Edge cloud resource management collaborative scheduling method based on artificial intelligence

Abstract

The invention discloses an edge cloud resource management collaborative scheduling method based on artificial intelligence, which relates to the technical field of resource scheduling, and comprises the following steps: obtaining resource states of edge nodes to obtain an edge resource state set; performing task matching according to the edge resource state set to obtain a matching feature set; performing matching score calculation according to the matching feature set to obtain a node matching score table; generating a minimum time delay scheduling path according to the node matching score table to obtain a task scheduling distribution table; performing task scheduling execution time prediction according to the task scheduling distribution table to obtain an execution time feedback set; and performing parameter optimization according to the execution time feedback set to obtain an optimization result. According to the method, the adaptability of scheduling is improved, the multi-target balance of scheduling is improved, scheduling conflicts and node congestion are effectively reduced, and the timeliness of scheduling and the resource utilization efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of resource scheduling technology, specifically to an artificial intelligence-based collaborative scheduling method for edge cloud resource management. Background Technology

[0002] With the convergence of edge computing and cloud computing, edge cloud architecture is gradually becoming a key infrastructure supporting low-latency, distributed deployment of complex computing tasks. To achieve multi-task execution and efficient collaboration among multiple nodes, edge cloud needs to perform reasonable resource management and task scheduling among heterogeneous computing nodes to meet the diverse needs of computing tasks.

[0003] In existing technologies, task scheduling and allocation have shortcomings: existing task scheduling methods are usually based on score-first or delay-first scheduling, only considering a single objective (such as maximum resource score or minimum delay), ignoring the queuing status of edge nodes when considering delay, which leads to a significant decrease in scheduling quality under resource redundancy or network bottlenecks and may cause scheduling conflicts or node congestion. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based collaborative scheduling method for edge cloud resource management, thereby resolving the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: In a first aspect, embodiments of the present invention provide an edge cloud resource management and collaborative scheduling method based on artificial intelligence, comprising the following steps: S1. Obtain the resource status of edge nodes to get the edge resource status set; S2. Perform task matching based on the edge resource status set to obtain the matching feature set; S3. Calculate the matching score based on the matching feature set to obtain the node matching score table; S4. Generate the minimum delay scheduling path based on the node matching score table to obtain the task scheduling allocation table; S5. Based on the task scheduling allocation table, predict the task scheduling execution time and obtain the execution time feedback set. S6. Optimize parameters based on the execution time feedback set to obtain the optimization results.

[0006] To further optimize this technical solution, the task matching in step S2 includes: Based on the obtained set of edge resource states, in order to achieve accurate mapping between resources and tasks, the requirement information of the task itself is matched with the resource state information of the node. A structured feature expression is established for all combinations between all tasks to be scheduled in the edge cloud and each available edge node, and a matching feature set is constructed.

[0007] To further optimize this technical solution, the matching score calculation in step S3 includes: Based on the obtained set of matching features, a comprehensive matching scoring mechanism is introduced through scoring model selection, model training data preparation, and scoring calculation. A matching scoring model is constructed to assign a score value to the matching relationship between each task and candidate node, evaluate the multi-dimensional matching relationship between tasks and nodes, reflect the adaptability of the current combination to task execution, and provide a basis for task scheduling.

[0008] To further optimize this technical solution, the minimum delay scheduling path generation in step S4 includes: Based on the obtained node matching score table and considering factors such as network communication latency, a scheduling score model is used to perform scheduling path reasoning, resulting in a task scheduling allocation table that indicates the edge node to which each task is assigned.

[0009] To further optimize this technical solution, the scheduling scoring model includes:

[0010] in: :Task Scheduled to node Comprehensive scheduling score; :Task With nodes Match score; :Task Scheduled to node Normalized delay; : Control factor.

[0011] To further optimize this technical solution, the normalized delay includes:

[0012] in: :Task Scheduled to node Total latency; Minimum total latency; Maximum total latency; By comparing the delay with the maximum and minimum delays of task scheduling to the node, the delay is normalized, and the normalized delay of task scheduling to the node is calculated.

[0013] To further optimize this technical solution, the total latency includes:

[0014] in: :Task Scheduled to node Network transmission latency; :node Waiting delay; The total latency of task scheduling to the target node is calculated by taking into account the network transmission latency of the task to the target node and the queuing latency of the target node.

[0015] To further optimize this technical solution, the waiting delay includes:

[0016] in: :node The number of tasks in the task queue; :node The task processing speed per unit time; The waiting time for a task to be scheduled to the target node is calculated based on the number of pending tasks and the task processing speed at the target node.

[0017] To further optimize this technical solution, the task scheduling execution time prediction in step S5 includes: Based on the obtained task scheduling and allocation table, a dynamic prediction mechanism is established for each task and target node scheduling pair. Through scheduling feature extraction, historical execution data analysis, prediction model selection, time sensitivity adjustment, and feedback set generation, the scheduling execution time prediction is further carried out. Based on the current node status, task type, and scheduling history, the task completion time is predicted to obtain the execution time feedback set.

[0018] To further optimize this technical solution, the parameter optimization in step S6 includes: Based on the obtained execution time feedback set, the parameters of the matching scoring model in step S3 are updated through scheduling deviation aggregation, parameter set update, and versioned model management. This makes the scoring model output closer to the actual scheduling performance, better reflects the changes in edge cloud resources and task characteristics, and improves the accuracy of the model.

[0019] In a second aspect, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program instructions, when executed by the processor, implement the steps of an artificial intelligence-based edge cloud resource management collaborative scheduling method as described in the first aspect of the present invention.

[0020] Thirdly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, they implement the steps of the collaborative scheduling method for edge cloud resource management based on artificial intelligence as described in the first aspect of the present invention.

[0021] Compared with existing technologies, this invention provides an artificial intelligence-based collaborative scheduling method for edge cloud resource management, which has the following beneficial effects: This AI-based collaborative scheduling method for edge cloud resource management improves scheduling adaptability and enhances the multi-objective balance of scheduling by using a scheduling scoring model that simultaneously considers the matching score between tasks and nodes and the delay from task to node. It also takes into account the delay of task queuing at nodes, effectively reducing scheduling conflicts and node congestion, and improving the timeliness and resource utilization efficiency of scheduling. Attached Figure Description

[0022] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a collaborative scheduling method for edge cloud resource management based on artificial intelligence proposed in this invention. Figure 2 This is a schematic diagram of the matching score calculation process for an artificial intelligence-based collaborative scheduling method for edge cloud resource management proposed in this invention. Figure 3 This is a flowchart illustrating the scheduling scoring model of an artificial intelligence-based collaborative scheduling method for edge cloud resource management proposed in this invention. Detailed Implementation

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0027] Example 1: Reference Figures 1-3 This is the first embodiment of the present invention, which provides an artificial intelligence-based edge cloud resource management and collaborative scheduling method, including the following steps: S1. Obtain the resource status of the edge nodes to get the edge resource status set.

[0028] In this embodiment, the resource status acquisition includes: Edge computing environments contain a large number of heterogeneous edge nodes. These nodes vary significantly in computing power, storage capacity, network capabilities, and operational stability, and their resource states are extremely dynamic. Without real-time monitoring of the actual operational status of each node, efficient task scheduling and resource coordination are impossible, easily leading to task failures, resource congestion, or latency issues. Furthermore, because edge nodes are typically deployed close to end users, their computing resources are limited and highly volatile, lacking the redundancy and fault tolerance capabilities of cloud centers. Therefore, constructing an accurate resource status description for each node is not only the foundation for scheduling decisions but also a crucial prerequisite for ensuring system service reliability.

[0029] The purpose of this step is to obtain the resource status of edge nodes and construct an edge node resource status set, which is an edge resource status set formed by uniformly summarizing the current available resource status of all edge nodes in the system. This unifies the resource quantification standard, allows for horizontal comparison of the resource status of different nodes, provides basic resource information input for subsequent steps, reflects changes in node operating status in real time, and improves the real-time performance and robustness of scheduling.

[0030] The implementation methods for this step include: Resource Acquisition: A resource status acquisition agent is deployed for each edge node. Resources are read through the Linux kernel's procfs interface and the container monitoring component cAdvisor. Each agent collects resources such as CPU utilization, available memory, network bandwidth usage, and disk I / O rate at fixed time intervals (e.g., once per second). An instantaneous resource fluctuation factor is introduced to characterize the stability of resource status. The fluctuation amplitude (e.g., standard deviation) is calculated by taking sample values ​​of the same resource dimension from several past time slices (e.g., the past 3 samples). If a node's resource fluctuates frequently, even if the current status value seems sufficient, it is marked as a fluctuating node. This is used to achieve more robust node selection during node scheduling and improve scheduling stability. Data standardization: After completing the collection of resource status data (including CPU, memory, bandwidth, disk, etc.) for each edge node, standardization processing is required to ensure that parameters with different physical units and dimensions can be used for subsequent processing under a unified scale. First, define the standard upper limit for each resource dimension, such as 100% CPU full load, 8GB maximum memory, 1000Mbps maximum network bandwidth, and 500MB / s maximum disk I / O rate. Then, normalize the collected resource status data by dividing the original collected value of each dimension by the corresponding standard upper limit value to obtain the dimensionless ratio value in the range of [0,1]. State vector construction: The current state values ​​and fluctuation information labels of each normalized resource parameter are encapsulated into a resource state vector with a unified structure; Resource state set construction: All resource state vectors are aggregated to construct a unified edge resource state set.

[0031] S2. Perform task matching based on the set of edge resource states to obtain a set of matching features.

[0032] In this embodiment, the task matching includes: In edge computing scenarios, the execution quality of a task depends not only on the resource capabilities of the nodes but also on whether the resource requirements of the task match the current resource status of the nodes. In step S1, the real-time resource status of all edge nodes is standardized and structured into a vector set to obtain an edge resource status set, which provides a clear description of the availability of each node in the edge cloud. In order to achieve accurate mapping between resources and tasks, it is necessary to match the task's own requirement information with the node's resource status information, and then construct a matching feature vector between the task and the node.

[0033] Traditional methods often simply compare the remaining resources of nodes to determine if they meet task requirements, neglecting the comprehensive correlation between multiple dimensions, making it difficult to guarantee the accuracy and interpretability of scheduling decisions. The purpose of this step is to establish a structured feature representation of all possible combinations between all tasks to be scheduled in the edge cloud and each available edge node, construct a matching feature set, thereby generating matching vectors for multiple candidate nodes for each task, evaluating all possible combinations, capturing the coupling relationship between resources and requirements, and achieving refined and intelligent scheduling decisions.

[0034] The implementation methods for this step include: Task requirement information extraction: For each task to be assigned in the scheduling queue, extract its key operating parameters, including computational requirements (such as maximum CPU utilization or instruction cycle count), memory usage requirements, expected network bandwidth usage, maximum tolerable task processing latency (i.e., service response time requirements, which are related to CPU performance, network bandwidth, and disk I / O rate), and task type or priority information (such as real-time tasks, batch processing tasks, etc.; real-time tasks have low demand fluctuations and more idle resources on nodes, while batch processing tasks can tolerate large resource fluctuations on nodes). Node resource combination mapping: For each task, select each currently available node from the edge resource state set and combine them to form a task-node pair; Constructing Feature Vectors: The resource status vectors of nodes (such as CPU utilization, memory remaining, bandwidth availability, I / O performance, whether it is a fluctuating node, etc.) are concatenated with the task's requirement parameters to form a structured feature vector. In order to improve the coupling and expressive ability between task and node features, a resource requirement ratio feature is added to the feature vector. That is, the resource requirement of the task is compared with the current resource availability of the node, which is used to describe the proportion of the current node's supply capacity relative to the task requirement.

[0035] Matching feature set generation: For each task, construct its combined features with all candidate edge nodes, and finally generate the matching feature set of the entire system, which is used as input for subsequent steps.

[0036] S3. Calculate the matching score based on the matching feature set to obtain the node matching score table.

[0037] In this embodiment, the matching score calculation includes: In edge computing environments, task scheduling faces complex constraints such as heterogeneous resource conditions, task priority conflicts, and strict real-time requirements. Simply matching resources based on whether they meet certain rules fails to reflect the differences between nodes, the adaptability of tasks to resources, and the expected execution efficiency. Therefore, a comprehensive matching and scoring mechanism is needed to uniformly evaluate the multi-dimensional matching relationship between tasks and nodes, providing a basis for judgment before resource selection and scheduling.

[0038] Based on the matching feature set obtained in step S2, this step assigns a score value to the matching relationship between each task and the candidate node to reflect the degree of suitability of the current combination for task execution, thereby providing a basis for task scheduling and facilitating horizontal selection of tasks among all nodes.

[0039] The implementation methods for this step include: Scoring model selection: Gradient Boosting Tree (GBDT) is used for scoring calculation. GBDT is a known stable and interpretable machine learning algorithm that can handle mixed data types and is suitable for learning complex but regular feature patterns in resource scheduling and matching. Model training data preparation: Real data from the system's historical scheduling records are used for training. The training samples include an input part and a label part. The input part is the matching feature vector of the task and the node. The label part refers to the comprehensive score converted from the results of whether the task was completed, the completion time, and whether it timed out after being scheduled to the node in the past. This comprehensive score is the judgment standard for whether the prediction is reasonable during model training. For example, for binary labels such as whether it was completed or whether it timed out, it is converted into a score of 1 or 0 according to whether the conditions are met, that is, 1 for completed and not timed out, and 0 for not completed and timed out. For continuous labels such as completion time, it is converted into a score in the range of 0 to 1 according to the proportion of the remaining available time to the maximum allowed time. Scoring Calculation: For all candidate nodes of each task, input their feature vectors into the pre-trained GBDT model, obtain the matching score of each node from the GBDT model (score range is [0, 1]), and retrieve the number of tasks or execution pressure level that each candidate node has been scheduled in the past period (e.g., the past 10 minutes). If a node has been frequently scheduled recently, the score will be lowered by a certain amount to avoid the node being over-called and causing resource congestion or hidden delays. Construct a corresponding node matching score table for each task, with node ID as the row and matching score as the column, to record the degree of fit of each node to the task.

[0040] S4. Generate the minimum latency scheduling path based on the node matching score table to obtain the task scheduling allocation table.

[0041] In this embodiment, the minimum latency scheduling path generation includes: Edge computing scheduling scenarios not only need to consider resource matching but also multiple practical factors such as network latency to effectively guarantee service response time and overall system throughput. Therefore, based on the node matching score table obtained in step S3, and combined with influencing factors such as network communication latency, a scheduling score model is used to perform scheduling path reasoning to obtain a task scheduling allocation table, indicating the edge node to which each task is assigned, thus ensuring the feasibility, adaptability, and resource utilization of scheduling.

[0042] Furthermore, the scheduling scoring model includes:

[0043] in: :Task Scheduled to node Comprehensive scheduling score; :Task With nodes The matching score, which reflects the degree to which a node is adapted to the task, is obtained through step S3; :Task Scheduled to node The normalized latency, ranging from [0, 1], represents the relative latency of task scheduling to different nodes; Control factor: Set according to the real-time nature of the task and resource sensitivity. For example, if real-time tasks are more concerned about latency, the control factor can be lowered to increase the impact of latency. If high-load tasks are more concerned about resources, the control factor can be increased to increase the impact of node adaptability.

[0044] Furthermore, the normalized delay includes:

[0045] in: :Task Scheduled to node The total latency includes network transmission and queuing latency; Minimum total latency represents the minimum latency value for task scheduling to all nodes; Maximum total latency: This represents the maximum latency required for a task to be scheduled to all nodes. By comparing the delay with the maximum and minimum delays of task scheduling to the node, the delay is normalized, and the normalized delay of task scheduling to the node is calculated.

[0046] Furthermore, the total delay includes:

[0047] in: :Task Scheduled to node The network transmission latency represents the network communication delay for task data to be transmitted to the target node, and is obtained through dynamic network monitoring. :node The waiting delay represents the queuing time for task processing at the target node; The total latency of task scheduling to the target node is calculated by taking into account the network transmission latency of the task to the target node and the queuing latency of the target node.

[0048] Furthermore, the waiting delay includes:

[0049] in: :node The number of tasks in the task queue, i.e., the number of tasks to be processed at the current node; :node The task processing speed per unit time represents the task processing capacity of the current node, that is, how many tasks can be processed per second or millisecond. The waiting time for a task to be scheduled to the target node is calculated based on the number of pending tasks and the task processing speed at the target node.

[0050] This model describes how to allocate task scheduling nodes based on a comprehensive scheduling score obtained from node matching scores and network latency.

[0051] Traditional task scheduling methods are typically based on score-first or delay-first scheduling, considering only a single objective (such as maximum resource score or minimum delay). This leads to a significant decrease in scheduling quality under resource redundancy or network bottlenecks. Furthermore, when considering delay, they ignore the queuing status of edge nodes, resulting in scheduling conflicts or node congestion. In contrast, this model considers both the matching score between tasks and nodes and the delay from task to node, exhibiting stronger scheduling adaptability. It significantly improves the multi-objective balance and environmental adaptability of scheduling. Moreover, by adding the delay of task queuing at nodes when calculating node delay, it better reflects the current state of target nodes, effectively improving the timeliness and resource utilization efficiency of scheduling.

[0052] The steps for using this model include: Data Acquisition: Obtain the matching score between tasks and nodes from step S3. The resource status of the node is obtained from step S1, including the number of tasks in the node's task queue. The task processing speed per unit time of a node Network transmission latency of task scheduling to nodes , as input to the model; Scheduling score calculation: Calculate the normalized latency of task scheduling to the node based on the acquired data. And combine task and node matching scoring and control factors Calculate the overall scheduling score for task scheduling to nodes. ; Scheduling and allocation: based on the calculated comprehensive scheduling score For each task, the node with the comprehensive scheduling score is selected as the scheduling target node in turn to obtain the task scheduling allocation table.

[0053] S5. Based on the task scheduling allocation table, predict the execution time of task scheduling and obtain the execution time feedback set.

[0054] In this embodiment, the task scheduling execution time prediction includes: Edge computing environments are highly dynamic, with frequent fluctuations in resource availability, network status, and node load. Even if the allocation scheme is optimal at the current moment, factors such as sudden load spikes and queue backlogs can lead to performance degradation during execution, resulting in unexpected high latency or resource conflicts, thus increasing task execution time. Therefore, based on the task scheduling allocation table obtained in step S4, a dynamic prediction mechanism needs to be established for each task-target node scheduling pair to further predict scheduling execution time. This involves predicting task completion time based on the current node status, task type, and scheduling history, thereby assessing the stability and rationality of the current scheduling strategy, providing feedback for the next stage of scheduling strategy, and assisting in optimizing scheduling effectiveness.

[0055] The implementation methods for this step include: Scheduling feature extraction: Extract scheduling information for each task from the task scheduling and allocation table, including task type, task resource requirements, target node number, current node load, network bandwidth, and the target node's recent execution records. Historical execution data analysis: The task scheduling history database is accessed to retrieve historical task execution samples that have a high degree of matching with the characteristics of the current task, including their allocation node, start time, actual completion time, execution duration, etc., to form a sample set; Prediction Model Selection: In edge computing scenarios, computing resources are limited, so it is necessary to select a lightweight model with high computing efficiency and low resource consumption to quickly predict the task execution time. This step selects a mature lightweight Bayesian regression model. This model constructs a Bayesian inference process by applying Gaussian priors to the input features to predict the execution time. Moreover, this model has few parameters, fast training convergence, and can be updated online. It is suitable for low-sample learning in edge environments and only requires a small amount of training data to obtain an effective prediction model. Time Sensitivity Adjustment: To improve prediction accuracy, a task urgency factor is introduced for fine-tuning based on the model prediction results. That is, after the model predicts the execution time of the task, the predicted value is dynamically corrected according to the time sensitivity of the task (i.e., the urgency weight set for each task type, for example, the urgency weight of ultra-real-time tasks is 1.2, and the urgency weight of relaxed tasks is 0.8). This makes the model output closer to the actual scheduling constraints, reflecting the different requirements of tasks for completion time. Based on the maximum tolerable time threshold for each task type (for example, the maximum tolerable time threshold for ultra-real-time tasks is 100ms, and the maximum tolerable time threshold for relaxed tasks is 1000ms), tasks exceeding the threshold are marked as potentially high-latency tasks. Feedback set generation: The predicted execution time for each task is summarized to form an execution time feedback set.

[0056] S6. Optimize parameters based on the execution time feedback set to obtain the optimization results.

[0057] In this embodiment, the parameter optimization includes: As time progresses, task structures, network environments, and node load distributions will shift. If the matching and scoring model remains static for an extended period, it will be unable to continuously adapt to scheduling requirements. The execution time feedback set obtained in step S5 records the difference between the actual completion time of a task after it is actually assigned and the predicted score. This is the core basis for correcting and enhancing the matching and scoring model. Therefore, this step updates the parameters of the matching and scoring model in step S3 based on the execution time feedback set, making the scoring model output closer to the actual performance of the current scheduling, conforming to the changes in edge cloud resources and the evolution of task characteristics, thereby improving the quality of task allocation decisions, enhancing the accuracy of the model in the next round of scheduling, improving the robustness of the model, and reducing the risk of model failure in edge scenarios.

[0058] The implementation methods for this step include: Scheduling Deviation Aggregation: Normalize each actual completion time obtained in each round of scheduling (to convert it into a dimensionless number in the range [0, 1], the closer to 1, the smaller the actual completion time), and then compare the deviation with the matching score of each task and corresponding node obtained by the matching scoring model in step S3. Identify the matching result of the task and node with the largest deviation between the output of the matching scoring model and the actual result. After multiple scheduling deviation comparisons, statistically analyze the matching feature combinations of tasks and nodes with frequent errors, mark them as error samples, and statistically analyze the mean, variance, and distribution of each input feature in error samples and non-error samples respectively. Identify the main error features in the frequently occurring matching feature combinations (i.e., input features whose actual impact on scheduling delay is greater than or less than the predicted impact). For example, the mean queue length of nodes in error samples is 9.2, while the mean queue length of nodes in non-error samples is only 2.1, indicating that the actual impact of the queue length of nodes on scheduling delay is greater than the predicted impact, and is marked as the main error feature. Parameter set update: The weights of the main error features in the matching scoring model are fine-tuned in a gradient manner. The adjustment range is determined by the difference between the actual execution time and the predicted execution time obtained in step S5. The adjustment range is obtained by multiplying the time difference by a conversion coefficient (used to convert the time difference into a dimensionless adjustment range, which is set empirically). The maximum allowable adjustment range for each adjustment is set (e.g., ±5%, to prevent model oscillation caused by sudden errors in extreme cases). The adjustment of weights prioritizes improving scheduling stability, with the aim of reducing task failure rate and reducing the range of scheduling delay fluctuations. Versioned model management: A new matching scoring model version is generated after each parameter update. A / B evaluation is performed in subsequent scheduling. If the new model performs better in real task scheduling (such as reduced average latency and improved task response rate), the new version is switched to.

[0059] Example 2: This embodiment also provides a computer device applicable to an AI-based edge cloud resource management collaborative scheduling method, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the AI-based edge cloud resource management collaborative scheduling method proposed in the above embodiment.

[0060] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, it implements an artificial intelligence-based collaborative scheduling method for edge cloud resource management as proposed in the above embodiment.

[0061] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0062] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0063] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0064] More specific examples (a non-exhaustive list) of computer-readable media include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0065] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0066] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A collaborative scheduling method for edge cloud resource management based on artificial intelligence, characterized in that, Includes the following steps: S1. Obtain the resource status of edge nodes to get the edge resource status set; S2. Perform task matching based on the edge resource status set to obtain the matching feature set; S3. Calculate the matching score based on the matching feature set to obtain the node matching score table; S4. Generate the minimum delay scheduling path based on the node matching score table to obtain the task scheduling allocation table; S5. Based on the task scheduling allocation table, predict the task scheduling execution time and obtain the execution time feedback set. S6. Optimize parameters based on the execution time feedback set to obtain the optimization results.

2. The edge cloud resource management and collaborative scheduling method based on artificial intelligence according to claim 1, characterized in that, The task matching in step S2 includes: Based on the obtained set of edge resource states, in order to achieve accurate mapping between resources and tasks, the requirement information of the task itself is matched with the resource state information of the node. A structured feature expression is established for all combinations between all tasks to be scheduled in the edge cloud and each available edge node, and a matching feature set is constructed.

3. The edge cloud resource management and collaborative scheduling method based on artificial intelligence according to claim 1, characterized in that, The matching score calculation in step S3 includes: Based on the obtained set of matching features, a comprehensive matching scoring mechanism is introduced through scoring model selection, model training data preparation, and scoring calculation. A matching scoring model is constructed to assign a score value to the matching relationship between each task and candidate node, evaluate the multi-dimensional matching relationship between tasks and nodes, reflect the adaptability of the current combination to task execution, and provide a basis for task scheduling.

4. The edge cloud resource management and collaborative scheduling method based on artificial intelligence according to claim 1, characterized in that, The minimum delay scheduling path generation in step S4 includes: Based on the obtained node matching score table and considering factors such as network communication latency, a scheduling score model is used to perform scheduling path reasoning, resulting in a task scheduling allocation table that indicates the edge node to which each task is assigned.

5. The edge cloud resource management and collaborative scheduling method based on artificial intelligence according to claim 4, characterized in that, The scheduling scoring model includes: ， in: :Task Scheduled to node Comprehensive scheduling score; :Task With nodes Match score; :Task Scheduled to node Normalized delay; : Control factor.

6. The edge cloud resource management and collaborative scheduling method based on artificial intelligence according to claim 5, characterized in that, The normalized delay includes: ， in: :Task Scheduled to node Total latency; Minimum total latency; Maximum total latency; By comparing the delay with the maximum and minimum delays of task scheduling to the node, the delay is normalized, and the normalized delay of task scheduling to the node is calculated.

7. The edge cloud resource management and collaborative scheduling method based on artificial intelligence according to claim 6, characterized in that, The total delay includes: ， in: :Task Scheduled to node Network transmission latency; :node Waiting delay; The total latency of task scheduling to the target node is calculated by taking into account the network transmission latency of the task to the target node and the queuing latency of the target node.

8. The edge cloud resource management and collaborative scheduling method based on artificial intelligence according to claim 7, characterized in that, The waiting delay includes: ， in: :node The number of tasks in the task queue; :node The task processing speed per unit time; The waiting time for a task to be scheduled to the target node is calculated based on the number of pending tasks and the task processing speed at the target node.

9. The edge cloud resource management and collaborative scheduling method based on artificial intelligence according to claim 1, characterized in that, The task scheduling execution time prediction in step S5 includes: Based on the obtained task scheduling and allocation table, a dynamic prediction mechanism is established for each task and target node scheduling pair. Through scheduling feature extraction, historical execution data analysis, prediction model selection, time sensitivity adjustment, and feedback set generation, the scheduling execution time prediction is further carried out. Based on the current node status, task type, and scheduling history, the task completion time is predicted to obtain the execution time feedback set.

10. The edge cloud resource management and collaborative scheduling method based on artificial intelligence according to claim 1, characterized in that, The parameter optimization in step S6 includes: Based on the obtained execution time feedback set, the parameters of the matching scoring model in step S3 are updated through scheduling deviation aggregation, parameter set update, and versioned model management. This makes the scoring model output closer to the actual scheduling performance, better reflects the changes in edge cloud resources and task characteristics, and improves the accuracy of the model.

Images