Multi-unmanned aerial vehicle dynamic task scheduling method and system based on double-layer collaborative optimization

Abstract

The invention relates to the technical field of unmanned aerial vehicle systems, in particular to a multi-unmanned aerial vehicle dynamic task scheduling method based on double-layer collaborative optimization, and the method comprises the following steps: S1, constructing scheduling input data; s2, upper-layer task allocation optimization is carried out; s3, lower-layer path planning and feasibility evaluation are carried out; s4, bidirectional collaborative feedback optimization is carried out; s5, dynamic task processing and incremental scheduling; s6, carrying out robustness recovery and adaptive adjustment; and S7, outputting and executing a scheduling result. According to the invention, by introducing a double-layer collaborative optimization architecture, the problem that task allocation and path planning are mutually separated in traditional multi-unmanned aerial vehicle scheduling is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) system technology, and in particular to a method and system for dynamic task scheduling of multiple UAVs based on two-layer collaborative optimization. Background Technology

[0002] With the widespread application of multi-drone systems in fields such as inspection and monitoring, disaster response, emergency rescue, urban governance, and logistics distribution, how to efficiently schedule multiple drones in multi-task scenarios has become a core issue. Existing multi-drone task scheduling technologies mainly include the following solutions: (1) Single-layer optimization method (mainly based on static scheduling) In scenarios with small task size and relatively stable environmental changes, single-level allocation or static programming methods are typically employed. For example, the Hungarian algorithm is suitable for linear programming allocation in small-scale static tasks; auction algorithms or consensus-based binding algorithms are suitable for distributed decision-making, but their adaptability to dynamic environmental changes is limited; rule-based greedy methods are simple to implement and efficient, but it is difficult to guarantee global optimum. While these methods still have some practical value in static task scenarios, they generally lack effective support for complex constraints and collaborative structures.

[0003] (2) Two-stage separation method of "task allocation - path planning" The current mainstream technical approach is to first allocate tasks and then plan paths. The first stage often uses algorithms such as the Hungarian algorithm, auction algorithm, or consensus binding algorithm to allocate tasks; the second stage calls A*, Dijkstra's algorithm, genetic algorithm, or local optimization methods to calculate the UAV path. A common characteristic of these methods is that the transition between the two stages is unidirectional; the path planning result cannot influence the task allocation decision in reverse. Therefore, it is difficult to obtain a globally optimal solution under multi-tasking, large-scale, or strongly constrained conditions, easily leading to decreased feasibility or insufficient overall efficiency.

[0004] (3) Research on multi-agent dynamic scheduling is still immature Some studies have attempted to address dynamic issues such as real-time task arrival, cancellation, priority changes, and UAV status changes, but most of them are based on local patching, task insertion, or simple replanning mechanisms. They lack a unified framework that can handle multiple types of disturbances simultaneously, resulting in limited overall system robustness and adaptability.

[0005] (4) The multi-algorithm scheduling platform lacks engineering design. Existing research largely focuses on verifying the performance of single algorithms, with code primarily experimental in nature. Algorithm structures are typically fixed and difficult to replace, and data formats vary across algorithms, making it challenging to build scalable platforms for comparing and integrating multiple algorithms. While this approach can be used for algorithm reproduction and paper verification, it is unsuitable for engineering scheduling systems and limits further exploration and application of collaborative optimization methods.

[0006] The shortcomings of existing technology: (1) The scheduling steps are complicated and lack an overall coordination mechanism. Traditional methods typically treat task allocation and path planning as two separate optimization problems, resulting in a "split allocation-path" process. Allocation algorithms focus only on task matching, neglecting path feasibility; path planning algorithms only accept a fixed set of tasks and cannot influence higher-level decisions. This process involves numerous steps, a long computational chain, and a lack of information fusion between modules, leading to cumbersome engineering processes, low optimization efficiency, and difficulty in reaching globally consistent optimal decisions.

[0007] (2) The overall efficiency is low, and it is difficult to maintain performance under medium to large-scale tasks. As the number of drones and missions increases, traditional single-layer optimization or heuristic scheduling methods struggle to handle high-dimensional combinatorial problems. Because path planning is performed only after the allocation result is fixed, upper-layer decisions do not fully consider the execution cost of lower layers, lack a global feedback mechanism, and are prone to getting trapped in local optima. This leads to infeasible paths, increased overall costs, and decreased mission completion rates, resulting in significant performance degradation of the system in large-scale scenarios.

[0008] (3) It has weak dynamic task processing capabilities and cannot respond efficiently to real-time changes. In practical applications, tasks often become new, canceled, delayed, or have changed priorities, and drones may also experience status changes such as low battery or communication interruptions. Existing scheduling systems are mostly based on static optimization ideas and lack event-triggered and incremental optimization mechanisms, often requiring overall replanning, resulting in slow response times and difficulty in meeting the real-time requirements of engineering projects.

[0009] (4) Insufficient robustness; abnormal situations can easily lead to scheduling failure. Existing systems typically lack real-time monitoring and health management mechanisms for drone status, and do not integrate task recovery, redistribution, or path restoration strategies. When drones experience low battery, execution failures, or obstacle avoidance anomalies, the system often lacks adaptive recovery capabilities, leading to the interruption or even failure of the entire scheduling process.

[0010] (5) The algorithm platform is closed, with poor scalability and maintainability. Most systems only support a single, fixed scheduling algorithm. The algorithm is tightly coupled with the system, lacking a unified data interface and modular design. Each algorithm relies on different input and output formats, making them difficult to replace or combine. This results in poor platform scalability, hinders algorithm upgrades and performance comparisons, and limits the long-term evolution and engineering reuse of the system.

[0011] (6) The project lacks feasibility and is difficult to support actual deployment scenarios. Existing scheduling systems are mostly still in the theoretical research or simulation experiment stage, lacking compatibility, stability, and the ability to integrate multiple algorithms. They cannot adapt to different scenarios, scales, and constraints on a unified platform. The systems lack standardized data feedback and multi-dimensional evaluation systems, making them difficult to apply directly to real engineering environments such as emergency response and inspection monitoring. Summary of the Invention

[0012] This invention provides a method and system for dynamic task scheduling of multiple unmanned aerial vehicles (UAVs) based on two-layer cooperative optimization. See the description below for details: A dynamic task scheduling method for multiple UAVs based on two-layer collaborative optimization includes the following steps: Step S1: Construction of scheduling input data: Collect and integrate task information, UAV status information, and environmental and dynamic event information to form a standardized scheduling input dataset; Step S2: Upper-level task allocation optimization: Based on the scheduling input dataset, the task allocation algorithm is invoked to generate candidate task allocation schemes; Step S3: Lower-level path planning and feasibility assessment: For each UAV in the candidate task allocation scheme, call the path planning algorithm to calculate its flight trajectory, execution cost and output path feasibility marker; Step S4: Two-way collaborative feedback optimization: The path execution cost and feasibility marker obtained in step S3 are fed back to the upper-level task allocation optimization process; the upper-level task allocation algorithm adjusts the optimization objective or search strategy based on the feedback information, generates new candidate task allocation schemes, and iteratively executes steps S3 to S4 until the collaborative optimization termination condition is met. Step S5: Dynamic Task Processing and Incremental Scheduling: When a dynamic event is detected, an incremental optimization based on the current scheduling scheme is triggered to update task allocation and path planning; Step S6: Robust recovery and adaptive adjustment: Monitor the drone status during execution. When an anomaly is detected, trigger the recovery mechanism to reallocate or replan the affected tasks. Step S7: Output and execution of scheduling results: Output the final task allocation scheme and the flight paths of each UAV, and send them to the execution layer.

[0013] Preferably, the task allocation algorithm includes at least one of genetic algorithm, particle swarm optimization algorithm, ant colony optimization algorithm, Hungarian algorithm, auction algorithm, consensus-based bundling algorithm, or scheduling method based on deep reinforcement learning; the path planning algorithm includes at least one of path optimization based on genetic algorithm, A* search algorithm, Dijkstra algorithm, 2-opt / 3-opt local optimization algorithm, or path prediction model based on deep learning.

[0014] Preferably, the dynamic events include at least one of the following: new task arrival, task cancellation, task priority change, UAV status abnormality, or environmental constraint change; the incremental optimization includes local task reallocation, local path replanning, or global re-optimization based on the previous round's global solution.

[0015] Preferably, in step S6, the anomaly includes the drone's battery level being below a threshold, communication interruption, task execution failure, or sudden path infeasibility; the recovery mechanism includes: marking the task undertaken by the abnormal drone as a task to be reclaimed, and re-inserting it into the task pool to be assigned in the currently available drone set, triggering local or global rescheduling.

[0016] Preferably, in step S4, the termination condition for collaborative optimization is reaching the preset maximum number of iterations, or the improvement of the overall system cost function F in multiple consecutive iterations is less than a set threshold; the overall system cost function F is the sum of the comprehensive costs f(i) of all UAVs, where the comprehensive cost f(i) = w1•D(i) + w2•T(i) + w3•R(i) + w4•U(i), D(i), T(i), R(i), and U(i) are the flight distance, execution time, risk cost, and load imbalance of UAV i, respectively, and w1 to w4 are adjustable weight coefficients.

[0017] Preferably, after step S6 or execution feedback, the weight coefficient of the next scheduling cycle is adaptively updated according to the monitored deviation information. The update formula is: w(k+1) = (1 − η)•w(k) + η•ŵ(k), where w(k) is the current cycle weight vector, ŵ(k) is the expected weight vector generated according to the current state, η is the adjustment step size, and 0<η≤1.

[0018] The present invention also discloses a multi-UAV dynamic task scheduling system based on two-layer cooperative optimization for implementing the method, comprising: The input layer module is used to collect and manage task information, UAV status information, and environmental and dynamic event information to form scheduling input data. The scheduling core layer module includes: The upper-level task allocation module is used to run a task allocation algorithm based on the scheduling input data and generate a task allocation scheme; The lower-level path planning module is used to run a path planning algorithm based on the task allocation scheme, generate UAV tracks and evaluate feasibility. The bidirectional collaboration module is used to establish a feedback channel between the upper-level task allocation module and the lower-level path planning module, so as to realize iterative collaborative optimization of task allocation and path planning. The adaptive and robust layer module is used to monitor the system status and trigger incremental scheduling or robust recovery processes when dynamic events or anomalies occur. The execution layer module is used to convert the optimized scheduling scheme into control commands and send them to the UAV, and to collect execution feedback information.

[0019] Preferably, the upper-layer task allocation module and the lower-layer path planning module adopt a pluggable algorithm interface design, which supports the rapid replacement and combination of different algorithm implementations that conform to the interface specification.

[0020] Preferably, the adaptive and robust layer module includes an anomaly detection unit and a recovery strategy unit; the anomaly detection unit analyzes the UAV's battery level, communication status, and task execution deviation in real time; the recovery strategy unit invokes the corresponding task recycling, reallocation, or path replanning strategy according to the anomaly type.

[0021] The beneficial effects of the technical solution provided by this invention are as follows: By introducing a two-layer collaborative optimization architecture, this invention effectively solves the problem of the separation between task allocation and path planning in traditional multi-UAV scheduling. The upper-layer task allocation module and the lower-layer path planning module iteratively optimize through a bidirectional feedback mechanism, ensuring that task allocation decisions fully consider the actual path cost and feasibility, thereby quickly generating a scheduling scheme with better global performance and higher feasibility under complex constraints. This method significantly improves the overall scheduling efficiency and quality in medium-to-large-scale task scenarios and has the ability to respond quickly to dynamic task arrival, cancellation, and environmental disturbances.

[0022] Furthermore, this invention integrates robust recovery and a pluggable algorithm platform design, greatly enhancing the system's engineering practicality and scalability. Through real-time status monitoring and anomaly handling mechanisms, the system can maintain mission continuity in the event of UAV failures or unexpected situations. The modular, unified interface platform architecture supports flexible replacement and comparative testing of different scheduling and planning algorithms, facilitating algorithm iteration and research, and enabling the framework to quickly adapt to various practical application scenarios, achieving a leap from theoretical algorithms to stable and reliable engineering systems. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the present invention; Figure 2 This is a flowchart of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0025] This invention provides a dynamic task scheduling method for multiple unmanned aerial vehicles (UAVs) based on two-layer collaborative optimization, applicable to multi-task, multi-constraint, and dynamically changing environment scenarios, comprising the following steps: Step S1: Constructing the scheduled input data The system collects task information from the task management module, UAV resource status information from the UAV status management module, and environmental constraint information from the environment and dynamic events module. It then performs unified modeling and time synchronization on the collected information to form a standardized scheduling input dataset.

[0026] Step S2: Optimize upper-level task allocation Based on the scheduling input data, the tasks and UAVs are uniformly coded, and a task allocation algorithm is invoked to generate candidate task allocation schemes; the task allocation algorithm includes, but is not limited to, genetic algorithm, particle swarm optimization algorithm, ant colony algorithm, Hungarian algorithm, auction algorithm, CBBA algorithm or scheduling methods based on deep reinforcement learning.

[0027] Step S3: Lower-level path planning and feasibility assessment Based on the candidate task allocation scheme, a path planning algorithm is executed for each UAV to calculate the corresponding flight trajectory, execution time, energy consumption and risk cost, and output the path feasibility label; the path planning algorithm includes, but is not limited to, path optimization algorithms based on genetic algorithms, graph search algorithms based on A* or Dijkstra, local optimization algorithms based on 2-opt or 3-opt, and path prediction models based on deep learning.

[0028] Step S4: Two-way collaborative feedback optimization The path cost, task execution time, risk indicators, and infeasibility markers obtained in the path planning stage are fed back to the task allocation process. The upper-level task allocation algorithm adjusts the fitness function, constraint weights, or search space based on the feedback information to achieve bidirectional collaborative optimization between task allocation and path planning.

[0029] Step S5: Dynamic Task Processing and Incremental Scheduling When dynamic events such as new tasks, canceled tasks, changes in task priority, or changes in UAV status are detected, a local replanning or overall re-optimization process is triggered, and incremental optimization is performed using the previous round of scheduling results as the initial solution.

[0030] Step S6: Robust Recovery and Adaptive Adjustment During execution, the drone's operating status is monitored in real time. When low battery, loss of connection, path failure, or mission execution abnormality is detected, a robust recovery mechanism is triggered to reclaim and redistribute the affected missions and dynamically adjust the path or scheduling strategy.

[0031] Step S7: Output and execution of scheduling results Output the final task – the UAV allocation matrix and the feasible flight paths of each UAV – and send them to the execution layer for task execution; continuously collect execution feedback information during the execution process, and enter the next scheduling cycle.

[0032] The present invention also provides a multi-UAV dynamic task scheduling system for implementing the above method, comprising: The input layer module is used to collect and manage task information, UAV status information, and environmental and dynamic event information. The scheduling core layer module includes an upper-layer task allocation module, a lower-layer path planning module, and a two-way collaboration module; The adaptive and robust layer module is used to perform anomaly detection, state assessment, and schedule recovery; The execution layer module is used to convert scheduling results into flight commands that the UAV can execute and to collect execution feedback.

[0033] in: The upper-layer task allocation module is used to generate a task allocation scheme based on the task set and the UAV resource status, and to call a pluggable task allocation algorithm for optimization. The lower-level path planning module is used to generate UAV flight paths based on the task allocation scheme, and to calculate path costs, risk indicators and feasibility results; The bidirectional collaborative module is used to establish a feedback mechanism between the task allocation module and the path planning module to achieve joint optimization; The adaptive and robust layer module is used to trigger scheduling correction or replanning when dynamic disturbances or drone anomalies occur. The execution layer module is used to issue scheduling results and continuously collect system operation feedback.

[0034] like Figure 2 As shown, the overall operation of the system unfolds sequentially from left to right horizontally and from top to bottom vertically, with each functional module achieving real-time collaboration and feedback loop through directed or bidirectional data paths.

[0035] First, the task management module, UAV status management module, and environment and dynamic event module collect task requirements, onboard status information, and external disturbance events, respectively, and then converge them into the input data aggregation module to form a standardized scheduling input dataset. This dataset includes key elements such as task sets, UAV capability vectors, remaining battery power, position coordinates, and environmental risk factors. The input data is then sent to the scheduling core layer. The scheduling core layer consists of a top-down three-stage structure: an upper-level task allocation module, a two-way collaboration module, and a lower-level path planning module. Specifically, the upper-level task allocation module generates an initial task allocation scheme based on the overall objective; the two-way collaboration module performs an adaptive evaluation of the scheme and passes necessary supplementary constraints to the lower-level path planning module; the lower-level path planning module generates the optimal trajectory that meets cost, risk, and feasibility constraints. The upper and lower layers exchange data in real-time through the two-way collaboration module, achieving deep coupling optimization between task allocation and path planning. The path planning results are then passed to the execution layer, where the scheduling output module issues flight commands to the multi-UAV execution system. Deviations, abnormal events, or state changes generated by the execution system during execution are monitored in real time and synchronously fed back to the adaptive and robust layer. The robust recovery module evaluates the feedback information to determine whether it triggers task replanning, path fine-tuning, or behavioral constraint correction. If adjustments are needed, the robust layer will push the corrected constraints or deviation compensation amounts back to the bidirectional collaborative module of the scheduling core layer, thereby achieving closed-loop robust optimization of the entire system.

[0036] The multi-UAV dual-layer collaborative scheduling system of this invention consists of an input layer, a scheduling core layer, an adaptive and robust layer, and an execution layer. Each module continuously interacts via a data bus to complete the entire scheduling process from task input to flight execution. The system emphasizes the design principles of "synchronous optimization, continuous feedback, and dynamic adaptation" during operation, enabling the UAV swarm to maintain efficient, stable, and controllable collaborative capabilities in complex environments.

[0037] After the system starts, the input layer first aggregates external environment and task information. The task management module is responsible for receiving, parsing, and standardizing tasks from the upper-level system, including the work area, task type, and time requirements; the UAV status management module collects the location, battery level, payload, and health status of each UAV in real time to form a set of currently available resources; the environment and dynamic events module further supplements external conditions such as no-fly zones, weather changes, and emergencies. After the three types of data are standardized in format and aligned in time sequence, they form the input data packets required for scheduling, providing an accurate foundation for core scheduling.

[0038] After the data enters the scheduling core layer, the two-layer collaborative optimization mechanism of this invention will be activated. The upper-layer task allocation module generates an initial allocation scheme based on the task set and UAV resources, while the lower-layer path planning module calculates the UAV's flight route, trajectory cost, and reachability constraints based on this. The bidirectional collaborative module is responsible for establishing a closed-loop feedback between the two layers: the lower layer returns the actual path cost, execution time, risk indicators, and load balancing status to the upper layer, and the upper layer readjusts the allocation results based on this feedback. The entire two-layer structure is evaluated through a unified fitness function, which integrates indicators such as distance, time, risk, and imbalance, and can automatically adjust weights according to the scenario, enabling the system to continuously iterate between global and local aspects, ultimately converging a scheduling scheme that balances overall efficiency and individual executability.

[0039] After the scheduling core layer generates candidate solutions, the system enters the adaptive and robust layer for robustness review. The robustness recovery module verifies the feasibility of the solution based on the drone's remaining battery power, path safety, execution time margin, and task conflict status. If potential risks are detected, such as insufficient resources for a drone, flight paths crossing dangerous areas, or overall unreasonable allocation, the robust layer generates adjustment instructions and returns them to the scheduling core layer. The scheduling core layer then adjusts the fitness weights or performs local re-optimization until the robust layer confirms that the solution has sufficient safety margin and fault tolerance under the current environmental conditions. This layer maintains dynamic coupling with the core scheduling through bidirectional feedback, enabling the system to quickly self-adjust in the event of sudden environmental changes or resource fluctuations.

[0040] The final scheduling scheme, after robustness verification, will enter the execution layer. The scheduling output module converts the optimized task allocation and trajectory planning into flight control commands that the UAVs can directly execute, including task sequence, waypoints, speed limits, and time window requirements. The multi-UAV execution system carries out tasks according to the commands and transmits status information back in real time. During execution, the system continuously monitors the UAVs' battery levels, attitude deviations, path deviations, and unexpected events. Once an anomaly occurs, the relevant information will immediately re-enter the input layer and robust layer, triggering local rescheduling or path replanning to ensure mission continuity and safety.

[0041] This invention constructs a dynamic, stable, and efficient intelligent collaborative scheduling system for UAV swarms through data integration at the input layer, two-layer collaborative optimization at the scheduling core layer, robustness guarantees at the adaptive and robust layers, and real-time monitoring at the execution layer. The core advantages of this system include: deep coupling of task allocation and path planning, adaptive adjustment of multi-index fitness functions, and a dedicated robust recovery mechanism with bidirectional closed-loop feedback between the scheduling core layer and the system. This enables the system to achieve high-precision, high-security, and highly scalable swarm collaboration in complex environments.

[0042] To make the overall operational logic of the multi-UAV dynamic task scheduling system of this invention more complete, and to further explain the unified optimization framework upon which this invention is based, a mathematical modeling description is given for the key objects, constraints, and two-level collaborative optimization relationships involved in the system. This section aims to characterize the methodological principles of this invention at the system level, including the collaborative effects of three levels: task allocation, path planning, and robust adaptive adjustment, to aid in understanding the aforementioned technical solutions, and does not constitute a limitation on the scope of protection of this invention. It provides complete results for engineering deployment, simulation verification, or task execution.

[0043] (1) System objects and basic sets UAV ensemble: U = {1, 2, …, N} The dynamic task set is: T(k) = {1, 2, …, M(k)}, where k represents the scheduling period; when a new task τ_new arrives, T(k+1) = T(k) ∪ {τ_new}.

[0044] The environmental state set, E(k), is provided by sensors or external data sources and includes information such as wind field, risk area, and obstacles.

[0045] (2) Task allocation decision variables Define variables for task j and drone i: x(i,j) = 1 indicates that drone i performs task j; x(i,j) = 0 means that drone i does not perform task j.

[0046] This results in the task allocation matrix X = [x(i,j)].

[0047] (3) Key variables in path planning and execution layer The path planning module outputs the following parameters: P(i): The flight path of UAV i; D(i): Total flight distance; T(i): Total execution time; R(i): Risk cost; U(i): Task load imbalance.

[0048] The above variables are the output of PathSolver(i, X, E(k)), and are determined simultaneously with the task assignment matrix and the environment state.

[0049] (4) System constraints Task uniqueness constraint: For all tasks j, Σ(i∈U) x(i,j) = 1.

[0050] Capacity constraint: For all UAVs i, Σ(j∈T(k)) x(i,j) ≤ C(i).

[0051] Range and time constraints: T(i) ≤ Tmax(i), D(i) ≤ Dmax(i).

[0052] Safety constraint: Conflict(P(i), P(l)) = 0 (i ≠ l), indicating that there is no conflict between the two drones in space and time.

[0053] (5) Path layer multi-index cost Define a comprehensive cost function for each drone: f(i) = w1•D(i) + w2•T(i) + w3•R(i) + w4•U(i). Where w1 to w4 are adaptively adjustable comprehensive weights.

[0054] (6) Overall optimization objective of two-level collaborative scheduling The path layer outputs the optimal index f*(i) for each UAV, and the overall system cost is: F = Σ(i∈U) f*(i).

[0055] It can be written as: F = Σ(i∈U)[w1·D(i) + w2·T(i) + w3·R(i) + w4·U(i)].

[0056] The goal of the upper-level GA is to minimize F and search for the optimal task assignment matrix X while satisfying all constraints.

[0057] (7) Robust adaptive mechanism under dynamic perturbation During the execution phase, the deviation δ(i,k) is calculated. When the deviation exceeds the threshold θ, the robust recovery process is triggered. The desired weight vector ŵ(k) is generated based on the environmental state and deviation information, and the weights for the next cycle are updated using the following formula: w(k+1) = (1 − η)•w(k) + η•ŵ(k).

[0058] Where η is the weight adjustment step size (0 < η ≤ 1).

[0059] (8) System-level optimization closed loop The scheduling system forms the following loop structure: Task set T(k) → Upper-level task allocation X → Lower-level path planning outputs P*(i) and f*(i) → Overall Cost F → Feedback δ(k) during the execution phase → Adaptive weight update w(k+1) → Proceed to the next cycle T(k+1).

[0060] This closed loop enables a dynamic adaptive process of planning, execution, feedback, and replanning, allowing the system to maintain scheduling stability and global optimization capabilities under conditions of task changes and environmental disturbances.

[0061] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A dynamic task scheduling method for multiple unmanned aerial vehicles (UAVs) based on two-layer collaborative optimization, characterized in that, Includes the following steps: Step S1: Construction of scheduling input data: Collect and integrate task information, UAV status information, and environmental and dynamic event information to form a standardized scheduling input dataset; Step S2: Upper-level task allocation optimization: Based on the scheduling input dataset, the task allocation algorithm is invoked to generate candidate task allocation schemes; Step S3: Lower-level path planning and feasibility assessment: For each UAV in the candidate task allocation scheme, call the path planning algorithm to calculate its flight trajectory, execution cost and output path feasibility marker; Step S4: Two-way collaborative feedback optimization: Feedback the path execution cost and feasibility marker obtained in step S3 to the upper-level task allocation optimization process; The upper-level task allocation algorithm adjusts the optimization target or search strategy based on the feedback information, generates new candidate task allocation schemes, and iteratively executes steps S3 to S4 until the collaborative optimization termination condition is met. Step S5: Dynamic Task Processing and Incremental Scheduling: When a dynamic event is detected, an incremental optimization based on the current scheduling scheme is triggered to update task allocation and path planning; Step S6: Robust recovery and adaptive adjustment: Monitor the drone status during execution. When an anomaly is detected, trigger the recovery mechanism to reallocate or replan the affected tasks. Step S7: Output and execution of scheduling results: Output the final task allocation scheme and the flight paths of each UAV, and send them to the execution layer.

2. The method of claim 1, wherein, The task allocation algorithm includes at least one of the following: genetic algorithm, particle swarm optimization algorithm, ant colony optimization algorithm, Hungarian algorithm, auction algorithm, consensus-based bundling algorithm, or scheduling method based on deep reinforcement learning; the path planning algorithm includes at least one of the following: path optimization based on genetic algorithm, A* search algorithm, Dijkstra algorithm, 2-opt / 3-opt local optimization algorithm, or path prediction model based on deep learning.

3. The method according to claim 1 or 2, characterized in that, The dynamic events include at least one of the following: new task arrival, task cancellation, task priority change, abnormal UAV status, or change in environmental constraints; the incremental optimization includes local task reassignment, local path replanning, or global re-optimization based on the previous round's global solution.

4. The method of claim 1, wherein, In step S6, the anomalies include the drone's battery level being below a threshold, communication interruption, task execution failure, or sudden path infeasibility; the recovery mechanism includes: marking the tasks undertaken by the abnormal drone as tasks to be reclaimed, and re-inserting them into the pool of tasks to be assigned in the currently available drone set, triggering local or global rescheduling.

5. The method of claim 1, wherein, In step S4, the termination condition for collaborative optimization is reaching the preset maximum number of iterations, or the improvement of the overall system cost function F in multiple consecutive iterations is less than a set threshold; the overall system cost function F is the sum of the comprehensive costs f(i) of all UAVs, where the comprehensive cost f(i) = w1·D(i) + w2·T(i) + w3·R(i) + w4·U(i), D(i), T(i), R(i), and U(i) are the flight distance, execution time, risk cost, and load imbalance of UAV i, respectively, and w1 to w4 are adjustable weight coefficients.

6. The method of claim 5, wherein, After step S6 or execution feedback, the weight coefficients of the next scheduling cycle are adaptively updated based on the monitored deviation information. The update formula is: w(k+1) = (1 − η)·w(k) +η·ŵ(k), where w(k) is the current cycle weight vector, ŵ(k) is the expected weight vector generated based on the current state, and η is the adjustment step size, 0 < η ≤ 1.

7. A multi-UAV dynamic task scheduling system based on double-layer collaborative optimization for implementing the method of any one of claims 1-6, characterized in that, include: The input layer module is used to collect and manage task information, UAV status information, and environmental and dynamic event information to form scheduling input data. The scheduling core layer module includes: The upper-level task allocation module is used to run a task allocation algorithm based on the scheduling input data and generate a task allocation scheme; The lower-level path planning module is used to run a path planning algorithm based on the task allocation scheme, generate UAV tracks and evaluate feasibility. The bidirectional collaboration module is used to establish a feedback channel between the upper-level task allocation module and the lower-level path planning module, so as to realize iterative collaborative optimization of task allocation and path planning. The adaptive and robust layer module is used to monitor the system status and trigger incremental scheduling or robust recovery processes when dynamic events or anomalies occur. The execution layer module is used to convert the optimized scheduling scheme into control commands and send them to the UAV, and to collect execution feedback information.

8. The system of claim 7, wherein, The upper-layer task allocation module and the lower-layer path planning module adopt a pluggable algorithm interface design, which supports the rapid replacement and combination of different algorithm implementations that conform to the interface specifications.

9. The system of claim 7 or 8, wherein, The adaptive and robust layer module includes an anomaly detection unit and a recovery strategy unit; the anomaly detection unit analyzes the UAV's battery level, communication status, and task execution deviation in real time; the recovery strategy unit calls the corresponding task recycling, reallocation, or path replanning strategy according to the anomaly type.

Images