An edge computing method, system, terminal and storage medium for real-time supervision tasks of low-altitude unmanned aerial vehicles

Abstract

The present invention discloses an edge computing method, system, terminal and storage medium for real-time supervision tasks of low-altitude unmanned aerial vehicles. The method includes: quantitatively expressing the spatio-temporal dependence characteristics of low-altitude unmanned aerial vehicle supervision tasks, and quantitatively evaluating the low-altitude unmanned aerial vehicle supervision tasks to obtain a quantitative evaluation result; performing fine-grained decomposition of edge-side spatio-temporal dependence tasks according to the quantitative evaluation result, and decomposing the low-altitude unmanned aerial vehicle supervision tasks into multiple fine-grained subtasks; constructing a fine-grained task execution plan that dynamically adapts to resource constraints and computing real-time performance in the edge environment, and establishing a multi-grained parallel execution strategy for fine-grained tasks at the edge side with a process-thread multi-grained mixture; initially allocating edge-side computing resources based on priorities and constraint conditions, and dynamically allocating and adaptively optimizing the edge-side computing resources. The present invention can significantly improve the real-time performance of edge computing for large-scale low-altitude unmanned aerial vehicle supervision tasks and improve the utilization rate of edge computing resources.

Description

Technical Field

[0001] The present invention relates to the technical field of unmanned aerial vehicle (UAV) data processing, and particularly to an edge computing method, system, terminal and computer-readable storage medium for real-time supervision tasks of low-altitude UAVs. Background Art

[0002] Performing real-time calculations and timely feedback on supervision tasks such as UAV anomaly detection and conflict warning is the core key to ensuring the safe flight of UAVs. In solving the problem of multi-UAV collision conflicts, currently, path planning algorithms and various artificial intelligence-based collision avoidance decision algorithms are mainly used, such as the A* search algorithm (heuristic search algorithm), ant colony algorithm, particle swarm optimization algorithm, etc. The above algorithms all focus on pre-planning before the task, and have a certain delay when dealing with emergencies encountered during the implementation of UAV tasks, and it is impossible to obtain a safe flight path through the planning algorithm in the face of an unfamiliar task environment.

[0003] Currently, research on behavior anomaly detection mostly focuses on objects such as large UAVs, large manned aircraft, vehicles, and crowds. For behavior anomaly detection algorithms for low-altitude small UAVs, there are isolation forest algorithms and support vector machines. Such methods require a large amount of UAV data and consume a large amount of training time, and cannot achieve real-time calculation and real-time feedback processing. Currently, relevant UAV supervision platforms have been developed, but most of the supervision tasks for UAVs have problems such as poor real-time performance, and cannot respond in a timely manner to emergencies encountered during the operation of UAVs, and there is often a large delay when calculating a large amount of data.

[0004] Therefore, in view of the problems of large calculation delay and inability to respond in a timely manner for large-scale low-altitude UAV supervision tasks, it is urgent to use edge computing technology to optimize the spatio-temporal flow calculation mode of supervision tasks such as UAV anomaly detection and early warning conflicts, so as to improve the real-time performance of large-scale low-altitude UAV supervision task calculations. Performing real-time calculations and timely feedback on supervision tasks such as anomaly detection and conflict prediction at the edge has obvious advantages. However, under the constraints of limited computing and storage resources of edge devices, there are still many challenges in achieving real-time response to local UAV supervision tasks.

[0005] Therefore, the prior art still needs to be improved and developed. Summary of the Invention

[0006] The main purpose of the present invention is to provide an edge computing method, system, terminal and computer-readable storage medium for real-time supervision tasks of low-altitude UAVs, aiming to solve the problems of large calculation delay and inability to respond in a timely manner for large-scale low-altitude UAV supervision tasks in the prior art.

[0007] To achieve the above object, the present invention provides an edge computing method for real-time supervision tasks of low-altitude unmanned aerial vehicles. The edge computing method for real-time supervision tasks of low-altitude unmanned aerial vehicles includes the following steps:

[0008] Quantitatively express the spatio-temporal dependence characteristics of the low-altitude UAV supervision task, and quantitatively evaluate the low-altitude UAV supervision task according to the quantitative expression result to obtain a quantitative evaluation result;

[0009] According to the quantitative evaluation result, perform fine-grained decomposition of the edge-side spatio-temporal dependence task, and decompose the low-altitude UAV supervision task into multiple fine-grained subtasks;

[0010] According to the multiple fine-grained subtasks, construct a fine-grained task execution plan that dynamically adapts to resource constraints and computing real-time performance in the edge environment, and establish a multi-grained parallel execution strategy for fine-grained tasks at the edge side with a process-thread multi-grained hybrid;

[0011] Based on priorities and constraint conditions, perform preliminary allocation of edge-side computing resources, and perform dynamic allocation and adaptive optimization of edge-side computing resources.

[0012] Optionally, in the edge computing method for real-time supervision tasks of low-altitude unmanned aerial vehicles, where the spatio-temporal dependence characteristics of the low-altitude UAV supervision task are quantitatively expressed, and the low-altitude UAV supervision task is quantitatively evaluated according to the quantitative expression result to obtain a quantitative evaluation result, specifically including:

[0013] Comprehensively analyze the timestamps, spatial location information, and the front-back dependence relationships between tasks of the real-time computing task, construct a spatio-temporal graph model expressing task spatio-temporal dependence, and the spatio-temporal graph model is used to construct task dependence relationships, quantify task execution constraints, and dynamically adjust task execution strategies;

[0014] Based on the changing characteristics of the real-time computing task over time and space, use a time series prediction method to model the task change trend, and combine historical data analysis to depict the dynamic evolution law of spatio-temporal dependence characteristics;

[0015] Based on the quantitative expression of spatio-temporal dependence characteristics, obtain computable modes of different tasks, and according to the computable modes, quantitatively evaluate the low-altitude UAV supervision task under the condition of limited resources at the edge side to obtain a quantitative evaluation result.

[0016] Optionally, in the edge computing method for real-time supervision tasks of low-altitude unmanned aerial vehicles, where the low-altitude UAV supervision task is quantitatively evaluated under the condition of limited resources at the edge side, specifically including:

[0017] Construct an objective function that reflects the requirements of spatio-temporal flow computing tasks and is suitable for optimization, and integrate resource consumption metrics into a unified computational volume evaluation model. The resource consumption metrics include CPU usage, memory occupancy, network transmission overhead, and energy consumption budget.

[0018] Define the computational volume evaluation model:

[0019] ;

[0020] Among them, represents the computational volume evaluation model, represents the task volume function, represents the number of drones, represents the total number of break points of the flight trajectory, representing the complexity of the trajectory, represents the time complexity calculation function of the drone supervision algorithm.

[0021] Optionally, for the edge computing method of the low-altitude drone real-time supervision task, the quantization evaluation results include functional requirements, input-output requirements, performance indicators, and key performance indicators.

[0022] Optionally, for the edge computing method of the low-altitude drone real-time supervision task, the fine-grained decomposition of the edge-side spatio-temporal dependent tasks according to the quantization evaluation results, and the low-altitude drone supervision task is decomposed into multiple fine-grained subtasks, specifically including:

[0023] According to functional requirements, input-output requirements, performance indicators, and key performance indicators, draw a detailed spatio-temporal dependency relationship map, including data flow, control flow, and resource dependencies, to visually display the associations between various parts of the task;

[0024] Select a modeling tool according to the task characteristics to intuitively represent the task structure and spatio-temporal characteristics, and build a comprehensive state space for the task, including all task states and the transition rules between task states;

[0025] According to the functional logic of the task, divide the large task into multiple small-scale and logically independent fine-grained subtasks, define clear input-output interfaces, use the dynamic programming algorithm to construct a state transition equation, and find the optimal decomposition scheme that satisfies the spatio-temporal dependency constraints.

[0026] Optionally, for the edge computing method of the low-altitude drone real-time supervision task, the construction of a fine-grained task execution plan that dynamically adapts to resource constraints and computational real-time in the edge environment, and the establishment of a multi-granularity hybrid parallel execution strategy for fine-grained tasks at the edge side, specifically including:

[0027] For the fine-grained subtasks after fine-grained decomposition, prioritize them according to the time urgency, data importance, and computational complexity of each fine-grained subtask, evaluate the resource requirements for each fine-grained subtask, and combine the real-time performance metrics of the edge device to construct a time window planning model. Arrange the execution order and start and end times of each fine-grained subtask through the time window planning model to ensure that all fine-grained subtasks are completed within the specified time window;

[0028] According to the characteristics of different types of fine-grained subtasks, establish a fine-grained task parallel execution model with a multi-granularity mixture of processes and threads, and implement hybrid-level parallel computing on the multi-core heterogeneous edge device through the fine-grained task parallel execution model.

[0029] Optionally, in the edge computing method for the real-time supervision task of low-altitude unmanned aerial vehicles, the preliminary allocation of edge computing resources based on priorities and constraint conditions, and the dynamic allocation and adaptive optimization of edge computing resources specifically include:

[0030] Perform real-time monitoring and quantitative analysis on limited computing resources, combine the characteristics and historical execution situations of spatio-temporal stream data processing tasks, use machine learning algorithms to predict the resource requirements in the next period of time, estimate the resource supply-demand relationship, optimize the priorities of each to-be-executed fine-grained subtask according to the urgency, timeliness, and importance of the tasks, and perform the initial computing resource allocation for each fine-grained subtask on demand using strategies such as fair scheduling, preemptive scheduling, and allocation on demand according to the current total resource amount, the resource requirements of each task, and the priorities;

[0031] After the preliminary allocation, establish a real-time feedback mechanism for edge resource utilization. When resource tension, load imbalance, or new tasks are detected, immediately trigger the resource dynamic reallocation process, adjust the allocated resources through the load balancing algorithm, design relevant constraint functions to balance the constraints on the computing resource distribution state and computing task completion state within an edge cluster, and construct an elastic resource allocation system performance model to dynamically adjust the edge resource allocation on the premise of meeting the task spatio-temporal dependence;

[0032] Evaluate the allocation effect according to the task completion time, resource utilization rate, and response speed, continuously iterate and optimize the resource allocation strategy, and combine the dynamic change trend of spatio-temporal stream data characteristics and the task demand fluctuations that will occur in the future to predict the change law of resource requirements, and automatically adjust and optimize the resource allocation strategy according to the evaluation feedback information to form a closed loop from monitoring, allocation, execution to feedback optimization.

[0033] In addition, to achieve the above object, the present invention also provides an edge computing system for the real-time supervision task of low-altitude unmanned aerial vehicles, wherein the edge computing system for the real-time supervision task of low-altitude unmanned aerial vehicles includes:

[0034] A task evaluation module, configured to quantitatively express the spatio-temporal dependence characteristics of the low-altitude UAV supervision task, and quantitatively evaluate the low-altitude UAV supervision task according to the quantitative expression result to obtain a quantitative evaluation result;

[0035] A task decomposition module, configured to perform fine-grained decomposition of the spatio-temporal dependence tasks at the edge end according to the quantitative evaluation result, and decompose the low-altitude UAV supervision task into multiple fine-grained subtasks;

[0036] A task execution module, configured to construct a fine-grained task execution plan that dynamically adapts to resource constraints and computing real-time performance in the edge environment according to the multiple fine-grained subtasks, and establish a fine-grained task parallel execution strategy with a multi-grained mixture of processes and threads at the edge end;

[0037] A resource allocation module, configured to perform preliminary allocation of edge-end computing resources based on priorities and constraint conditions, and perform dynamic allocation and adaptive optimization of edge-end computing resources.

[0038] In addition, to achieve the above object, the present invention further provides a terminal, wherein the terminal includes: a memory, a processor, and an edge computing program for the low-altitude UAV real-time supervision task stored on the memory and executable on the processor. When the edge computing program for the low-altitude UAV real-time supervision task is executed by the processor, the steps of the edge computing method for the low-altitude UAV real-time supervision task as described above are implemented.

[0039] In addition, to achieve the above object, the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores an edge computing program for the low-altitude UAV real-time supervision task. When the edge computing program for the low-altitude UAV real-time supervision task is executed by a processor, the steps of the edge computing method for the low-altitude UAV real-time supervision task as described above are implemented.

[0040] In the present invention, the spatio-temporal dependence characteristics of the low-altitude UAV supervision task are quantitatively expressed, and the low-altitude UAV supervision task is quantitatively evaluated according to the quantitative expression result to obtain a quantitative evaluation result; according to the quantitative evaluation result, fine-grained decomposition of the spatio-temporal dependence tasks at the edge end is performed, and the low-altitude UAV supervision task is decomposed into multiple fine-grained subtasks; according to the multiple fine-grained subtasks, a fine-grained task execution plan that dynamically adapts to resource constraints and computing real-time performance in the edge environment is constructed, and a fine-grained task parallel execution strategy with a multi-grained mixture of processes and threads at the edge end is established; preliminary allocation of edge-end computing resources is performed based on priorities and constraint conditions, and dynamic allocation and adaptive optimization of edge-end computing resources are performed. The present invention can significantly improve the real-time performance of edge computing for large-scale low-altitude UAV supervision tasks and improve the utilization rate of edge computing resources. Description of the Drawings

[0041] Figure 1 is a flowchart of a preferred embodiment of the edge computing method for the real-time supervision task of low-altitude unmanned aerial vehicles in the present invention;

[0042] Figure 2 is a schematic diagram of the overall process in a preferred embodiment of the edge computing method for the real-time supervision task of low-altitude unmanned aerial vehicles in the present invention;

[0043] Figure 3 is the overall technical roadmap in a preferred embodiment of the edge computing method for the real-time supervision task of low-altitude unmanned aerial vehicles in the present invention;

[0044] Figure 4 is a schematic diagram of the spatio-temporal dependence graph model of the unmanned aerial vehicle computing task in a preferred embodiment of the edge computing method for the real-time supervision task of low-altitude unmanned aerial vehicles in the present invention;

[0045] Figure 5 is a schematic diagram of the fine-grained decomposition of the unmanned aerial vehicle subtasks in a preferred embodiment of the edge computing method for the real-time supervision task of low-altitude unmanned aerial vehicles in the present invention;

[0046] Figure 6 is a schematic diagram of the parallel execution strategy method of the edge-end unmanned aerial vehicle subtasks in a preferred embodiment of the edge computing method for the real-time supervision task of low-altitude unmanned aerial vehicles in the present invention;

[0047] Figure 7 is a flowchart of the dynamic allocation strategy of the edge-end computing resources in a preferred embodiment of the edge computing method for the real-time supervision task of low-altitude unmanned aerial vehicles in the present invention;

[0048] Figure 8 is a structural diagram of a preferred embodiment of the edge computing system for the real-time supervision task of low-altitude unmanned aerial vehicles in the present invention;

[0049] Figure 9 is a structural diagram of a preferred embodiment of the terminal of the present invention. Detailed Embodiments

[0050] To make the objectives, technical solutions and advantages of the present invention clearer and more definite, the following further describes the present invention in detail with reference to the accompanying drawings and by way of examples. It should be understood that the specific examples described herein are only used to explain the present invention and are not used to limit the present invention.

[0051] The present invention is an edge computing method for real-time supervision tasks of low-altitude unmanned aerial vehicles. By quantitatively expressing the spatio-temporal dependence characteristics of supervision tasks such as abnormal detection and conflict warning of unmanned aerial vehicles, it realizes the fine-grained decomposition of spatio-temporal dependence tasks at the edge side, reduces the complexity of task processing at the edge side, and improves the balance and efficiency of task allocation; establishes a multi-granularity hybrid parallel execution strategy for fine-grained tasks at the edge side with processes and threads, and establishes an adaptive adjustment mechanism for the execution plan of fine-grained tasks at the edge side, effectively improving the concurrent execution ability at the edge side; aiming at the computing requirements and computing resource status of edge-side tasks, studies a dynamic allocation method for edge-side computing resources under limited resources, and establishes an optimization mechanism for dynamic allocation of edge-side computing resources to ensure reasonable and efficient resource allocation of computing tasks at the edge side.

[0052] The edge computing method for real-time supervision tasks of low-altitude unmanned aerial vehicles according to a preferred embodiment of the present invention is as Figure 1 、 Figure 2 (schematic diagram of the overall process) and Figure 3 (schematic diagram of the technical route) shown, and the edge computing method for real-time supervision tasks of low-altitude unmanned aerial vehicles includes the following steps:

[0053] Step S10: Quantitatively express the spatio-temporal dependence characteristics of the low-altitude unmanned aerial vehicle supervision task, and quantitatively evaluate the low-altitude unmanned aerial vehicle supervision task according to the quantitative expression result to obtain a quantitative evaluation result.

[0054] Specifically, first, comprehensively analyze the time stamps, spatial position information of real-time computing tasks, and the front-back dependence relationship between tasks, and construct a spatio-temporal graph model expressing task spatio-temporal dependence. The spatio-temporal graph model can express the spatio-temporal dependence of tasks, and is used to construct task dependence relationships, quantify the execution constraints of tasks, and dynamically adjust task execution strategies. According to factors such as the front-back execution order between tasks, data dependence relationships (such as causality, synchrony, or asynchrony), and data transmission delays, define and assign weights to the edges in the spatio-temporal graph. These weight parameters can quantify key characteristics such as the time sequence constraint strength, data dependence tightness, and delay tolerance between tasks, forming a multi-layer and multi-dimensional spatio-temporal dependence network.

[0055] Then, based on the changing characteristics of real-time computing tasks over time and space, use methods such as time series prediction to model the task change trend, and combine historical data analysis to depict the dynamic evolution law of spatio-temporal dependence characteristics. Spatio-temporal dependence characteristics refer to the characteristic dependencies in the time dimension, space dimension, and spatio-temporal interaction. In different computing tasks (such as Figure 4The manifestation forms of spatio-temporal dependence features in computing tasks 1 and 2 are also different. For example, in task scheduling, the manifestation form of spatio-temporal dependence features in the time dimension is the execution order of tasks, and in the space dimension, it is the distribution of computing resources of tasks on different edge computing nodes. At the spatio-temporal interaction level, it is manifested that the execution efficiency of tasks not only depends on the current time state but also on the spatial position of the UAV. Finally, based on the feedback of the actual operation situation, the edge weights are continuously adjusted and optimized to reflect the changes in the system state. For example, the delay tolerance threshold is dynamically updated according to factors such as the current resource load status and network condition changes. The construction of the spatio-temporal graph model first establishes data dependencies according to the relationships such as the execution order, synchronous and asynchronous, and transmission delay between different computing tasks, and then converts these data dependencies into edge set relationships between different computing tasks according to the weights, and distributes the computing tasks into the graph network and connects them with data dependencies to establish the spatio-temporal graph model (for example, including computing tasks 1 - computing task 7). That is, the construction of the spatio-temporal graph model refers to Figure 4 .

[0056] Finally, based on the quantitative expression of spatio-temporal dependence features, computable patterns of different tasks are obtained. According to the computable patterns, under the condition of limited resources at the edge, the low-altitude UAV supervision task is quantitatively evaluated to obtain a quantitative evaluation result. That is, through the quantitative expression of spatio-temporal dependence features of different tasks, abstract tasks are transformed into specific computable data forms, and then the quantitative evaluation results of these computable quantitative expressions are used for the quantitative evaluation of the task computing amount.

[0057] Since edge devices are usually limited by various factors such as computing power, storage capacity, and battery life, the calculation amount evaluation needs to consider factors such as CPU usage rate, memory occupancy, network transmission overhead, and energy consumption budget. By constructing an objective function that can reflect the requirements of spatio-temporal flow computing tasks and is suitable for optimization, the above resource consumption indicators are integrated into a unified calculation amount evaluation model (referring to the model used for quantitative evaluation of computing tasks at the edge). For large-scale UAV spatio-temporal flow computing tasks, the calculation amount estimation is also affected by factors such as the number of UAVs and the operation complexity of UAVs. Among them, the number of UAVs directly affects the task parallelism, cooperation complexity, and overall resource requirements. The flight trajectory planning, data collection, and processing strategies of each UAV will affect the calculation amount of the entire system. Therefore, the calculation amount will be reflected as the time complexity of the spatial calculation operator, which is determined by the specific process of the algorithm, the number of UAVs, and the operation complexity of UAVs. Define the calculation amount evaluation model as:

[0058] ;

[0059] Among them, represents the calculation amount evaluation model, represents the task amount function, represents the number of drones, represents the total number of turning points of the operation trajectory, which represents the complexity of the trajectory, represents the time complexity calculation function of the drone supervision algorithm.

[0060] Among them, the quantitative evaluation results include functional requirements, input / output requirements, performance indicators, and key performance indicators.

[0061] Step S20: Perform fine-grained decomposition of the edge-side spatio-temporal dependence tasks according to the quantitative evaluation results, and decompose the low-altitude drone supervision task into multiple fine-grained subtasks.

[0062] Specifically, first conduct a comprehensive quantitative evaluation of the supervision task, clarify its functional requirements, input / output requirements, performance indicators (such as response time, throughput, accuracy, etc.), and determine the key performance indicators. There are various types of computing tasks in the drone supervision task, and the spatio-temporal dependence characteristics of each task (such as Figure 5 spatio-temporal dependence characteristic 1, spatio-temporal dependence characteristic 2, spatio-temporal dependence characteristic n in it) are all different. It is necessary to perform quantitative expression of the spatio-temporal dependence characteristics of these tasks and then conduct quantitative evaluation of the computational complexity. Comprehensive quantitative evaluation means evaluating all different computing tasks; draw a detailed spatio-temporal dependence relationship map, including data flow, control flow, and resource dependence, to visually display the associations between various parts of the task and ensure that these associations can be completely retained during the subsequent decomposition process; select an appropriate modeling tool according to the task characteristics to more intuitively represent the task structure and its spatio-temporal characteristics; build a comprehensive state space for the task, including all task states and the conversion rules between them.

[0063] Then, according to the functional logic of the task, divide the large task into multiple small-scale and logically independent fine-grained subtasks (such as Figure 5 task 1, task 2, task n in it), and at the same time define clear input / output interfaces to ensure loose coupling between fine-grained subtasks; use the dynamic programming algorithm to construct a state transition equation, and describe how the execution state of the fine-grained subtasks changes with time and resources through the state transition equation, ensuring that the task decomposition scheme optimizes the overall scheduling strategy while satisfying the spatio-temporal dependence constraints and finding the optimal decomposition scheme under the spatio-temporal dependence constraints. This process should consider the balance between time complexity and space complexity; set specific time windows, resource limitations, and other necessary conditions for each fine-grained subtask to guide the interaction order and method between fine-grained subtasks, ensuring that the decomposed fine-grained subtasks can effectively cooperate under the specified conditions. The fine-grained decomposition of the drone subtasks refers to Figure 5 .

[0064] Step S30: Based on the multiple fine-grained subtasks, a fine-grained task execution plan that dynamically adapts to resource constraints and computing real-time in the edge environment is constructed, and a process-thread multi-granularity mixed edge-end fine-grained task parallel execution strategy is established.

[0065] Specifically, first, for the fine-grained subtasks after fine-grained decomposition, we first prioritize them according to factors such as the time urgency, data importance, and computational size of each fine-grained subtask, and then evaluate the resource requirements of each fine-grained subtask. Then, combined with the real-time performance indicators of the edge device, we build a time window planning model. Through the time window planning model, we reasonably arrange the execution order and start and end time of each fine-grained subtask to ensure that all fine-grained subtasks are completed within the specified time window, while avoiding resource contention and conflict. Based on the above steps, we build a fine-grained task execution plan that dynamically adapts to the resource constraints of the edge environment and the real-time requirements of computing. The execution plan needs to ensure the maximization of resource utilization efficiency and the efficient completion of tasks according to the preset time window.

[0066] Then, according to the characteristics of different types of fine-grained subtasks, a fine-grained task parallel execution model with multi-granularity hybrid of process and thread is established. Through the fine-grained task parallel execution model, hybrid-level parallel computing is implemented on multi-core heterogeneous edge devices to make full use of concurrency mechanisms at different granularity levels. For fine-grained subtasks with high computing intensity or large memory consumption, independent processes are created to avoid global resource competition. At the same time, synchronization mechanisms such as mutexes and semaphores between processes are used to ensure the correct and orderly execution of related operations. For lightweight fine-grained subtasks with high response speed requirements, multiple threads are used in a single process to quickly switch execution, reduce context switching overhead, and improve the real-time response capability of the system. Between the processes of each step, a directed acyclic graph is constructed using memory iteration. This method can avoid huge hard disk I / O overhead and additional calculations. In thread-level parallel management, the amount of tasks to be processed by the process to which the thread belongs is determined. If the amount of tasks built is greater than the set threshold, the drone anomaly detection and conflict warning are divided into multiple fine-grained subtasks. The calculation chain sequence is further segmented according to the complexity of the geometric object and entrusted to the sub-threads for processing, so as to make full use of idle computing resources with finer parallel granularity. The multi-granularity hybrid parallel execution strategy formed in this way can flexibly respond to various types of task requirements and can be dynamically adjusted according to the real-time environment. For example, when the edge device resources are sufficient, the number of processes is increased, and when resources are tight, more threads are used, thereby improving the overall computing performance. For example, Figure 6As shown, in each computing node (such as node 0, node 1, node 2), first, the spatio-temporal stream data is processed by the stream data processing process. After basic cleaning, addition, deletion, and screening of the data (data preprocessing), a large amount of data is divided into different threads (such as thread 0 - thread i) for synchronous processing. After being processed by the data division algorithm, subsequently, in the real-time supervision process, different computing tasks are assigned to different threads (such as thread 0 - thread i) to obtain the required computing data, realizing the synchronous execution of different algorithms, and finally fed back to the drone. The multi-granularity hybrid parallel execution strategy of the drone task for process - thread refers to Figure 6 .

[0067] Step S40: Conduct preliminary allocation of edge-side computing resources based on priorities and constraint conditions, and perform dynamic allocation and adaptive optimization of edge-side computing resources.

[0068] Specifically, first, through real-time monitoring and quantitative analysis of limited computing resources, combined with the characteristics and historical execution situations of spatio-temporal stream data processing tasks, machine learning algorithms are used to predict the resource requirements in a future period of time to accurately estimate the resource supply-demand relationship. According to factors such as the urgency, timeliness, and importance of tasks, optimize the priorities of each fine-grained subtask to be executed. According to the current total resources, the resource requirements and priorities of each task, adopt strategies such as fair scheduling, preemptive scheduling, and on-demand allocation to perform the initial computing resource allocation for each fine-grained subtask as needed, ensuring that critical tasks can obtain necessary resources in a timely manner. The flow chart of the edge-side computing resource allocation strategy refers to Figure 7 , which is divided into resource quantification of computing tasks (such as computing task A, computing task B, computing task C), resource requirement prediction, and dynamic allocation and optimization of edge-side computing resources.

[0069] Then, after the initial allocation, a real-time feedback mechanism for edge-side resource utilization is established. Once situations such as resource tension, load imbalance, or new tasks are detected, the resource dynamic reallocation process is immediately triggered. The allocated resources are adjusted through load balancing algorithms (such as round-robin scheduling algorithm, least-connections scheduling algorithm, etc.), such as releasing the resources occupied by completed tasks and reallocating idle resources. For key metrics such as processing latency, resource utilization rate, and data processing speed collected, relevant constraint functions (such as inclination willingness function, rate willingness function, etc.) are designed to balance the constraints on the distribution state of computing resources and the completion state of computing tasks within an edge cluster. On the premise of meeting the spatio-temporal dependencies of tasks, an elastic resource allocation system performance model is constructed to dynamically adjust the edge-side resource allocation, minimizing the degree of resource fragmentation and improving resource utilization rate. When the resources of a single edge device are insufficient to meet all task requirements, cloud-edge coordination can also be used later to expand to multi-edge collaboration, build a distributed resource pool, and further optimize the resource allocation effect through cross-device task migration and resource sharing.

[0070] Finally, based on the evaluation of the task completion time, resource utilization rate, and response speed distribution effect, the resource allocation strategy is continuously iteratively optimized. Combining the dynamic change trend of spatio-temporal flow data characteristics and the future task demand fluctuations, the change law of resource demand is predicted. According to the evaluation feedback information, the resource allocation strategy is automatically adjusted and optimized, forming a closed loop from monitoring, allocation, execution to feedback optimization, so that the edge-side computing resources can always achieve dynamic and efficient management and allocation under limited resource conditions.

[0071] The fine-grained task decomposition method based on the quantization modeling of UAV supervision tasks proposed by the present invention quantifies the spatio-temporal dependence characteristics of tasks according to the task execution order and the dependence relationship between data, establishes a spatio-temporal graph data model, and then provides model quantization support for subsequent spatio-temporal task fine-grained decomposition, achieving a reasonable balance between the execution time and space requirements of subtasks and improving the efficiency and timeliness of computing for a large number of UAV supervision tasks.

[0072] The prediction allocation method for edge-side task parallel computing resources proposed by the present invention, based on the parallel execution of the above-mentioned UAV fine-grained subtasks, uses deep learning methods to achieve the pre-prediction allocation of computing resources, ensuring the priority of key subtask execution. Combining with the dynamic allocation optimization strategy of computing resources, the resource allocation effect is further optimized, and the computing efficiency of UAV supervision tasks under limited computing resources is achieved.

[0073] Furthermore, as Figure 8As shown, based on the above-mentioned edge computing method for real-time supervision tasks of low-altitude UAVs, the present invention also correspondingly provides an edge computing system for real-time supervision tasks of low-altitude UAVs. Among them, the edge computing system for real-time supervision tasks of low-altitude UAVs includes:

[0074] A task evaluation module 51, configured to quantitatively express the spatio-temporal dependence characteristics of the low-altitude UAV supervision task, and quantitatively evaluate the low-altitude UAV supervision task according to the quantitative expression result to obtain a quantitative evaluation result;

[0075] A task decomposition module 52, configured to perform fine-grained decomposition of the spatio-temporal dependence tasks at the edge end according to the quantitative evaluation result, and decompose the low-altitude UAV supervision task into multiple fine-grained subtasks;

[0076] A task execution module 53, configured to construct a fine-grained task execution plan that dynamically adapts to resource constraints and computing real-time performance in the edge environment according to the multiple fine-grained subtasks, and establish a multi-grained parallel execution strategy for fine-grained tasks at the edge end with a process-thread multi-grained mixture;

[0077] A resource allocation module 54, configured to perform preliminary allocation of edge-end computing resources based on priorities and constraint conditions, and perform dynamic allocation and adaptive optimization of edge-end computing resources.

[0078] Furthermore, as Figure 9 shown, based on the above-mentioned edge computing method and system for real-time supervision tasks of low-altitude UAVs, the present invention also correspondingly provides a terminal, and the terminal includes a processor 10, a memory 20, and a display 30. Figure 9 Only some components of the terminal are shown, but it should be understood that it is not required to implement all the shown components, and more or fewer components can be alternatively implemented.

[0079] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as the hard disk or memory of the terminal. In some other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc., equipped on the terminal. Further, the memory 20 may also include both the internal storage unit and the external storage device of the terminal. The memory 20 is used to store the application software installed on the terminal and various types of data, such as the program code of the installed terminal, etc. The memory 20 may also be used to temporarily store the data that has been output or will be output. In one embodiment, an edge computing program 40 for the real-time supervision task of low-altitude unmanned aerial vehicles is stored on the memory 20, and the edge computing program 40 for the real-time supervision task of low-altitude unmanned aerial vehicles can be executed by the processor 10, so as to implement the edge computing method for the real-time supervision task of low-altitude unmanned aerial vehicles in the present application.

[0080] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other data processing chips, and is used to run the program code stored in the memory 20 or process data, such as executing the edge computing method for the real-time supervision task of low-altitude unmanned aerial vehicles, etc.

[0081] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode) toucher, etc. The display 30 is used to display the information in the terminal and to display a visual user interface. The processor 10, the memory 20 and the display 30 of the terminal communicate with each other through a system bus.

[0082] In one embodiment, when the processor 10 executes the edge computing program 40 for the real-time supervision task of low-altitude unmanned aerial vehicles in the memory 20, the steps of the edge computing method for the real-time supervision task of low-altitude unmanned aerial vehicles as described above are implemented.

[0083] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores an edge computing program for the real-time supervision task of low-altitude unmanned aerial vehicles, and when the edge computing program for the real-time supervision task of low-altitude unmanned aerial vehicles is executed by a processor, the steps of the edge computing method for the real-time supervision task of low-altitude unmanned aerial vehicles as described above are implemented.

[0084] In summary, the present invention provides an edge computing method, system, terminal, and computer-readable storage medium for real-time supervision tasks of low-altitude unmanned aerial vehicles. The method includes: quantitatively expressing the spatio-temporal dependence characteristics of the low-altitude unmanned aerial vehicle supervision tasks, and quantitatively evaluating the low-altitude unmanned aerial vehicle supervision tasks according to the quantitative expression results to obtain quantitative evaluation results; performing fine-grained decomposition of the edge-side spatio-temporal dependence tasks according to the quantitative evaluation results, and decomposing the low-altitude unmanned aerial vehicle supervision tasks into multiple fine-grained subtasks; constructing a fine-grained task execution plan that dynamically adapts to resource constraints and computing real-time performance in the edge environment according to the multiple fine-grained subtasks, and establishing a multi-granularity hybrid parallel execution strategy for fine-grained tasks at the edge side with processes and threads; initially allocating edge-side computing resources based on priorities and constraint conditions, and dynamically allocating and adaptively optimizing the edge-side computing resources. The present invention can significantly improve the real-time performance of edge computing for large-scale low-altitude unmanned aerial vehicle supervision tasks and improve the utilization rate of edge computing resources.

[0085] It should be noted that in this article, the terms "including", "comprising", or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or terminal including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such a process, method, article, or terminal. Without further limitations, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, article, or terminal including that element.

[0086] Of course, those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware (such as a processor, a controller, etc.) through a computer program. The program can be stored in a computer-readable storage medium that can be read by a computer. When the program is executed, it can include the processes of the above method embodiments. The computer-readable storage medium can be a memory, a magnetic disk, an optical disc, etc.

[0087] It should be understood that the application of the present invention is not limited to the above examples. For those of ordinary skill in the art, improvements or transformations can be made according to the above description. All such improvements and transformations should fall within the protection scope of the appended claims of the present invention.

Claims

1. An edge computing method for real-time monitoring of low-altitude UAVs, characterized in that: The edge computing method for the real-time supervision task of low-altitude drones includes: Quantitatively express the spatiotemporal dependency characteristics of the low-altitude UAV supervision task, and quantitatively evaluate the low-altitude UAV supervision task according to the quantitative expression result to obtain a quantitative evaluation result; Perform fine-grained decomposition of the edge-end spatiotemporal dependent tasks according to the quantitative evaluation results, and decompose the low-altitude UAV supervision task into multiple fine-grained subtasks; According to the multiple fine-grained subtasks, a fine-grained task execution plan that dynamically adapts to resource constraints and computing real-time in the edge environment is constructed, and a process-thread multi-granularity mixed edge-end fine-grained task parallel execution strategy is established; Perform preliminary allocation of edge computing resources based on priorities and constraints, and dynamically allocate and adaptively optimize edge computing resources; The temporal and spatial dependence characteristics of the low-altitude UAV supervision task are quantitatively expressed, and the low-altitude UAV supervision task is quantitatively evaluated according to the quantitative expression result to obtain a quantitative evaluation result, which specifically includes: Comprehensively analyze the timestamps, spatial location information, and dependencies between tasks in real-time computing tasks, and build a spatiotemporal graph model that expresses the spatiotemporal dependencies of tasks. The spatiotemporal graph model is used to build task dependencies, quantify task execution constraints, and dynamically adjust task execution strategies. Based on the changing characteristics of real-time computing tasks over time and space, the time series prediction method is used to model the task change trend, and combined with historical data analysis, the dynamic evolution law of time and space dependency characteristics is depicted; Based on the quantitative expression of spatiotemporal dependency features, computable patterns of different tasks are obtained. According to the computable patterns, the low-altitude UAV supervision task is quantitatively evaluated under the condition of limited resources at the edge to obtain quantitative evaluation results.

2. The edge computing method for real-time monitoring of low-altitude UAVs according to claim 1 is characterized in that: The quantitative evaluation of the low-altitude UAV supervision task under the limited resource conditions at the edge includes: Construct an objective function that reflects the requirements of spatiotemporal flow computing tasks and is suitable for optimization, and integrate resource consumption indicators into a unified computing workload evaluation model. The resource consumption indicators include CPU usage, memory usage, network transmission overhead, and energy consumption budget. Define the computational evaluation model: ； in, represents the computational workload evaluation model, represents the task load function, Indicates the number of drones, Indicates the total number of turning points in the running trajectory, representing the complexity of the trajectory, Function representing the time complexity calculation of the drone supervision algorithm.

3. The edge computing method for real-time monitoring of low-altitude UAVs according to claim 1 is characterized in that: The quantitative evaluation results include functional requirements, input and output requirements, performance indicators and key performance indicators.

4. The edge computing method for real-time monitoring of low-altitude UAVs according to claim 3 is characterized in that: The fine-grained decomposition of the edge-end spatiotemporal dependent tasks is performed according to the quantitative evaluation results, and the low-altitude UAV supervision task is decomposed into multiple fine-grained subtasks, specifically including: Draw detailed spatiotemporal dependency graphs based on functional requirements, input and output requirements, performance indicators, and key performance indicators, including data flow, control flow, and resource dependencies, to visually display the relationships between the various parts of the task; Select modeling tools based on task characteristics, intuitively represent task structure and temporal and spatial characteristics, and build a comprehensive state space for the task, including all task states and transition rules between task states; According to the functional logic of the task, the large task is subdivided into multiple small-scale and logically independent fine-grained subtasks, clear input and output interfaces are defined, and the state transfer equation is constructed using the dynamic programming algorithm to find the optimal decomposition solution that meets the time and space dependency constraints.

5. The edge computing method for real-time monitoring of low-altitude UAVs according to claim 1 is characterized in that: According to the plurality of the fine-grained subtasks, a fine-grained task execution plan that dynamically adapts to resource constraints and computing real-time in the edge environment is constructed, and a process-thread multi-granularity mixed edge-end fine-grained task parallel execution strategy is established, which specifically includes: For the fine-grained subtasks after fine-grained decomposition, the priorities are divided according to the time urgency, data importance, and computing amount of each fine-grained subtask, and the resource demand of each fine-grained subtask is evaluated. Combined with the real-time performance indicators of the edge device, a time window planning model is constructed. The execution order and start and end time of each fine-grained subtask are arranged through the time window planning model to ensure that all fine-grained subtasks are completed within the specified time window; According to the characteristics of different types of fine-grained subtasks, a process-thread multi-granularity hybrid fine-grained task parallel execution model is established, and hybrid-level parallel computing is implemented on multi-core heterogeneous edge devices through the fine-grained task parallel execution model.

6. The edge computing method for real-time monitoring of low-altitude UAVs according to claim 1 is characterized in that: The preliminary allocation of edge computing resources based on priorities and constraints, and dynamic allocation and adaptive optimization of edge computing resources, specifically include: Real-time monitoring and quantitative analysis of limited computing resources are carried out. Combined with the characteristics of spatiotemporal data processing tasks and historical execution status, machine learning algorithms are used to predict resource demand in the future and estimate the relationship between resource supply and demand. According to the urgency, timeliness and importance of the tasks, the priority of each fine-grained subtask to be executed is optimized. According to the current total amount of resources, the resource demand and priority of each task, fair scheduling, preemptive scheduling and on-demand allocation strategies are used to allocate initial computing resources to each fine-grained subtask on demand. After the initial allocation, a real-time feedback mechanism for edge resource utilization is established. When resource shortage, load imbalance or new tasks are detected, the dynamic resource reallocation process is triggered immediately. The allocated resources are adjusted through the load balancing algorithm. The relevant constraint functions are designed to balance the computing resource distribution status and computing task completion status within an edge cluster. Under the premise of satisfying the spatiotemporal dependency of tasks, an elastic resource allocation system performance model is constructed to dynamically adjust the edge resource allocation. Based on the evaluation of allocation effects based on task completion time, resource utilization and response speed, the resource allocation strategy is continuously iterated and optimized. In combination with the dynamic change trend of spatiotemporal flow data characteristics and future fluctuations in task requirements, the changing pattern of resource demand is predicted, and the resource allocation strategy is automatically adjusted and optimized based on the evaluation feedback information, forming a closed loop from monitoring, allocation, execution to feedback optimization.

7. An edge computing system for real-time monitoring of low-altitude drones, characterized in that: The edge computing system for the real-time supervision task of low-altitude UAVs is applied to the edge computing method for the real-time supervision task of low-altitude UAVs according to any one of claims 1 to 6, and the edge computing system for the real-time supervision task of low-altitude UAVs includes: A task evaluation module is used to quantitatively express the spatiotemporal dependency characteristics of the low-altitude UAV supervision task, and to quantitatively evaluate the low-altitude UAV supervision task according to the quantitative expression result to obtain a quantitative evaluation result; A task decomposition module, used to perform fine-grained decomposition of the edge-end spatiotemporal dependent tasks according to the quantitative evaluation results, and decompose the low-altitude UAV supervision task into multiple fine-grained subtasks; A task execution module is used to construct a fine-grained task execution plan that dynamically adapts to resource constraints and computing real-time in the edge environment based on multiple fine-grained subtasks, and establish a process-thread multi-granularity mixed edge-end fine-grained task parallel execution strategy; The resource allocation module is used to perform preliminary allocation of edge computing resources based on priorities and constraints, and to dynamically allocate and adaptively optimize edge computing resources.

8. A terminal, characterized in that: The terminal includes: a memory, a processor, and an edge computing program for low-altitude UAV real-time supervision tasks stored in the memory and run on the processor. When the edge computing program for low-altitude UAV real-time supervision tasks is executed by the processor, the steps of the edge computing method for low-altitude UAV real-time supervision tasks are implemented as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores an edge computing program for the low-altitude UAV real-time supervision task. When the edge computing program for the low-altitude UAV real-time supervision task is executed by the processor, the steps of the edge computing method for the low-altitude UAV real-time supervision task are implemented as described in any one of claims 1-6.

Images