Unmanned aerial vehicle edge network control method and system

Abstract

The application relates to a kind of unmanned aerial vehicle edge network control method and system, it is related to network control technical field, including the following steps, the state data of each edge node in the edge network of unmanned aerial vehicle group is collected, the computing task obtained by unmanned aerial vehicle group is classified based on the state data, and task priority list is obtained;Based on the task priority list, the matching degree of unmanned aerial vehicle group and edge node is calculated, and a task allocation scheme is obtained;Based on historical task execution data, resource consumption and time delay prediction are carried out on the task allocation scheme, and resource consumption information and time delay prediction result are obtained;Based on the resource consumption information and time delay prediction result, the task allocation scheme is dynamically adjusted, and a task allocation instruction is obtained;The technical problem of the performance of the uneven distribution of computing resources between different unmanned aerial vehicle nodes is solved.

Description

Technical Field

[0001] This invention relates to the field of network control technology, and in particular to a method and system for edge network control of unmanned aerial vehicles (UAVs). Background Technology

[0002] With the rapid development and widespread application of drone technology, drone swarm collaborative operations have become an important research direction. In practical applications, drone swarms often need to process large amounts of real-time data and complex computing tasks, posing a serious challenge to traditional centralized control methods. In traditional methods, all data needs to be transmitted to remote cloud servers for processing, which not only increases system response latency but also easily causes network congestion and bandwidth waste.

[0003] During drone swarm missions, the uneven distribution of computing resources among different drone nodes and the low efficiency of task processing are becoming increasingly prominent issues. Some drone nodes may be overloaded, while others are relatively idle. This improper utilization of resources seriously affects the performance of the entire system. Furthermore, due to the dynamic changes in the environment and the uncertainty of task requirements, how to achieve real-time task scheduling and flexible resource allocation has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to at least partially solve one of the technical problems existing in the prior art.

[0005] To achieve the above objectives, the present invention provides a method for controlling an unmanned aerial vehicle (UAV) via an edge network, comprising the following steps:

[0006] The status data of each edge node in the edge network of the drone swarm is collected, and the computing tasks acquired by the drone swarm are classified and divided according to the status data to obtain a task priority list.

[0007] Based on the task priority list, the matching degree of the drone swarm and edge nodes is calculated to obtain the task allocation scheme;

[0008] Based on historical task execution data, the resource requirements of the computing tasks allocated to each edge node in the task allocation scheme are cumulatively predicted to obtain the expected load value of the edge node. Based on the expected load value, the available resources of each edge node are evaluated to obtain the available resource curve. The difference between the available resource curve and the status data of the edge node is calculated to obtain the resource contention degree. Based on the resource contention degree, the execution time of each computing task is estimated to obtain the task execution table. The time data in the task execution table is weighted and analyzed with the resource contention degree to obtain the task completion index. Based on the task completion index, the performance of each edge node is evaluated to obtain resource consumption information and latency prediction results.

[0009] Based on the resource consumption information and latency prediction results, the task allocation scheme is dynamically adjusted to obtain task allocation instructions.

[0010] Furthermore, the collection of state data from each edge node in the edge network of the drone swarm includes:

[0011] The processor utilization, memory usage, and network bandwidth of each edge node of the drone swarm are periodically sampled to obtain node performance data;

[0012] The response time data of each edge node is measured based on the node performance data, and the response time data is weighted and averaged to obtain the status data.

[0013] Furthermore, the computational tasks acquired by the UAV based on the state data are hierarchically divided to obtain a task priority list, including:

[0014] The task header information of the computing tasks acquired by the UAV is parsed to obtain a task attribute table including task identifier, task type, task deadline and resource requirements. Based on the task type, a type weight is assigned to each computing task to obtain the basic weight value of the task.

[0015] The urgency of each computation task is calculated based on the difference between the task deadline and the current system time, and the urgency score of the task is obtained.

[0016] The task priority score is obtained by weighting and summing the task base weight value, the task urgency score and the available resource quantity of the node in the status data.

[0017] The task priority scores are sorted from high to low to obtain a task sorting sequence; based on a preset priority threshold, each calculation task in the task sorting sequence is classified into levels to obtain the task priority list.

[0018] Furthermore, based on the task priority list, a matching degree calculation is performed on the drone and edge nodes to obtain a task allocation scheme, including:

[0019] The resource requirements in the task priority list are compared with the available computing resources of the edge nodes to obtain resource matching degree data. Based on the resource matching degree data, the computing load of the edge nodes is evaluated to obtain a node load distribution map.

[0020] The node load distribution map is weighted and calculated with the physical distance from the UAV to the edge node to obtain a node affinity score. The edge nodes are then sorted based on the node affinity score to obtain a node priority queue. The node affinity score represents the degree of adaptation between the task and the computing resources of the edge node.

[0021] A one-to-one mapping is performed between the edge nodes in the node priority queue and the computational tasks in the task priority list to obtain an initial allocation table. Load balancing verification is then performed based on the initial allocation table to obtain the task allocation scheme.

[0022] Furthermore, the step of predicting resource consumption and latency of the task allocation scheme based on historical task execution data to obtain resource consumption information and latency prediction results includes:

[0023] Based on historical task execution data, the resource requirements of the computing tasks allocated to each edge node in the task allocation scheme are cumulatively predicted to obtain the expected load value of the edge node.

[0024] Based on the expected load value, the available resources of each edge node are evaluated to obtain the available resource curve;

[0025] The difference between the available resource curve and the state data of the edge node is calculated to obtain the resource contention degree, and the execution time of each computing task is estimated based on the resource contention degree to obtain the task execution table;

[0026] The time data in the task execution table is weighted and analyzed with the resource contention level to obtain the task completion index. Based on the task completion index, the performance of each edge node is evaluated to obtain the resource consumption information and latency prediction results.

[0027] Furthermore, based on the task completion metrics, performance evaluation is performed on each edge node to obtain the resource consumption information and latency prediction results, including:

[0028] The task completion index is compared with the processing capacity of the edge nodes to obtain the node load rate, and the resource usage of each edge node is calculated based on the node load rate.

[0029] The resource usage and the response time data of the edge nodes are combined to calculate the resource consumption information and latency prediction results.

[0030] Furthermore, based on the node load rate, the resource usage of each edge node is calculated, including:

[0031] Based on a preset load rate threshold table, the node load rate of each edge node is classified into different load rate zones. For edge nodes in the same load rate zone, resource usage frequency data over a historical period is collected and combined with the current resource usage to perform resource usage trend analysis, resulting in a resource usage trend map for each edge node.

[0032] Based on the resource usage trend chart, the resource usage of each edge node in different time periods is statistically analyzed in segments. The resource usage changes within adjacent time periods that are less than a preset value are merged to obtain the resource usage of each edge node.

[0033] Furthermore, based on the resource consumption information and latency prediction results, the task allocation scheme is dynamically adjusted to obtain task allocation instructions, including:

[0034] The resource usage in the resource consumption information and latency prediction results is sorted to obtain a resource usage sorting table, and the difference in resource usage between adjacent edge nodes in the resource usage sorting table is calculated to obtain a resource usage difference sequence.

[0035] Based on the resource usage difference sequence, the resource stress of each edge node is evaluated to obtain a resource stress evaluation value. When the resource stress evaluation value exceeds a preset stress threshold, the computing tasks in the task allocation scheme are reassigned based on the resource stress evaluation value to obtain a task reassignment priority table.

[0036] Based on the task redistribution priority table, the task allocation of each edge node is adjusted to obtain a task adjustment allocation table, and the task adjustment allocation table is converted into task allocation instructions.

[0037] Furthermore, based on the resource usage difference sequence, resource stress is assessed for each edge node to obtain a resource stress assessment value, including:

[0038] The interval distribution statistics of the difference data between adjacent edge nodes in the resource usage difference sequence are performed to obtain the node resource distribution curve;

[0039] Based on the resource distribution curve, the resource occupancy status of each edge node is calculated in segments to obtain a resource occupancy distribution map. Then, fluctuation analysis is performed on the data of adjacent time periods in the resource occupancy distribution map to obtain a load fluctuation curve.

[0040] The load fluctuation curve is compared with a preset load threshold to obtain the load over-limit range, and the stress of each edge node is quantified based on the load over-limit range to obtain the resource stress assessment value.

[0041] The present invention also provides an edge network control system for unmanned aerial vehicles, comprising:

[0042] The acquisition module is used to collect the status data of each edge node in the edge network of the UAV swarm, and to classify the computing tasks acquired by the UAV swarm according to the status data to obtain a task priority list.

[0043] The calculation module is used to calculate the matching degree between the drone swarm and edge nodes based on the task priority list to obtain a task allocation scheme;

[0044] The prediction module is used to perform cumulative prediction of resource requirements for computing tasks allocated to each edge node in the task allocation scheme based on historical task execution data, to obtain the expected load value of the edge node, and to evaluate the available resources of each edge node based on the expected load value, to obtain an available resource curve. The module performs a difference calculation between the available resource curve and the status data of the edge node to obtain the resource contention degree, and to estimate the execution time of each computing task based on the resource contention degree, to obtain a task execution table. The module performs a weighted analysis of the time data in the task execution table and the resource contention degree to obtain a task completion index, and to evaluate the performance of each edge node based on the task completion index, to obtain resource consumption information and latency prediction results.

[0045] The adjustment module is used to dynamically adjust the task allocation scheme based on the resource consumption information and latency prediction results to obtain task allocation instructions.

[0046] This invention provides a method for controlling an edge network of unmanned aerial vehicles (UAVs), comprising the following steps: collecting state data of each edge node in the edge network of a UAV swarm; classifying the computing tasks acquired by the UAV swarm based on the state data to obtain a task priority list; calculating the matching degree between the UAV swarm and edge nodes based on the task priority list to obtain a task allocation scheme; predicting resource consumption and latency based on historical task execution data to obtain resource consumption information and latency prediction results; and dynamically adjusting the task allocation scheme based on the resource consumption information and latency prediction results to obtain a task allocation instruction. The task allocation instruction guides the edge nodes to execute computing tasks according to the adjusted allocation scheme, thereby achieving balanced scheduling of edge node computing resources and efficient collaborative processing of tasks, solving the technical problem of uneven distribution of computing resources among different UAV nodes. This method achieves balanced scheduling of edge node computing resources, effectively avoiding situations where some nodes are overloaded while others are idle. This balanced resource utilization not only improves the overall system performance but also extends the service life of the equipment. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a schematic diagram of the steps of an UAV edge network control method in one embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of an edge network control system for a drone according to an embodiment of the present invention;

[0050] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0051] The embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The step numbers in the following embodiments are set only for ease of explanation, and there is no limitation on the order between the steps. The execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0052] The following describes in detail, with reference to the accompanying drawings, a method for controlling an unmanned aerial vehicle (UAV) edge network according to an embodiment of the present invention.

[0053] Figure 1 This invention provides an edge network control method for unmanned aerial vehicles (UAVs) according to one embodiment, comprising the following steps:

[0054] Step S1: Collect the status data of each edge node in the edge network of the UAV swarm, and classify the computing tasks acquired by the UAV swarm according to the status data to obtain a task priority list.

[0055] Specifically, when the drone swarm is performing a mission, the system first activates the edge node status data acquisition module. This module periodically acquires key indicators such as CPU utilization, memory usage, and network bandwidth of each node. After collection, the system establishes a node health scoring matrix based on this real-time status data, which includes quantitative indicators in three dimensions: processing capacity index, load level, and network connection quality. For example, when the CPU utilization of an edge node exceeds 85%, its processing capacity index will decrease accordingly, which directly affects subsequent task allocation decisions.

[0056] After acquiring a computational task, the system analyzes its urgency, resource requirements, and computational complexity, calculating task weights using predefined scoring rules. Specifically, urgent tasks are assigned higher weights; for example, a real-time target tracking task might have twice the weight of a regular data processing task. The system then sorts the tasks by weight, creating a dynamically updated priority list.

[0057] To ensure the accuracy of task classification, the system also references historical execution records and fine-tunes the priority of current tasks based on the completion status of similar tasks in the past. For example, if historical data shows that a certain type of task frequently experiences delays, the system will appropriately increase the priority of that type of task to allow more processing time. This dynamic classification mechanism based on multi-dimensional data can effectively balance system resource utilization and task processing efficiency.

[0058] Step S2: Calculate the matching degree between the drone swarm and edge nodes based on the task priority list to obtain a task allocation scheme.

[0059] Specifically, after obtaining the task priority list, the system constructs a matching matrix based on the real-time status characteristics of edge nodes and the task requirement characteristics. The matching score is calculated using a weighted scoring method, primarily considering hardware metrics such as processor speed, memory capacity, and network bandwidth, while also taking into account the current load status of the nodes. For example, if a computationally intensive task requires a large amount of CPU resources, nodes with more idle CPU resources will receive a higher matching score.

[0060] In the implementation, each edge node maintains a task queue, recording the number of tasks currently being processed and those waiting to be processed. The system calculates the remaining computing power of each node and, combined with predictions of task resource requirements, identifies the most suitable task-node pairing. For example, when an image processing task arrives, the system prioritizes nodes with GPU acceleration capabilities and currently lighter loads.

[0061] After the matching degree calculation is completed, the system will generate a preliminary task allocation plan. This plan is not static, but will be continuously updated as new tasks are added and node status changes. To avoid overloading certain nodes, the system also sets a load balancing threshold. Once the load of a node exceeds the threshold, the task redistribution mechanism will be triggered.

[0062] Step S3: Based on historical task execution data, perform cumulative prediction of resource requirements for the computing tasks allocated to each edge node in the task allocation scheme to obtain the expected load value of the edge node. Based on the expected load value, evaluate the available resources of each edge node to obtain an available resource curve. Perform difference calculation between the available resource curve and the status data of the edge node to obtain the resource contention degree. Based on the resource contention degree, estimate the execution time of each computing task to obtain a task execution table. Perform weighted analysis on the time data in the task execution table and the resource contention degree to obtain a task completion index. Based on the task completion index, evaluate the performance of each edge node to obtain resource consumption information and latency prediction results.

[0063] Specifically, the system first reads historical task execution records and extracts actual resource consumption information for various tasks. For multiple tasks assigned to the same edge node, the system calculates their total resource requirements. For example, if a node is assigned three image processing tasks, each consuming an average of 2GB of memory, then the expected total memory load is 6GB. This cumulative prediction takes into account the time overlap of tasks; if the task execution times are staggered, the actual peak load may be lower than the simple cumulative value.

[0064] Based on the expected load, the system plots the trend of available resources for each edge node over a future period. This curve reflects the remaining resources of the node at different points in time. For example, if a node has a total memory of 16GB, currently uses 8GB, and new tasks are expected to consume 6GB, then the available resource curve will show 2GB of available memory remaining. This dynamic prediction helps the system identify potential resource bottlenecks in a timely manner.

[0065] The system compares the available resource curve with the current node status data to calculate the resource contention level. If the expected available resources for a certain period are significantly less than the current available resources, it indicates intense resource contention during that period. For example, if the current CPU idle rate is 50%, but the prediction curve shows only 10% idle time for a certain period, then the CPU resource contention level for that period is 0.8. This metric directly affects the estimation of task execution time; the higher the contention level, the longer the expected execution time.

[0066] Finally, the system evaluates task completion. The task execution table records the estimated start time, runtime, and completion deadline for each task. The system analyzes this time data along with resource contention levels to derive task completion metrics. For example, if a task is expected to take 20 minutes to complete, but due to a resource contention level as high as 0.8, the actual execution time may extend to 30 minutes. The system generates a detailed performance evaluation report for each edge node, including key metrics such as resource utilization efficiency and task latency risk.

[0067] This predictive mechanism continuously learns and optimizes from actual operational data, and the accuracy of the prediction results improves with the accumulation of historical data. The system also periodically evaluates prediction biases and adjusts the parameters of the prediction model in a timely manner to ensure the scientific nature of resource scheduling decisions.

[0068] Step S4: Based on the resource consumption information and latency prediction results, dynamically adjust the task allocation scheme to obtain task allocation instructions.

[0069] Specifically, once the system obtains resource consumption information and latency prediction results, it will initiate a dynamic adjustment mechanism for the task allocation scheme. During the adjustment process, resource utilization thresholds are first set, such as CPU utilization not exceeding 80% and memory usage not exceeding 75%. If the prediction results show that a node is about to exceed these thresholds, the system will trigger task migration, reallocating some tasks to nodes with lighter loads.

[0070] The specific adjustment strategy employs a sliding time window method. The system checks the actual load of each node against the predicted data at regular intervals. When it detects a significant deviation between the actual resource consumption of a node and the predicted value—for example, if the actual CPU utilization exceeds the predicted value by more than 15%—an emergency adjustment mechanism is immediately activated. The adjusted solution generates detailed task allocation instructions, including task ID, target node, execution priority, and other information.

[0071] Task allocation instructions are sent to each edge node via a message queue. Upon receiving the instructions, the nodes process the tasks according to their priority. To ensure a smooth task migration, the system also implements a task state saving and recovery mechanism. For example, when a video processing task needs to migrate from node A to node B, the system first saves the current processing progress to ensure that execution can resume from the breakpoint after the migration. This dynamic adjustment mechanism greatly improves the resource utilization efficiency and task processing speed of the entire system. The task allocation instructions guide the edge nodes to execute computing tasks according to the adjusted allocation scheme, thereby achieving balanced scheduling of edge node computing resources and efficient collaborative task processing.

[0072] In a specific embodiment, the collection of state data of each edge node in the edge network of the drone swarm includes:

[0073] The processor utilization, memory usage, and network bandwidth of each edge node of the drone swarm are periodically sampled to obtain node performance data;

[0074] The response time data of each edge node is measured based on the node performance data, and the response time data is weighted and averaged to obtain the status data.

[0075] Specifically, during the edge node status data acquisition phase, the system initiates a monitoring process, triggering timers to sample the performance metrics of each node. Processor utilization sampling is typically performed every 500 milliseconds, obtaining the usage of each CPU core by reading system performance counters and calculating the average value. Memory usage monitoring includes the usage of physical memory and virtual memory; the system records the specific values ​​of available memory, used memory, and cache memory. For network bandwidth, the system tracks uplink and downlink data transmission rates, as well as the packet throughput of network interfaces.

[0076] The raw data obtained from sampling needs to be preprocessed before it can be used for subsequent analysis. For example, CPU utilization data will have abnormal fluctuations removed, and a smoothed value from multiple consecutive sampling periods will be used. Memory usage data needs to consider the degree of memory fragmentation, because excessive fragmentation will affect the actual available memory space. Network bandwidth data needs to be combined with current network quality indicators; if the packet loss rate is found to be too high, the bandwidth data needs to be corrected.

[0077] Response time measurement employs probe technology. The system periodically sends probe packets to each node, recording the time interval between sending and receiving a response. To improve measurement accuracy, each node receives multiple probe requests with different priorities. For example, the system might send both normal-priority and high-priority probe packets simultaneously, comparing their response times to assess the node's load status.

[0078] When calculating the weighted average response time, the system assigns different weights to probe packets based on their priority and network conditions. Generally, higher-priority probe packets receive a larger weight in their response time because they better reflect the actual processing capacity of a node. The system also considers network congestion; when network quality is poor, the weight of the response time data for that period will be appropriately reduced.

[0079] The final status data is a comprehensive indicator that reflects not only the current performance of a node but also its trend over a certain period. This data is used to assess the health of nodes and serves as an important basis for task allocation. For example, if a node's status data shows significant fluctuations in response time, even if its current performance indicators are good, the system will lower the priority of assigning critical tasks to that node.

[0080] In a specific embodiment, the step of classifying the computational tasks acquired by the UAV based on the state data to obtain a task priority list includes:

[0081] The task header information of the computing tasks acquired by the UAV is parsed to obtain a task attribute table including task identifier, task type, task deadline and resource requirements. Based on the task type, a type weight is assigned to each computing task to obtain the basic weight value of the task.

[0082] The urgency of each computation task is calculated based on the difference between the task deadline and the current system time, and the urgency score of the task is obtained.

[0083] The task priority score is obtained by weighting and summing the task base weight value, the task urgency score and the available resource quantity of the node in the status data.

[0084] The task priority scores are sorted from high to low to obtain a task sorting sequence; based on a preset priority threshold, each calculation task in the task sorting sequence is classified into levels to obtain the task priority list.

[0085] Specifically, when a drone acquires a new computing task, the system first reads the key fields in the task header information. The task identifier is usually a unique code used to track the task status throughout the processing. Task types may include different categories such as image processing, path planning, and target tracking, each with a preset base weight value. For example, the weight value for a real-time target tracking task might be 1.5, while the weight value for a regular data storage task might be 0.8.

[0086] After parsing the task deadline, the system calculates the difference between the deadline and the current system time to assess the task's urgency. A logarithmic function is typically used to calculate the urgency score, which avoids distortion when the time difference is too large. For example, a task that needs to be completed within 10 minutes might have an urgency score of 0.9, while a task allowed to be completed within one hour might have an urgency score as low as 0.5. Resource requirements are directly extracted from the task attribute table, including estimated CPU time, memory space, and other specific values.

[0087] When calculating the overall priority, the system combines the previously obtained indicators. First, the task's base weight value is multiplied by a weighting coefficient α (usually 0.4). Then, the urgency score is multiplied by a coefficient β (usually 0.4). Finally, the normalized value of the node's available resources is multiplied by a coefficient γ (usually 0.2). The weighted sum of these three factors constitutes the final task priority score.

[0088] After obtaining the priority scores of all tasks, the system uses the quicksort algorithm to sort them in descending order. After sorting, the tasks are categorized according to preset priority thresholds. Typically, three thresholds are set: 0.8, 0.6, and 0.4, dividing tasks into four levels. Tasks with a priority score greater than 0.8 are marked as highest priority and require priority allocation of computing resources; tasks with a priority score between 0.6 and 0.8 are high priority; those between 0.4 and 0.6 are medium priority; and those less than 0.4 are low priority.

[0089] In a specific embodiment, a matching degree calculation is performed on the drone and edge nodes based on the task priority list to obtain a task allocation scheme, including:

[0090] The resource requirements in the task priority list are compared with the available computing resources of the edge nodes to obtain resource matching degree data. Based on the resource matching degree data, the computing load of the edge nodes is evaluated to obtain a node load distribution map.

[0091] The node load distribution map is weighted and calculated with the physical distance from the UAV to the edge node to obtain a node affinity score. The edge nodes are then sorted based on the node affinity score to obtain a node priority queue. The node affinity score represents the degree of adaptation between the task and the computing resources of the edge node.

[0092] A one-to-one mapping is performed between the edge nodes in the node priority queue and the computational tasks in the task priority list to obtain an initial allocation table. Load balancing verification is then performed based on the initial allocation table to obtain the task allocation scheme.

[0093] Specifically, the system first calculates the difference between the resource requirements of the task and the available resources of the edge node. This process involves matching calculations across three dimensions: CPU computing power, memory capacity, and network bandwidth. For example, if an image processing task requires 2GB of memory, and an edge node currently has 3GB of available memory, then the matching degree for the memory dimension is 1GB. The system assigns different weights to the matching results of each dimension, such as 0.5 for CPU, 0.3 for memory, and 0.2 for bandwidth, to comprehensively derive the resource matching degree data.

[0094] When assessing node load, the system considers the number of tasks currently being executed and the status of queued tasks. By normalizing the real-time load data for each node, an intuitive load distribution chart is generated. For example, if a node currently has a CPU load of 60% and a memory utilization of 75%, its overall load index might be 0.68. This data is updated in real-time as task execution progresses, ensuring the accuracy of load assessment.

[0095] Next, the system calculates the physical distance from the drone to each edge node. Here, GPS coordinates are used to calculate the Euclidean distance, and the distance values ​​are converted into normalized values ​​between 0 and 1. The closer the distance, the closer the normalized value is to 1. The system then performs a weighted sum of the node load value and the distance normalized value to obtain the node affinity score. Typically, the load weight is 0.7, and the distance weight is 0.3. For example, if a node has a load value of 0.68 and a distance normalized value of 0.85, its affinity score is 0.68 × 0.7 + 0.85 × 0.3 = 0.731.

[0096] The system sorts nodes in descending order based on affinity scores, forming a node priority queue. Higher-priority nodes receive task assignment opportunities first. When mapping tasks to nodes, the system follows the principle of "matching high-priority tasks with high-priority nodes." The resulting initial allocation table undergoes load balancing verification to ensure no nodes experience severe overload.

[0097] During the verification process, the system sets load thresholds, such as a maximum load of no more than 85% for a single node. If the expected load of a node exceeds the threshold, a task redistribution mechanism is triggered. Redistribution selects nodes with lighter loads, taking into account task migration costs. The final task allocation scheme aims to ensure both resource utilization efficiency and overall system stability. This scheme is periodically evaluated and updated to adapt to dynamic changes in the network environment.

[0098] In a specific embodiment, performance evaluation of each edge node is performed based on the task completion metrics to obtain the resource consumption information and latency prediction results, including:

[0099] The task completion index is compared with the processing capacity of the edge nodes to obtain the node load rate, and the resource usage of each edge node is calculated based on the node load rate.

[0100] The resource usage and the response time data of the edge nodes are combined to calculate the resource consumption information and latency prediction results.

[0101] Specifically, the system divides the task completion metric by the processing capacity of the edge node to obtain the actual node load rate. For example, if a node has a processing capacity of 1000 operations per second, but the current task completion metric shows that it can only achieve 700 operations per second, then the node's load rate is 70%. This ratio reflects the actual utilization efficiency of node resources and helps to identify performance bottlenecks.

[0102] When tracking resource usage, the system records specific metrics such as CPU time slice usage, peak memory usage, and network bandwidth consumption. For example, when a node is performing an image processing task, its average CPU utilization remains at 75%, memory usage reaches 12GB, and network bandwidth consumption is approximately 50Mbps. This data is used to evaluate the node's resource utilization, helping the system make more accurate task allocation decisions.

[0103] The system performs correlation analysis between the collected response time data and resource usage. A weighted average method is typically used, with response time weighted at 0.6 and resource usage weighted at 0.4. If a node's average response time is 100 milliseconds and its resource usage is at a moderate level, its overall score might be around 0.7. This score directly reflects the node's overall performance.

[0104] When generating the final prediction, the system considers the trends in historical data. For example, if it finds that the response time of a node is gradually increasing, even if the current load rate is not high, the system will make a relatively conservative prediction of its future performance. This prediction includes the expected value and possible fluctuation range of resource consumption information, as well as the estimated latency of task execution.

[0105] The accuracy of the prediction results directly impacts subsequent task scheduling decisions. The system periodically compares the prediction results with actual operational data, calculates the prediction error, and adjusts the parameters of the prediction model accordingly. For example, if the actual execution time of a certain type of task frequently exceeds the predicted value by more than 20%, the system will correspondingly increase the resource reservation for that type of task.

[0106] In a specific embodiment, the resource usage of each edge node is calculated based on the node load rate, including:

[0107] Based on a preset load rate threshold table, the node load rate of each edge node is classified into different load rate zones. For edge nodes in the same load rate zone, resource usage frequency data over a historical period is collected and combined with the current resource usage to perform resource usage trend analysis, resulting in a resource usage trend map for each edge node.

[0108] Based on the resource usage trend chart, the resource usage of each edge node in different time periods is statistically analyzed in segments. The resource usage changes within adjacent time periods that are less than a preset value are merged to obtain the resource usage of each edge node.

[0109] Specifically, the system sets multiple load rate thresholds to classify load levels, typically including four zones: light load (0-30%), medium load (30-60%), heavy load (60-85%), and overload (>85%). Each edge node is assigned to the corresponding zone based on its actual load rate. For example, a node with a load rate of 45% would be classified as a medium load node. This classification method facilitates the system's rapid identification of nodes requiring focused attention.

[0110] For nodes of the same load level, the system queries resource usage records for the past 24 hours. This historical data includes sampled values ​​of metrics such as CPU utilization, memory usage, and network traffic. For example, a medium-load node might record resource usage every 5 minutes over the past 24 hours, resulting in 288 data points. The system combines this historical data with current sampled values ​​and uses time series analysis to create a resource usage trend chart.

[0111] Trend analysis employs a sliding window method, with the window size typically set to one hour. Within each window, the system calculates the average and standard deviation of resource usage and represents the usage trend using a line graph. For example, if a node's CPU utilization remains stable at around 60% between 8 AM and 10 AM, but suddenly surges to 75% after noon, this trend will be clearly reflected on the trend graph.

[0112] To simplify the statistical results, the system segments resource usage. If the change in resource usage between two adjacent time periods does not exceed a preset threshold (usually 5%), these two time periods are merged. For example, if a node's average CPU usage is 62% from 9:00 to 9:30 and 63% from 9:30 to 10:00, since the difference is less than 5%, this hour will be merged into one statistical interval, recording an average usage of 62.5%.

[0113] This merging process effectively reduces data redundancy and highlights important nodes with changes in usage. For example, if a node's resource usage is relatively stable during working hours (9:00-17:00), it may be merged into 2-3 statistical intervals, while during periods of drastic load changes (such as peak task periods), more detailed segmented data will be retained.

[0114] The final resource usage statistics retain key load change information while avoiding overly trivial data recording.

[0115] In a specific embodiment, the task allocation scheme is dynamically adjusted based on the resource consumption information and latency prediction results to obtain task allocation instructions, including:

[0116] The resource usage in the resource consumption information and latency prediction results is sorted to obtain a resource usage sorting table, and the difference in resource usage between adjacent edge nodes in the resource usage sorting table is calculated to obtain a resource usage difference sequence.

[0117] Based on the resource usage difference sequence, the resource stress of each edge node is evaluated to obtain a resource stress evaluation value. When the resource stress evaluation value exceeds a preset stress threshold, the computing tasks in the task allocation scheme are reassigned based on the resource stress evaluation value to obtain a task reassignment priority table.

[0118] Based on the task redistribution priority table, the task allocation of each edge node is adjusted to obtain a task adjustment allocation table, and the task adjustment allocation table is converted into task allocation instructions.

[0119] Specifically, the resource usage of each edge node is first sorted from highest to lowest to form a resource usage ranking table. For example, if node A has a usage rate of 85%, node B has a usage rate of 70%, and node C has a usage rate of 40%, the sorted values ​​are A, B, and C. Then, the resource usage difference between adjacent nodes is calculated. For example, the difference between A and B is 15%, and the difference between B and C is 30%. These differences constitute a resource usage difference sequence.

[0120] The system assesses resource scarcity based on a sequence of resource usage differences. If the resource usage difference between adjacent nodes is too large, it indicates uneven resource allocation. For example, when the difference exceeds 25%, the system will increase the resource scarcity assessment value. Assuming a scarcity threshold of 0.7, when a node's assessment value reaches 0.8, a task reallocation mechanism will be triggered.

[0121] During task reassignment, the system recalculates the priority of each task. For example, an image processing task that was originally running on a high-load node A might be downgraded, while some critical tasks retain their original priorities. This adjustment creates a new task reassignment priority table, which contains the new priority value for each task and a suggested target node.

[0122] Based on the redistribution priority table, the system begins to adjust task allocation. Typically, this involves migrating some tasks from high-load nodes to low-load nodes. For example, two non-critical tasks on node A might be migrated to node C, thus balancing the load across nodes. This process generates a detailed task reassignment table, recording the migration path and timing for each task.

[0123] Finally, the system converts the task adjustment and allocation table into specific execution instructions. These instructions contain key information such as task ID, target node, and migration time. For example, an instruction might require "task T1 to migrate from node A to node C in the next scheduling cycle." To ensure a smooth migration process, the system sets the priority order and time window for task migrations.

[0124] In a specific embodiment, resource stress is assessed for each edge node based on the resource usage difference sequence to obtain a resource stress assessment value, including:

[0125] The interval distribution statistics of the difference data between adjacent edge nodes in the resource usage difference sequence are performed to obtain the node resource distribution curve;

[0126] Based on the resource distribution curve, the resource occupancy status of each edge node is calculated in segments to obtain a resource occupancy distribution map. Then, fluctuation analysis is performed on the data of adjacent time periods in the resource occupancy distribution map to obtain a load fluctuation curve.

[0127] The load fluctuation curve is compared with a preset load threshold to obtain the load over-limit range, and the stress of each edge node is quantified based on the load over-limit range to obtain the resource stress assessment value.

[0128] Specifically, the system processes resource data using a difference sequence, dividing the difference data into multiple intervals based on their magnitude. For example, it uses 0-10%, 10-20%, and 20-30% as statistical intervals, counting the number of nodes within each interval. The node resource distribution curve obtained through this statistical method can intuitively show the degree of balance in the system's resource allocation. For instance, if most differences are concentrated in the 0-10% interval, it indicates that the system's resource allocation is relatively balanced.

[0129] Next, we analyze the resource usage of each node at different times. The system divides the timeline into multiple observation windows, such as one window every 30 minutes, and records the resource usage within each window. This produces a resource usage distribution chart that changes over time. For example, if a node's resource usage is 65% from 9:00 to 9:30 and rises to 75% from 9:30 to 10:00, these changes will be recorded in detail.

[0130] By comparing data from adjacent time windows, the system generates a load fluctuation curve. This curve reflects the dynamic changes in resource usage. For example, if the load on a node jumps from 50% to 80% in a short period, this drastic fluctuation will create a noticeable peak on the curve. The system pays particular attention to these sudden load changes because they may indicate potential performance problems.

[0131] The system pre-sets multiple load thresholds, such as a warning threshold of 75% and a danger threshold of 85%. By comparing the load fluctuation curve with these threshold lines, the overload range can be identified. For example, if the load of a node consistently exceeds 85% between 10:00 and 11:00, this period will be marked as an overload range. The system will then track the overload duration and severity for each node.

[0132] Finally, the system calculates the resource stress assessment value based on the number, duration, and degree of exceeding limits. The calculation formula considers multiple factors: the percentage of exceeding limits, the degree of exceeding limits, and the fluctuation frequency, all with a weight of 0.4. For example, if a node is in an exceeding-limit state 30% of the time, with an average exceeding-limit degree of 10%, and experiences frequent load fluctuations, its stress assessment value may reach 0.75.

[0133] The above describes a UAV edge network control method according to an embodiment of the present invention. The following describes a UAV edge network control system according to an embodiment of the present invention. Please refer to [link / reference]. Figure 2 One embodiment of the UAV edge network control system of the present invention includes:

[0134] The acquisition module 21 is used to acquire the status data of each edge node in the edge network of the UAV swarm, and to classify the computing tasks acquired by the UAV swarm according to the status data to obtain a task priority list.

[0135] Calculation module 22 is used to calculate the matching degree between the UAV swarm and edge nodes based on the task priority list to obtain a task allocation scheme;

[0136] The prediction module 23 is used to perform cumulative prediction of resource requirements for computing tasks allocated to each edge node in the task allocation scheme based on historical task execution data, to obtain the expected load value of the edge node, and to evaluate the available resources of each edge node based on the expected load value, to obtain an available resource curve, to perform a difference operation between the available resource curve and the status data of the edge node, to obtain the resource contention degree, and to estimate the execution time of each computing task based on the resource contention degree, to obtain a task execution table, to perform a weighted analysis of the time data in the task execution table and the resource contention degree, to obtain a task completion index, and to perform a performance evaluation of each edge node based on the task completion index, to obtain resource consumption information and latency prediction results.

[0137] The adjustment module 24 is used to dynamically adjust the task allocation scheme based on the resource consumption information and latency prediction results to obtain task allocation instructions.

[0138] In this embodiment, the specific implementation of each module in the above system embodiment is described in the above method embodiment, and will not be repeated here.

Claims

1. A method for controlling an unmanned aerial vehicle (UAV) via an edge network, characterized in that, Includes the following steps: The status data of each edge node in the edge network of the drone swarm is collected, and the computing tasks acquired by the drone swarm are classified and divided according to the status data to obtain a task priority list. Based on the task priority list, the matching degree of the drone swarm and edge nodes is calculated to obtain the task allocation scheme; Based on historical task execution data, the resource requirements of the computing tasks allocated to each edge node in the task allocation scheme are cumulatively predicted to obtain the expected load value of the edge node. Based on the expected load value, the available resources of each edge node are evaluated to obtain the available resource curve. The difference between the available resource curve and the status data of the edge node is calculated to obtain the resource contention degree. Based on the resource contention degree, the execution time of each computing task is estimated to obtain the task execution table. The time data in the task execution table is weighted and analyzed with the resource contention degree to obtain the task completion index. Based on the task completion index, the performance of each edge node is evaluated to obtain resource consumption information and latency prediction results. Based on the resource consumption information and latency prediction results, the task allocation scheme is dynamically adjusted to obtain task allocation instructions. Based on the task priority list, the matching degree between the drone and the edge node is calculated to obtain a task allocation scheme, including: The resource requirements in the task priority list are compared with the available computing resources of the edge nodes to obtain resource matching degree data. Based on the resource matching degree data, the computing load of the edge nodes is evaluated to obtain a node load distribution map. The node load distribution map is weighted and calculated with the physical distance from the UAV to the edge node to obtain a node affinity score. The edge nodes are then sorted based on the node affinity score to obtain a node priority queue. The node affinity score represents the degree of adaptation between the task and the computing resources of the edge node. A one-to-one mapping is performed between the edge nodes in the node priority queue and the computation tasks in the task priority list to obtain an initial allocation table. Load balancing verification is then performed based on the initial allocation table to obtain the task allocation scheme. Based on the resource consumption information and latency prediction results, the task allocation scheme is dynamically adjusted to obtain task allocation instructions, including: The resource usage in the resource consumption information and latency prediction results is sorted to obtain a resource usage sorting table, and the difference in resource usage between adjacent edge nodes in the resource usage sorting table is calculated to obtain a resource usage difference sequence. Based on the resource usage difference sequence, the resource stress of each edge node is evaluated to obtain a resource stress evaluation value. When the resource stress evaluation value exceeds a preset stress threshold, the computing tasks in the task allocation scheme are reassigned based on the resource stress evaluation value to obtain a task reassignment priority table. Based on the task redistribution priority table, the task allocation of each edge node is adjusted to obtain a task adjustment allocation table, and the task adjustment allocation table is converted into task allocation instructions. Based on the resource usage difference sequence, resource stress is assessed for each edge node to obtain resource stress assessment values, including: The interval distribution statistics of the difference data between adjacent edge nodes in the resource usage difference sequence are performed to obtain the node resource distribution curve; Based on the resource distribution curve, the resource occupancy status of each edge node is calculated in segments to obtain a resource occupancy distribution map. Then, fluctuation analysis is performed on the data of adjacent time periods in the resource occupancy distribution map to obtain a load fluctuation curve. The load fluctuation curve is compared with a preset load threshold to obtain the load over-limit range, and the stress of each edge node is quantified based on the load over-limit range to obtain the resource stress assessment value.

2. The UAV edge network control method according to claim 1, characterized in that, The status data of each edge node in the edge network of the drone swarm includes: The processor utilization, memory usage, and network bandwidth of each edge node in the edge network of the drone swarm are periodically sampled to obtain node performance data; The response time data of each edge node is measured based on the node performance data, and the response time data is weighted and averaged to obtain the status data.

3. The UAV edge network control method according to claim 1, characterized in that, The computational tasks acquired by the UAV based on the state data are classified into levels to obtain a task priority list, including: The task header information of the computing tasks acquired by the UAV is parsed to obtain a task attribute table including task identifier, task type, task deadline and resource requirements. Based on the task type, a type weight is assigned to each computing task to obtain the basic weight value of the task. The urgency of each computation task is calculated based on the difference between the task deadline and the current system time, and the urgency score of the task is obtained. The task priority score is obtained by weighting and summing the task base weight value, the task urgency score and the available resource quantity of the node in the status data. The task priority scores are sorted from high to low to obtain a task sorting sequence; based on a preset priority threshold, each calculation task in the task sorting sequence is classified into levels to obtain the task priority list.

4. The UAV edge network control method according to claim 1, characterized in that, Based on the task completion metrics, the performance of each edge node is evaluated to obtain the resource consumption information and latency prediction results, including: The task completion index is compared with the processing capacity of each edge node to obtain the node load rate, and the resource usage of each edge node is calculated based on the node load rate. The resource usage and the response time data of the edge nodes are combined to calculate the resource consumption information and latency prediction results.

5. The UAV edge network control method according to claim 4, characterized in that, Based on the node load rate, the resource usage of each edge node is calculated, including: Based on a preset load rate threshold table, the node load rate of each edge node is classified into different load rate zones. For edge nodes in the same load rate zone, resource usage frequency data over a historical period is collected and combined with the current resource usage to perform resource usage trend analysis, resulting in a resource usage trend map for each edge node. Based on the resource usage trend chart, the resource usage of each edge node in different time periods is statistically analyzed in segments. The resource usage changes within adjacent time periods that are less than a preset value are merged to obtain the resource usage of each edge node.

6. A drone edge network control system, characterized in that, The method for performing the UAV edge network control according to any one of claims 1 to 5 includes: The acquisition module is used to collect the status data of each edge node in the edge network of the UAV swarm, and to classify the computing tasks acquired by the UAV swarm according to the status data to obtain a task priority list. The calculation module is used to calculate the matching degree between the drone swarm and edge nodes based on the task priority list to obtain a task allocation scheme; The prediction module is used to perform cumulative prediction of resource requirements for computing tasks allocated to each edge node in the task allocation scheme based on historical task execution data, to obtain the expected load value of the edge node, and to evaluate the available resources of each edge node based on the expected load value, to obtain an available resource curve. The module performs a difference calculation between the available resource curve and the status data of the edge node to obtain the resource contention degree, and to estimate the execution time of each computing task based on the resource contention degree, to obtain a task execution table. The module performs a weighted analysis of the time data in the task execution table and the resource contention degree to obtain a task completion index, and to evaluate the performance of each edge node based on the task completion index, to obtain resource consumption information and latency prediction results. The adjustment module is used to dynamically adjust the task allocation scheme based on the resource consumption information and latency prediction results to obtain task allocation instructions; Based on the task priority list, the matching degree between the drone and the edge node is calculated to obtain a task allocation scheme, including: The resource requirements in the task priority list are compared with the available computing resources of the edge nodes to obtain resource matching degree data. Based on the resource matching degree data, the computing load of the edge nodes is evaluated to obtain a node load distribution map. The node load distribution map is weighted and calculated with the physical distance from the UAV to the edge node to obtain a node affinity score. The edge nodes are then sorted based on the node affinity score to obtain a node priority queue. The node affinity score represents the degree of adaptation between the task and the computing resources of the edge node. A one-to-one mapping is performed between the edge nodes in the node priority queue and the computation tasks in the task priority list to obtain an initial allocation table. Load balancing verification is then performed based on the initial allocation table to obtain the task allocation scheme. Based on the resource consumption information and latency prediction results, the task allocation scheme is dynamically adjusted to obtain task allocation instructions, including: The resource usage in the resource consumption information and latency prediction results is sorted to obtain a resource usage sorting table, and the difference in resource usage between adjacent edge nodes in the resource usage sorting table is calculated to obtain a resource usage difference sequence. Based on the resource usage difference sequence, the resource stress of each edge node is evaluated to obtain a resource stress evaluation value. When the resource stress evaluation value exceeds a preset stress threshold, the computing tasks in the task allocation scheme are reassigned based on the resource stress evaluation value to obtain a task reassignment priority table. Based on the task redistribution priority table, the task allocation of each edge node is adjusted to obtain a task adjustment allocation table, and the task adjustment allocation table is converted into task allocation instructions. Based on the resource usage difference sequence, resource stress is assessed for each edge node to obtain resource stress assessment values, including: The interval distribution statistics of the difference data between adjacent edge nodes in the resource usage difference sequence are performed to obtain the node resource distribution curve; Based on the resource distribution curve, the resource occupancy status of each edge node is calculated in segments to obtain a resource occupancy distribution map. Then, fluctuation analysis is performed on the data of adjacent time periods in the resource occupancy distribution map to obtain a load fluctuation curve. The load fluctuation curve is compared with a preset load threshold to obtain the load over-limit range, and the stress of each edge node is quantified based on the load over-limit range to obtain the resource stress assessment value.

Images