vehicle is treated as an independent intelligent agent, and a multi-agent collaborative training environment is constructed. Training data such as the state, actions, and rewards of each agent are stored in an experience replay pool. The decision-making model of the agent is trained in batch processing, and the decision-making strategy of the agent is iteratively updated. Throughout the training process, task efficiency and resource consumption are always taken as the comprehensive optimization goal. The scheduling strategies such as task allocation, path planning, and replenishment decisions are continuously optimized, so that the scheduling system can achieve a dynamic trade-off between task completion rate, high priority task satisfaction, and system energy consumption, and continuously improve the robustness and autonomous decision-making ability of the scheduling system in dynamic task environments.

[0098] In practical applications, the parameter values ​​in the method of this invention can be adaptively adjusted according to the specific task scenario, such as the time weight coefficient α and the environmental factor weight coefficient. and Resource priority weight coefficient , , The parameters can be set to specific values ​​according to the needs of different mission scenarios such as emergency rescue, regional reconnaissance, target search and strike. At the same time, the relevant training parameters of the ant colony algorithm and the MADDPG algorithm can also be adjusted according to the size of the operation area and the number of task points. Such parameter adjustments and detailed optimizations that do not depart from the basic principles of this invention are all within the protection scope of this invention.

[0099] To further illustrate the practical application effect of the method of the present invention, the implementation process of the present invention will be described in detail below in conjunction with the specific task scenario of forest fire emergency rescue.

[0100] This embodiment selects a 10km × 10km forest fire operation area, setting up 50 task points to be handled, of which 10 task points are areas where personnel are trapped, with an initial priority. Set at level 9-10, with 40 task points designated for fire monitoring and fire detection, initial priority... Set to levels 1-8; maximum allowed waiting time for mission points. Based on the task type, the time limit is 60 minutes for personnel search and rescue and 120 minutes for fire monitoring.

[0101] Environmental factors The temperature reference threshold is calculated based on the real-time temperature and CO concentration at the task location. The reference threshold for CO concentration is 40℃. It is 50 ppm. =0.2、 =0.2; The time weight coefficient α is set to 0.2 so that the task priority increases linearly with the waiting time.

[0102] This embodiment is configured with 4 multi-rotor drones, each with a full charge. The maximum workload is 100%, with energy consumption of 2% / min per unit time. It carries 8 fire extinguishing bombs and has a maximum flight range of 30 minutes; it is equipped with one ground unmanned vehicle with a maximum speed of 60 km / h, carrying 20 spare batteries and 40 fire extinguishing bombs, enabling automatic battery swapping and payload resupply for the drone.

[0103] First, the initial collection of task information and resource status is performed. The scheduling system obtains the three-dimensional coordinates, initial priority, generation time, maximum allowable waiting time and real-time environmental data of all task points. At the same time, it obtains the real-time location, current battery level and remaining fire extinguishing bombs of the four drones, as well as the real-time location, number of backup batteries and remaining fire extinguishing bombs of the unmanned vehicle and other resource status information.

[0104] Based on the collected information, a dynamic task priority model is constructed, according to the formula. Calculate the real-time dynamic priority of each task point. Among them, the task point in the area where personnel are trapped has a high initial priority due to its excessive ambient temperature and CO concentration. The value is greater than 1, and the priority increases continuously with the waiting time; the initial priority of fire monitoring task points is low. When the value is close to 1, the priority increase is relatively gradual.

[0105] Simultaneously, a dynamic resource priority model is constructed, according to the formula... Calculate the resource priority for each drone, where the weighting coefficient is set to... =0.5、 =0.3、 =0.2, which satisfies This makes resource priority positively correlated with the drone's current battery level, with drones having sufficient battery power, ample payload capacity, and proximity to the mission location receiving higher resource priority.

[0106] Construct a matching degree matrix based on dynamic task priority and dynamic resource priority. The task allocation problem is formalized as a 0-1 integer linear programming problem, with the optimization objective being to maximize the total matching degree, and the constraints being:

[0107] , , ;

[0108] The optimal task allocation scheme is obtained by using the Hungarian algorithm.

[0109] When new personnel are reported to be stranded at mission locations, or when the drone's battery level or payload capacity changes significantly, the dynamic priority is recalculated and the matching algorithm is executed to achieve dynamic task redistribution.

[0110] After task allocation, the 50 task points were spatially clustered using a weighted K-means clustering algorithm.

[0111] According to the formula Update: Prioritize tasks by task priority to move higher priority tasks closer to the cluster center.

[0112] Calculate the spatial radius of each cluster. If the radius exceeds half of the maximum flight range of the UAV (i.e., the flight range corresponding to 15 minutes), increase the number of clusters and re-cluster them. Finally, obtain 4 clusters that satisfy the flight range constraint. The cluster center serves as the take-off and landing point of the UAV and the docking point of the unmanned vehicle.

[0113] For intra-cluster path planning of UAVs, with optimization as the goal, the ant colony algorithm is used to plan the optimal flight path to ensure that high-priority task points are executed first; for inter-cluster path planning of unmanned vehicles, with the goal of minimizing the total travel distance, the ant colony algorithm is used to plan the travel path of the unmanned vehicle to visit the four cluster centers in sequence, so as to provide resupply support for the UAVs.

[0114] The dispatch system monitors the drone's status data in real time at a frequency of 1Hz and sets multi-level resupply trigger thresholds.

[0115] The second battery threshold is a non-negotiable safety baseline; the drone's current battery level must be below 25% of its full charge.

[0116] Regardless of the dynamic return-to-home decision, the drone will immediately abort its mission and return to base.

[0117] The first battery threshold is the trigger condition for dynamic return-to-home decision-making; it occurs when the drone's current battery level is below 40% of its full charge.

[0118] When the failure rate is not less than 25%, a trade-off calculation for dynamic return-to-home decision-making is initiated;

[0119] The payload threshold is the trigger condition for dynamic return-to-home decision-making. It is activated when the UAV's mission payload margin is less than 2 units.

[0120] The trade-off calculation of dynamic return decision.

[0121] Simultaneously, a return-to-base decision is made based on the dynamic priority of the task and the remaining executable capacity, according to the formula. A judgment is made, where γ is set to 0.8.

[0122] For example, the dynamic priority of the current task being performed by a certain drone. With 30% battery remaining and a power consumption of 2% / min, the mission can last for 15 minutes. The return flight to the nearest cluster center will take 10 minutes. This is the highest priority mission to be executed. Substituting into the calculation, we get the left side. right side Since the left side is larger than the right side, the return-to-home command is not triggered, and the drone continues to perform its current task.

[0123] If the highest priority of the task to be executed is increased to 11, and the calculated value on the right is 99, which is still less than that on the left, the task continues to be executed.

[0124] If the highest priority of the task to be executed is increased to 12, the calculated value on the right is 108, which is greater than that on the left, triggering a return trip.

[0125] When the drone triggers the return-to-home command, the unmanned vehicle plans the optimal path based on the drone's estimated arrival time and moves to the corresponding cluster center's reception position. After the drone lands accurately, the unmanned vehicle completes battery replacement and fire extinguishing bomb reloading within 3 minutes. After resupply, the drone regains mission execution authority, and the scheduling system dynamically reallocates unfinished tasks, forming a closed-loop scheduling process.

[0126] The scheduling strategy is optimized online based on the MADDPG multi-agent deep reinforcement learning algorithm. The constructed state space includes information such as the dynamic priority of task points, remaining time windows, UAV battery and payload margins, unmanned vehicle resource margins, and environmental conditions. The action space includes the UAV's task selection and return-to-home decision, and the unmanned vehicle's movement decision. After training for 2000 episodes, the algorithm achieves policy convergence, realizing self-learning and adaptive optimization of the scheduling strategy, and achieving an optimal trade-off between task completion rate, priority satisfaction, and system energy consumption.

[0127] The above description is only a preferred embodiment of the present invention. It should be noted that those skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be considered within the scope of protection of the present invention.

Claims

1. A closed-loop cooperative scheduling method for UAVs and unmanned vehicles based on dynamic task priority, characterized in that: Includes the following steps: S1: Obtain a set of task points to be processed. Each task point includes spatial coordinates, initial priority attribute, generation time, maximum allowable waiting time, and real-time environmental information. Obtain resource status information of UAVs and unmanned vehicles. The resource status information of UAVs includes real-time location, current battery level, and remaining task payload. The resource status information of unmanned vehicles includes real-time location, current battery level, and remaining payload. S2: Construct a task dynamic priority model based on initial priority attributes, generation time, maximum allowable waiting time, and real-time environmental information; construct a resource dynamic priority model based on the UAV's real-time location, current battery level, and task payload margin, wherein the resource priority determined by the resource dynamic priority model is positively correlated with the UAV's current battery level; and assign tasks to corresponding UAVs according to the matching relationship between the task dynamic priority model and the resource dynamic priority model. S3: Based on the initial priority attribute, perform spatial clustering on each task point, and determine the cluster center as the take-off and landing point of the UAV and the parking point of the unmanned vehicle; with the optimization objective of minimizing the correlation between the initial priority attribute and the execution order, plan the flight path of the UAV in each cluster and the driving path of the unmanned vehicle between each cluster. S4: Establish a real-time monitoring mechanism for UAV status and set multi-level resupply trigger thresholds. The monitoring mechanism includes monitoring the current battery level and remaining mission payload of the UAV, and determining the remaining executable capacity based on the current battery level; determining the return decision based on the comparison result between the mission priority output by the mission dynamic priority model and the remaining executable capacity; adjusting the UAV to the pick-up position based on the real-time position of the UAV; forming a closed-loop scheduling process of mission execution, status monitoring, resupply decision and mission reallocation. S5: Based on a multi-agent cooperative optimization algorithm, the scheduling strategy is continuously optimized with task performance and resource consumption as optimization objectives.

2. The UAV-Vehicle Closed-Loop Cooperative Scheduling Method Based on Dynamic Task Priority as described in claim 1, characterized in that: The task dynamic priority model calculates the dynamic priority of task points according to the following formula: ， in, For the initial priority attribute, For the current moment, For the generation time, The maximum allowed waiting time, For time weighting coefficients, Environmental factors; Environmental factors are determined according to the following formula: ， in, For real-time ambient temperature, Real-time concentration of harmful gases This is a temperature reference threshold. This is the concentration reference threshold. , These are the weighting coefficients.

3. The UAV-Vehicle Closed-Loop Cooperative Scheduling Method Based on Dynamic Task Priority as described in claim 1 or 2, characterized in that: The resource dynamic priority model calculates the resource priority of the UAV according to the following formula: ， in, For resource priority, This is the current battery level. Full charge This represents the mission payload margin. For maximum load, Distance to the nearest mission point For the maximum acceptable distance, , , These are the weighting coefficients. A positive coefficient ensures that resource priority is positively correlated with current power consumption.

4. The UAV-Vehicle Closed-Loop Cooperative Scheduling Method Based on Dynamic Task Priority as described in claim 1, characterized in that: Based on the initial priority attribute, spatial clustering is performed on each task point, and the cluster centers are updated according to the following formula: ， in, For the first Cluster centers, For the first One cluster, This is the initial priority attribute. The coordinates of the task point; Calculate the spatial radius of each cluster ,when When the drone's range exceeds half of the stated range, increase the number of clusters and re-cluster the data.

5. The UAV-Vehicle Closed-Loop Cooperative Scheduling Method Based on Dynamic Task Priority as described in claim 1, characterized in that: The optimization objective is to minimize the correlation between the initial priority attribute and the execution order, and it is determined according to the following formula: ， in, To optimize the target value, For the number of clusters, This is the initial priority attribute. For the task point In its respective number The order of execution within each cluster.

6. The UAV-Vehicle Closed-Loop Cooperative Scheduling Method Based on Dynamic Task Priority as described in claim 1, characterized in that: The dynamic return-to-home decision only takes effect when the drone's battery level is between 40% and 25% of its full charge, or when the mission payload margin is less than 2 units. The return-to-home decision is determined according to the following formula: ， in, Priority of the current task point Remaining battery power Energy consumption per unit time A set of tasks to be executed. This refers to the flight time required for the drone to return from its current location to the nearest cluster center. This is an adjustment factor that triggers a return trip when the above conditions are met.

7. The UAV-Vehicle Closed-Loop Cooperative Scheduling Method Based on Dynamic Task Priority as described in claim 6, characterized in that: Multi-level supply trigger thresholds are executed in descending order of priority, including: The second battery threshold is an insurmountable safety baseline. When the drone's current battery level is less than 25% of its full charge, the drone will immediately abort its mission and return to base regardless of the dynamic return-to-home decision. The first battery threshold is the trigger condition for dynamic return-to-home decision-making. When the current battery level of the drone is lower than 40% of the full battery level but not lower than 25%, the trade-off calculation for dynamic return-to-home decision-making is initiated. The load threshold is the trigger condition for dynamic return-to-home decision-making. When the UAV's mission load margin is less than 2 units, the trade-off calculation for dynamic return-to-home decision-making is initiated.

8. The UAV-Vehicle Closed-Loop Cooperative Scheduling Method Based on Dynamic Task Priority as described in claim 1 or 6, characterized in that: Adjusting the autonomous vehicle to the pick-up location based on its real-time location includes: The drone calculates the estimated arrival time based on the return-to-home decision and sends the information to the unmanned vehicle. The driverless vehicle moves to the pick-up location based on the estimated arrival time and the driverless vehicle's real-time location; After landing, the drone performs battery replacement or payload loading.

9. The UAV-Vehicle Closed-Loop Cooperative Scheduling Method Based on Dynamic Task Priority as described in claim 1, characterized in that: The multi-agent cooperative optimization algorithm adopts the MADDPG algorithm framework, and the state space includes the remaining time window of each task point, the current battery level of the UAV, and the remaining amount of the unmanned vehicle's onboard supplies. The action space includes task allocation, path fine-tuning, and return-to-base decisions; The reward function is determined based on the weighted difference between task completion rate, priority satisfaction, and system energy consumption.

10. The UAV-Vehicle Closed-Loop Cooperative Scheduling Method Based on Dynamic Task Priority as described in claim 1, characterized in that: The set of tasks to be processed is located in emergency rescue scenarios, area reconnaissance scenarios, or target search and strike scenarios; The mission payload margin includes the amount of fire extinguishing munitions, the endurance of reconnaissance equipment, or the weapon payload.

Images