A wireless network scheduling optimization method and system for cloud-edge collaboration network

Abstract

The application provides a wireless network scheduling optimization method and system of a cloud-edge collaborative network. The method collects network behavior data of a communicable edge node in an initial granularity interval, and constructs a node behavior rule matrix. Feature extraction is performed based on the node behavior rule matrix to obtain a link feature vector of the communicable edge node. Feature learning is performed on the link feature vector to determine a task granularity adjustment factor corresponding to a current communication period. A corresponding communication sub-task allocation strategy is executed according to the network priority of the communication edge node. A tail delay evaluation index is constructed according to the maximum value of the completion time of each communication sub-task. The task granularity adjustment factor is adaptively updated based on the tail delay evaluation index, so that the communication task can be adaptively adjusted in granularity according to the node behavior rule of the current edge node in the initial granularity interval, and the accuracy and stability of the wireless network scheduling are improved.

Description

Technical Field

[0001] This application relates to the field of wireless network communication technology, and more specifically, to a wireless network scheduling optimization method and system for cloud-edge collaborative networks. Background Technology

[0002] With the continuous evolution of 5G wireless and future 6G technologies, cloud-edge collaborative networks based on mobile edge computing are gradually becoming an important architecture supporting low-latency, high-reliability communication services. In this architecture, computing and storage capabilities are shifted from the cloud to edge nodes, enabling smart terminal devices to interact directly with edge nodes through the wireless access network, thereby effectively shortening data transmission paths and reducing communication latency. In typical scenarios such as smart transportation, industrial internet, and drone collaboration, communication tasks often have characteristics such as large differences in data scale, diverse latency requirements, and complex resource dependencies, which place higher demands on wireless network scheduling.

[0003] In existing technologies, most wireless network scheduling methods process communication tasks using a fixed task segmentation granularity. That is, after tasks are generated, they are uniformly divided according to preset rules, and offloading and resource allocation are performed accordingly. However, tasks of different granularities have significantly different requirements for network and computing resources: coarse-grained tasks usually have large data volumes and are computationally intensive, requiring high link stability and node computing power, while fine-grained tasks have stronger decomposability and rely more on low-latency transmission and flexible scheduling capabilities. Under a fixed-granularity mechanism, resource allocation imbalances are prone to occur. For example, large-scale tasks may concentrate on high-performance nodes, leading to excessive load, while small-scale tasks may crowd out critical resources, affecting overall scheduling efficiency, thereby reducing system stability and increasing task completion latency.

[0004] Therefore, how to adaptively divide communication tasks into granularities based on the wireless link status and edge node resource conditions in a cloud-edge collaborative network environment has become a key technical problem that urgently needs to be solved. Summary of the Invention

[0005] This application provides a wireless network scheduling optimization method and system for cloud-edge collaborative networks, which can adaptively adjust the granularity of communication tasks based on the node behavior patterns of current edge nodes within the initial granularity range, thereby improving the accuracy of wireless network scheduling.

[0006] In a first aspect, this application provides a wireless network scheduling optimization method for cloud-edge collaborative networks. This method can be executed by a network device, or it can be executed by a chip configured in the network device. This application does not limit the method in this regard.

[0007] Specifically, the method includes: Obtain communication requests and corresponding communication task data from user devices, construct task granularity representation parameters based on the communication task data, and generate an initial granularity range for the communication task data based on the task granularity representation parameters. Within a preset historical communication period, network behavior data of communicable edge nodes are collected within the initial granularity range to construct a node behavior pattern matrix. Based on the node behavior pattern matrix, feature extraction is performed to obtain the link feature vector of the communicable edge node. Feature learning is then performed on the link feature vector to determine the corresponding task granularity adjustment factor within the current communication cycle. The initial granularity range of the communication task data is dynamically adjusted according to the task granularity adjustment factor to obtain multiple communication sub-tasks, and the corresponding communication sub-task allocation strategy is executed according to the network priority of the communication edge node. A tail delay evaluation index is constructed based on the maximum completion time of each communication subtask, and the task granularity adjustment factor is adaptively updated based on the tail delay evaluation index.

[0008] In conjunction with the first aspect, in certain implementations of the first aspect, collecting network behavior data of communicable edge nodes within the initial granularity range and constructing a node behavior pattern matrix specifically includes: Multiple candidate granularities are determined based on the initial granularity range. Within a preset historical communication period, network behavior data of each edge node at different time windows corresponding to the candidate granularities are collected. The network behavior data includes wireless link status information and resource status information of each edge node. The network behavior data is arranged according to the node dimension and the time dimension to construct a node behavior pattern matrix.

[0009] In conjunction with the first aspect, in certain implementations of the first aspect, feature extraction based on the node behavior pattern matrix to obtain the link feature vector of the communicable edge node specifically includes: Statistical analysis is performed on the node behavior pattern matrix in the time dimension to extract the behavioral statistical features corresponding to each candidate granularity, and the corresponding granularity statistical time series is formed according to the time sequence. Node statistical features within different time windows are obtained, and the overall behavior sequence of nodes is formed according to the window time sequence. Correlation analysis is performed between the granular statistical time series under different candidate granularity conditions and the overall behavior sequence of nodes to determine the behavior correlation degree under different candidate granularity conditions. Link feature vectors are formed based on the behavioral correlation under different candidate granularity conditions.

[0010] In conjunction with the first aspect, in some implementations of the first aspect, performing feature learning on the link feature vector to determine the corresponding task granularity adjustment factor within the current communication cycle specifically includes: using a pre-trained convolutional neural network model as a link state discrimination model, classifying the link feature vector, and determining the corresponding task granularity adjustment factor based on the classification result.

[0011] In conjunction with the first aspect, in certain implementations of the first aspect, the initial granularity range for generating the communication task data based on the task granularity representation parameters specifically includes: After normalizing the task granularity representation parameters, the granularity benchmark value of the communication task data is generated by parameter weighting. The granularity reference value is expanded according to the preset interval length to obtain the initial granularity interval of the communication task.

[0012] In conjunction with the first aspect, in some implementations of the first aspect, constructing task granularity representation parameters based on the communication task data specifically includes: extracting features from the communication task data to obtain task granularity representation parameters used to represent task characteristics. The task granularity representation parameters include at least data volume parameters, computational complexity parameters, latency sensitivity parameters, and service quality level parameters.

[0013] In conjunction with the first aspect, in some implementations of the first aspect, during the process of obtaining the communication request and corresponding communication task data of the user equipment, the communication request and corresponding communication task data originate from the user equipment report.

[0014] Secondly, this application provides a wireless network scheduling optimization system for a cloud-edge collaborative network, which includes a communication allocation unit, the communication allocation unit comprising: The data acquisition module is used to acquire communication requests and corresponding communication task data of user devices, construct task granularity representation parameters based on the communication task data, and generate an initial granularity range of the communication task data based on the task granularity representation parameters. The data analysis module is used to collect network behavior data of communicable edge nodes within the initial granularity range within a preset historical communication period, and to construct a node behavior pattern matrix. The data analysis module is also used to extract features based on the node behavior pattern matrix to obtain the link feature vector of the communicable edge node, perform feature learning on the link feature vector, and determine the corresponding task granularity adjustment factor within the current communication cycle. The task execution module is used to dynamically adjust the initial granularity range of the communication task data according to the task granularity adjustment factor to obtain multiple communication sub-tasks, and execute the corresponding communication sub-task allocation strategy according to the network priority of the communication edge node. The feedback adjustment module is used to construct a tail delay evaluation index based on the maximum completion time of each communication subtask, and to adaptively update the task granularity adjustment factor based on the tail delay evaluation index.

[0015] Thirdly, this application provides a computer terminal device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the aforementioned wireless network scheduling optimization method for a cloud-edge collaborative network.

[0016] Fourthly, this application provides a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor to implement the operations performed by the aforementioned wireless network scheduling optimization method for a cloud-edge collaborative network.

[0017] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: This application provides a wireless network scheduling optimization method and system for cloud-edge collaborative networks. First, it acquires communication requests and corresponding communication task data from user devices. Based on the communication task data, it constructs task granularity representation parameters and generates an initial granularity interval for the communication task data. It then collects network behavior data of communicable edge nodes within the initial granularity interval to construct a node behavior pattern matrix. Based on the node behavior pattern matrix, it extracts features to obtain link feature vectors of communicable edge nodes. It performs feature learning on the link feature vectors to determine the corresponding task granularity adjustment factor within the current communication cycle. Based on the task granularity adjustment factor, it dynamically adjusts the initial granularity interval of the communication task data to obtain multiple communication sub-tasks. It then executes the corresponding communication sub-task allocation strategy according to the network priority of the communication edge nodes. Finally, it constructs a tail latency evaluation index based on the maximum completion time of each communication sub-task and adaptively updates the task granularity adjustment factor based on the tail latency evaluation index.

[0018] Therefore, this application constructs task granularity representation parameters based on communication task data and generates an initial granularity interval, providing an adjustable range for task partitioning. Subsequently, network behavior data of edge nodes is collected within this granularity interval, and a node behavior pattern matrix is ​​constructed. This allows for a structured representation of information such as link latency, throughput, and resource load of nodes under different time windows and candidate granularity conditions. By extracting and learning features from the node behavior pattern matrix, link feature vectors are obtained, quantifying the correlation between network behavior changes and node resource status under different granularity conditions. This enables modeling of the granularity-node performance relationship, resulting in model output. The task granularity adjustment factor can reflect the optimal task partitioning tendency under the current network environment, enabling the initial granularity interval to dynamically shift according to the real-time network status, generating communication subtasks that conform to the current network carrying capacity. At the same time, the task execution results are fed back through the tail latency evaluation index. When the maximum subtask completion time is detected to deviate from the target latency, the task granularity adjustment factor is adaptively updated, thus forming a closed-loop optimization mechanism. This allows the granularity adjustment to continuously approach the optimal scheduling state, enabling the task partitioning granularity to dynamically match the real-time status of edge nodes, avoiding resource mismatch problems caused by fixed granularity, and thus significantly improving the accuracy and stability of wireless network scheduling in cloud-edge collaborative networks.

[0019] Based on the current behavior patterns of edge nodes within the initial granularity range, the communication task is adaptively adjusted in granularity, which improves the accuracy and stability of wireless network scheduling. Attached Figure Description

[0020] Figure 1 This is an exemplary flowchart of a wireless network scheduling optimization method for a cloud-edge collaborative network according to some embodiments of this application; Figure 2 This is a flowchart illustrating the segmentation and unloading of communication tasks in some embodiments of this application; Figure 3 This is a schematic diagram of the structure of a communication distribution unit according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a computer terminal device that implements a wireless network scheduling optimization method for a cloud-edge collaborative network according to some embodiments of this application. Detailed Implementation

[0021] This application acquires communication requests and corresponding communication task data from user equipment, constructs task granularity representation parameters based on the communication task data, and generates an initial granularity interval for the communication task data based on the task granularity representation parameters. It collects network behavior data of communicable edge nodes within the initial granularity interval to construct a node behavior pattern matrix. Based on the node behavior pattern matrix, it extracts features to obtain link feature vectors of communicable edge nodes, performs feature learning on the link feature vectors, and determines the corresponding task granularity adjustment factor within the current communication cycle. It dynamically adjusts the initial granularity interval of the communication task data according to the task granularity adjustment factor to obtain multiple communication sub-tasks, and executes the corresponding communication sub-task allocation strategy according to the network priority of the communication edge nodes. It constructs a tail latency evaluation index based on the maximum completion time of each communication sub-task, and adaptively updates the task granularity adjustment factor based on the tail latency evaluation index. This allows for adaptive granularity adjustment of communication tasks based on the node behavior patterns of the current edge nodes within the initial granularity interval, improving the accuracy and stability of wireless network scheduling.

[0022] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific implementation methods. (Reference) Figure 1 The figure is an exemplary flowchart of a wireless network scheduling optimization method for a cloud-edge collaborative network according to some embodiments of this application. The wireless network scheduling optimization method 100 for the cloud-edge collaborative network mainly includes the following steps: In step S101, the communication request of the user equipment and the corresponding communication task data are obtained, a task granularity representation parameter is constructed based on the communication task data, and an initial granularity range of the communication task data is generated based on the task granularity representation parameter.

[0023] Preferably, in some embodiments, during the process of obtaining the communication request and corresponding communication task data of the user equipment, the communication request and corresponding communication task data are reported by the user equipment, and the communication task data includes the data content to be transmitted, the service type, and the service quality level parameters.

[0024] Optionally, in some embodiments, during the process of obtaining communication requests and corresponding communication task data from user equipment, the user equipment includes, but is not limited to, terminal devices connected to the wireless network such as smartphones, tablets, laptops, wearable devices, vehicle terminals, industrial control equipment, IoT terminals, and sensor nodes; the communication requests and corresponding communication task data originate from reports submitted by the user equipment, and the communication task data includes the data content to be transmitted, service type, and service quality level parameters.

[0025] Furthermore, the communication tasks described in this application are tasks that transmit data or perform collaborative processing through a wireless network, including but not limited to real-time data transmission tasks, periodic data reporting tasks, video stream transmission tasks, control signaling interaction tasks, edge computing offloading tasks, and multimedia data processing tasks; among them, different types of communication tasks correspond to different latency requirements, bandwidth requirements, and reliability requirements.

[0026] Optionally, in some embodiments, constructing task granularity characterization parameters based on the communication task data specifically includes: extracting features from the communication task data to obtain task granularity characterization parameters used to characterize task characteristics. The task granularity characterization parameters include at least data volume parameters, computational complexity parameters, latency sensitivity parameters, and service quality level parameters.

[0027] In specific implementation, the data volume parameter is used to characterize the data scale of the communication task, which is determined by statistically analyzing the data content reported by the user equipment, or by obtaining the data volume cached by the wireless protocol stack; the computational complexity parameter is used to characterize the computing resources required for the communication task to execute at the edge node, which is determined by a preset mapping relationship between task type and size and computational complexity; the latency sensitivity parameter is used to characterize the sensitivity of the communication task to transmission latency, which is obtained by mapping according to the service type corresponding to the communication task and preset latency constraints; the service quality level parameter is used to characterize the priority of the communication task, which can be obtained through a preset service quality identifier in the wireless communication system.

[0028] It should be noted that this application, based on the traditional task offloading process, introduces a dynamic task splitting mechanism based on task granularity adjustment factors and a scheduling strategy based on node network priority. This enables the task offloading process to adapt to changes in the wireless network state, improving the stability and efficiency of task execution. (See reference...) Figure 2 As shown, Figure 2 This is a flowchart illustrating the task segmentation and unloading process in some embodiments of this application. The task segmentation and unloading process mainly includes data return, task generation, node search, task splitting, unloading decision, data transmission, task reception, resource scheduling, parallel processing, and resource reclamation. These stages coordinate with each other to form a closed-loop optimization process, specifically including: (1) Task generation stage: The user equipment generates communication task data according to the service request. The communication task may include data transmission task, edge computing task or hybrid task. At the same time, task-related feature parameters are extracted, including data scale, latency sensitivity and service quality requirements, etc., to provide a basis for subsequent granular division. (2) Node search phase: Based on the current wireless network environment, the user equipment or scheduling control unit searches for communicable edge nodes and obtains the network status information and resource status information of candidate nodes, including link latency, throughput, packet loss rate and node computing load, etc. (3) Task splitting stage: Based on the pre-calculated task granularity adjustment factor, the initial granularity range of the communication task is dynamically adjusted to determine the target granularity value, and the communication task is divided into multiple communication sub-tasks accordingly, so as to realize the adaptive control of task granularity. (4) Unloading decision stage: Based on the network priority of each edge node and the characteristics of the task subtask, determine the execution location of each subtask, including local execution or unloading to the corresponding edge node for execution, so as to achieve optimal task allocation; (5) Data transmission stage: The communication subtasks that need to be offloaded are sent to the corresponding edge nodes through the wireless communication link. During the transmission process, the transmission strategy can be dynamically adjusted in combination with the link status to improve the reliability and efficiency of data transmission. (6) Task receiving stage: Edge nodes receive the corresponding communication subtasks and cache and parse the tasks to provide a basis for subsequent resource scheduling and execution; (7) Resource scheduling phase: The edge node allocates resources to the received communication subtasks according to the current resource status and task priority, including the scheduling of computing resources, storage resources and communication resources, so as to ensure the time delay requirements of task execution; (8) Parallel processing stage: Each edge node or local device performs parallel processing on the assigned communication subtasks. Subtasks of different granularities can be executed collaboratively among multiple nodes to improve the overall task processing efficiency.

[0029] Preferably, in some embodiments, the initial granularity range for generating the communication task data based on the task granularity characterization parameters specifically includes: After normalizing the task granularity representation parameters, the granularity benchmark value of the communication task data is generated by parameter weighting. The granularity reference value is expanded according to the preset interval length to obtain the initial granularity interval of the communication task.

[0030] In specific implementation, the normalization process can adopt a linear normalization method based on historical statistical ranges to map data volume parameters, computational complexity parameters, latency sensitivity parameters, and service quality level parameters in the task granularity representation parameters to a unified numerical range. Then, the normalized parameters are weighted and fused according to preset weights to generate a granularity benchmark value representing the scale of communication task segmentation. Different weights reflect the degree of influence of each parameter on task granularity division. According to preset granularity mapping rules, the granularity benchmark value is mapped to a corresponding interval length. The interval length represents the size of the granularity interval range of the communication task under the current conditions. The upper and lower limits are set to KB, MB, and GB based on the granularity benchmark value. Then, based on the granularity benchmark value, an upward and downward floating expansion is performed to generate an initial granularity interval containing multiple candidate granularities. In this application, the upper limit of the initial granularity interval corresponds to a coarser granularity division method, and the lower limit corresponds to a finer granularity division method.

[0031] In step S102, within a preset historical communication period, network behavior data of communicable edge nodes within the initial granularity range are collected to construct a node behavior pattern matrix.

[0032] It should be noted that in a cloud-edge collaborative network environment, the link status and resource load of different edge nodes will change dynamically over time. In existing technologies, task splitting based on fixed granularity cannot reflect this time-varying characteristic, which can easily lead to a mismatch between task division and node carrying capacity, resulting in latency jitter or uneven resource utilization. This solution constructs task granularity representation parameters based on communication task data and generates an initial granularity range, so that task division has an adjustable range. Within this granularity range, network behavior data of edge nodes are collected, and a node behavior pattern matrix is ​​constructed, thereby structurally expressing information such as link latency, throughput, and resource load of nodes under different time windows and different candidate granularity conditions.

[0033] Optionally, in some embodiments, the preset historical communication period is set according to the wireless network scheduling period and service latency requirements, preferably a time range of several consecutive scheduling periods, and the network behavior data of the communicable edge nodes is collected through a sliding time window.

[0034] Preferably, in some embodiments, collecting network behavior data of communicable edge nodes within the initial granularity range and constructing a node behavior pattern matrix specifically includes: Multiple candidate granularities are determined based on the initial granularity range. Within a preset historical communication period, network behavior data of each edge node at different time windows corresponding to the candidate granularities are collected. The network behavior data includes wireless link status information and resource status information of each edge node. The network behavior data is arranged according to the node dimension and the time dimension to construct a node behavior pattern matrix.

[0035] The network behavior data includes wireless link status information and resource status information of each edge node; the wireless link status information includes at least the round-trip time, link throughput, packet loss rate and channel quality parameters of the corresponding candidate granularity task, and the resource status information includes at least the node computing load, task queue length and processing latency of the corresponding candidate granularity task.

[0036] In the process of collecting network behavior data of the communicable edge nodes through a sliding time window, the sliding time window is divided according to a fixed time interval, which matches the transmission time interval. For example, in a 5G NR system, the time window can be divided based on the transmission time interval or several time slots, and historical communication data is continuously updated through a sliding window. Within each time window, network behavior data of corresponding candidate granularity tasks are obtained through user equipment reporting information and wireless network side monitoring mechanisms. The wireless link status information is obtained through wireless protocol stack measurement, including round-trip delay calculated based on packet round-trip interaction, link throughput calculated based on actual transmitted data volume and time statistics, packet loss rate calculated through packet repetition, and channel quality parameters obtained through channel measurement. The resource status information is obtained through the edge node monitoring module, including computing load represented by node CPU or computing power utilization, task queue length represented by task queuing status, and processing delay obtained through actual task execution statistics. In some embodiments, this application determines network behavior data at different candidate granularities in the following ways: for historically executed tasks, the actual executed tasks are mapped to the corresponding granularity categories according to the task division method, and the corresponding network behavior data is extracted; in specific implementation, when there is a time delay in collecting the corresponding network behavior data, the network behavior data under different candidate granularity conditions can also be estimated through the scheduling simulation and prediction model of the current network state, and this application does not limit this; on this basis, the network behavior data of each edge node under different time windows and different candidate granularity conditions are arranged according to the node dimension and time dimension to construct a node behavior pattern matrix, which is used to characterize the network state change trend of each edge node under different granularity division methods.

[0037] It should be noted that, in this application, for executed communication tasks, the system obtains the task splitting method and the execution records of corresponding subtasks through task scheduling logs, thereby obtaining observation data such as transmission latency, processing latency, and task completion time under different granularity conditions. For granularity partitioning methods that do not appear directly in historical periods, estimation is performed based on the current network state and task data characteristics using simulation scheduling or equivalent mapping methods. The equivalent mapping method includes a latency amortization model based on data volume ratio. In some embodiments, equivalent mapping can also be performed simultaneously based on a resource allocation model with computational complexity weights to obtain network behavior data under the corresponding candidate granularity. This application will not elaborate on this further. In this application, for obtaining network behavior data of fine-grained tasks, the system records the task splitting process to achieve independent latency statistics at the subtask level. When the task is not explicitly split, the transmission latency and processing latency of the overall task are decomposed according to preset rules to obtain equivalent latency data corresponding to the fine-grained partitioning method, thereby ensuring the data comparability of network behavior data under different granularity conditions.

[0038] In step S103, feature extraction is performed based on the node behavior pattern matrix to obtain the link feature vector of the communicable edge node. Feature learning is then performed on the link feature vector to determine the corresponding task granularity adjustment factor within the current communication cycle.

[0039] It should be noted that this application extracts and learns features from the node behavior pattern matrix to obtain link feature vectors, enabling a quantitative expression of the correlation between network behavior changes and node resource status under different granularity conditions, thereby achieving modeling of the granularity-node performance relationship. Based on this, the task granularity adjustment factor output by the model can reflect the optimal task partitioning tendency under the current network environment, allowing the initial granularity interval to dynamically shrink or shift according to the real-time network status, thus generating communication subtasks that better suit the current network carrying capacity.

[0040] Preferably, in some embodiments, feature extraction based on the node behavior pattern matrix to obtain the link feature vector of the communicable edge node specifically includes: Statistical analysis is performed on the node behavior pattern matrix in the time dimension to extract the behavioral statistical features corresponding to each candidate granularity, and the corresponding granularity statistical time series is formed according to the time sequence. Node statistical features within different time windows are obtained, and the overall behavior sequence of nodes is formed according to the window time sequence. Correlation analysis is performed between the granular statistical time series under different candidate granularity conditions and the overall behavior sequence of nodes to determine the behavior correlation degree under different candidate granularity conditions. Link feature vectors are formed based on the behavioral correlation under different candidate granularity conditions.

[0041] In specific implementation, the node behavior pattern matrix is ​​statistically analyzed in the time dimension. For each candidate granularity, the normalized values ​​of the corresponding behavioral statistical features are extracted within different time windows. The behavioral statistical features include at least the mean round-trip delay, the variance of round-trip delay, the maximum delay, the mean throughput, and the packet loss rate. Based on the behavioral statistical features, they are arranged in the order of the time windows to construct the granularity statistical time series of the corresponding candidate granularity, which characterizes the dynamic characteristics of network performance changes over time under different granularity conditions.

[0042] Furthermore, the overall statistical characteristics of each edge node within different time windows are obtained. These overall statistical characteristics include the mean average round-trip time delay, mean average round-trip time delay variance, mean maximum delay, mean throughput, and mean packet loss rate for each node during the overall task transmission process. A sequence of overall node behavior is constructed based on the time window order to characterize the overall operational status changes of the edge nodes within historical communication cycles. Correlation analysis is performed between the granular statistical time series under different candidate granularity conditions and the overall node behavior sequence. This correlation analysis uses the Pearson correlation coefficient calculation method to measure the coupling relationship between network behavior changes and overall node status changes under different granularity conditions. The correlation calculation results are used to characterize the correlation strength between network latency fluctuations and node resource changes during task execution under corresponding candidate granularity conditions. Then, based on the behavioral correlation degree under different candidate granularity conditions, the nodes are sorted according to the granularity dimension to generate link feature vectors. These link feature vectors characterize the correlation characteristics between edge node communication performance and resource status under each candidate granularity condition.

[0043] Preferably, in some embodiments, performing feature learning on the link feature vector to determine the corresponding task granularity adjustment factor within the current communication cycle specifically includes: using a pre-trained convolutional neural network model as a link state discrimination model, classifying the link feature vector, and determining the corresponding task granularity adjustment factor based on the classification result.

[0044] The following is a specific embodiment of this application using a pre-trained convolutional neural network model to perform feature classification on the link feature vector: Link feature vectors are constructed as multi-dimensional feature data and input into a pre-trained convolutional neural network model. Each dimension of the link feature vector includes the correlation value between the granular statistical time series under different candidate granularity conditions and the overall node behavior sequence. The correlation value is used to characterize the coupling relationship between network behavior changes and overall node resource state changes under the corresponding candidate granularity conditions. The convolutional neural network model in this application includes an input layer, multiple convolutional layers, pooling layers, and a fully connected output layer. The convolutional layers of the convolutional neural network are used to extract local patterns from the correlation sequence features corresponding to different candidate granularities. The pooling layers are used to compress and enhance key correlation features of different granularity dimensions. The fully connected layers are used to perform nonlinear mapping on high-dimensional correlation features and output the corresponding granularity-adjusted classification results. In the model training phase, this application pre-generates a training sample dataset based on a node behavior pattern matrix constructed within historical communication cycles. It then labels the task completion latency stability under different candidate granularity conditions with granularity-adapted tags, combining the actual task execution results. These granularity-adapted tags include at least latency fluctuation level tags, tail latency level tags, and task execution stability level tags. The granularity-adapted tags are obtained by a pre-trained convolutional neural network model classifying the correlation features between granularity statistical time series and the overall node behavior sequence under different candidate granularity conditions, forming a labeled granularity-adapted training sample set. The link feature vector is input into the convolutional neural network model for forward propagation calculation to obtain the classification result. The output result is compared with the corresponding granularity-adapted tag annotation result. When the error between the model output result and the actual annotation exceeds a preset threshold, the model parameters are iteratively updated using a backpropagation algorithm to optimize the convolution kernel parameters and fully connected layer weight parameters until the model converges, completing the pre-trained model construction. In the model application stage, this application inputs the link feature vectors under different candidate granularity conditions within the current communication cycle into the trained convolutional neural network model, and obtains multiple corresponding granularity adaptation labels through the model output layer. Further, for the multiple granularity adaptation labels, the mapping relationship between the labels and granularity adjustment capabilities is obtained, and each granularity adaptation label is mapped to a corresponding label mapping value. The label mapping value is used to characterize the degree of adaptation of task partitioning under the corresponding granularity condition. Optionally, in some embodiments, a regularized hierarchical mapping method is adopted, directly assigning corresponding numerical ranges based on the hierarchical results of the granularity adaptation labels. For example, low-latency waveforms... The combined label of dynamic + low tail latency + high stability is mapped to a fine-grained priority adjustment value range (0.2~0.4); the combined label of medium volatility + medium tail latency + medium stability is mapped to a medium-grained range (0.4~0.7); and the combined label of high volatility + high tail latency + low stability is mapped to a coarse-grained range (0.7~0.9), realizing a direct numerical expression based on stability classification. In some other embodiments, a multi-index weighted mapping method can also be used to convert the latency volatility level, tail latency level and task execution stability index into numerical scores respectively, and generate label mapping values ​​through weighted fusion. This application does not limit this.

[0045] After mapping each granularity adaptation label to its corresponding label mapping value, a weighted fusion process is then performed based on the label mapping value corresponding to each granularity adaptation label. The weight coefficient is determined based on the classification probability or confidence level output by the model. By averaging and weighting each label mapping value, the task granularity adjustment factor within the current communication cycle is obtained, and the task granularity adjustment factor is used as the basis for subsequent dynamic adjustment of task granularity. Furthermore, the task granularity is dynamically adjusted based on the task granularity adjustment factor. Specifically, when the task granularity adjustment factor is adapted to a highly relevant category, the task granularity is increased to improve parallel processing capabilities; when the task granularity adjustment factor is adapted to a moderately relevant category, a moderate granularity partitioning strategy is adopted to balance communication overhead and processing stability; when the task granularity adjustment factor is adapted to a lowly relevant category, the task partitioning granularity is reduced to decrease the impact of network fluctuations on task completion latency. The task granularity adjustment factor in this application is determined based on the correlation between the stability of subtasks and the overall stability of nodes within the current granularity partitioning interval. This task granularity adjustment factor is used for rule-based control of task granularity and adaptively adjusted according to different wireless network load states and edge node resource states, thereby achieving cloud-edge collaborative task scheduling optimization based on granularity correlation learning.

[0046] In step S104, the initial granularity range of the communication task data is dynamically adjusted according to the task granularity adjustment factor to obtain multiple communication sub-tasks, and the corresponding communication sub-task allocation strategy is executed according to the network priority of the communication edge node.

[0047] Preferably, in some embodiments, during the process of dynamically adjusting the initial granularity range of the communication task data according to the task granularity adjustment factor to obtain multiple communication sub-tasks, a target granularity value is determined according to the task granularity adjustment factor and the initial granularity range, and the communication task data is divided based on the target granularity value.

[0048] In specific implementation, the target granularity value is determined based on the task granularity adjustment factor and the initial granularity interval, wherein the task granularity adjustment factor is used to characterize the relative position of the target granularity within the initial granularity interval. Preferably, the initial granularity interval is first subjected to interval shrinkage processing to reduce the instability caused by frequent granularity fluctuations. For example, the initial granularity interval is compressed by a preset shrinkage coefficient to make the granularity adjustment range more concentrated. In some embodiments, taking an initial granularity range of 10MB to 50MB as an example, when the system's preset shrinkage coefficient is 0.8, the granularity range can be shrunk to 14MB to 46MB. Based on the task granularity adjustment factor, interpolation calculation is performed within the shrunk granularity range to determine the target granularity value. For example, when the task granularity adjustment factor is 0.3, it indicates a preference for fine granularity, and the target granularity value can be obtained through interval linear interpolation: target granularity value ≈ 14MB + 0.3 × (46MB − 14MB) ≈ 23.6MB; In another embodiment, when the task granularity adjustment factor is 0.8, it indicates a preference for coarse granularity, and the corresponding target granularity value is approximately: target granularity value ≈ 14MB + 0.8 × (46MB − 14MB) ≈ 39.6MB. Further, the communication task data is divided according to the target granularity value. Assuming the total data volume of the current communication task is 120MB, when the target granularity value is approximately 24MB, the task can be divided into 5 sub-tasks; when the target granularity value is approximately 40MB, the task can be divided into 3 sub-tasks, thereby achieving dynamic adjustment of the task division granularity. In some embodiments, when the system adopts a discrete granularity scheduling mechanism, the target granularity value can be mapped to the closest granularity level in a preset granularity set. For example, if the preset granularity set is {10MB, 20MB, 30MB, 40MB, 50MB}, then when the calculated target granularity value is 23.6MB, it can be mapped to 20MB / 30MB to achieve matching with the actual scheduling system.

[0049] Preferably, in some embodiments, executing the corresponding communication subtask allocation strategy based on the network priority of the communication edge node specifically includes: The communication subtask is tagged with attributes, and the attribute tags include at least data size, latency requirements and priority information; Based on the network behavior data and resource status information of the communicable edge nodes, the network priority of each edge node is calculated. The network priority is used to characterize the communication and processing capabilities of each edge node in the current communication cycle. Each edge node is sorted according to the network priority, and the multiple communication subtasks are assigned to the corresponding edge nodes for execution according to a preset allocation strategy, wherein the allocation strategy includes an allocation method based on priority sorting.

[0050] In step S105, a tail delay evaluation index is constructed based on the maximum completion time of each communication subtask, and the task granularity adjustment factor is adaptively updated based on the tail delay evaluation index.

[0051] A tail latency evaluation index is constructed based on the maximum completion time of each communication subtask. The task granularity adjustment factor is then adaptively updated based on this tail latency evaluation index, specifically including: The completion time of each communication subtask is obtained, and a tail delay evaluation index is determined based on the completion time. The tail delay evaluation index is the maximum value among the completion times of each communication subtask. The tail delay evaluation index is then normalized with a preset target delay to obtain a delay deviation index.

[0052] The task granularity adjustment factor is updated based on the latency deviation index. When the tail latency is higher than the target latency, the task granularity adjustment factor is increased based on a preset first correction coefficient to reduce the number of task divisions. When the tail latency is lower than the target latency, the task granularity adjustment factor is decreased based on a preset second correction coefficient to improve the parallelism of tasks. In some embodiments, the task granularity adjustment factor can also be adjusted by incremental update, and the update result is subject to boundary constraints and smoothing to improve the stability and robustness of the granularity adjustment process.

[0053] In specific implementation, the completion time of each communication subtask is obtained, and a tail delay evaluation index is determined based on the completion time. The tail delay evaluation index is the maximum value among the completion times of each communication subtask. Further, the tail delay evaluation index is normalized with a preset target delay to obtain a delay deviation index. The delay deviation index can be expressed as the ratio of tail delay to target delay, which is used to characterize the degree of deviation of the current task execution delay from the target delay. In some embodiments, the time delay deviation index is expressed as a ratio: E = T_tail / T_ref Where T_tail represents the tail latency evaluation index, and T_ref represents the preset target latency. When E is greater than 1, it means that the latency exceeds the target requirement. When E is less than 1, it means that the latency meets or is better than the target requirement. The task granularity adjustment factor is updated based on the latency deviation index. When the tail latency is higher than the target latency, the task granularity adjustment factor is incrementally adjusted based on the preset first correction coefficient, and updated in the following manner: g_new = g_old + a × (E−1), Where 'a' is the first correction coefficient, used to control the magnitude of granularity increase, making task division tend towards coarse granularity, so as to reduce the number of subtasks and reduce scheduling overhead. When the tail latency is lower than the target latency, the task granularity adjustment factor is reduced based on the preset second correction coefficient, for example, updated in the following manner: g_new = g_old − b × (1 − E), Where b is the second correction coefficient, used to control the granularity refinement range, so that the task division tends to be finer-grained, thereby improving the parallel processing capability of the task. In some embodiments, the first correction coefficient and the second correction coefficient may be the same or different, and can be set according to the service type. For example, for latency-sensitive services, the first correction coefficient can be increased to accelerate the granularity convergence speed. Furthermore, after adjusting the task granularity adjustment factor using an incremental update method, boundary constraints are applied to the update result to ensure that the task granularity adjustment factor remains within a preset range, such as being limited to between 0 and 1, to guarantee the effectiveness of the granularity adjustment. Simultaneously, to reduce volatility during the granularity adjustment process, the update result can be smoothed, for example, by using exponential smoothing to merge the current update value with historical values. g_final=k×g_old+(1−k)×g_new, Where k is the smoothing coefficient, which controls the degree of integration between historical granularity and current adjustment results; In some embodiments, a maximum adjustment range limit can also be set. When the change in granularity adjustment factor between two consecutive updates exceeds a preset threshold, it is truncated to prevent abrupt changes in granularity from affecting system stability. For example, if the completion times of each subtask in a certain communication cycle are 120ms, 150ms, 180ms and 260ms respectively, then the tail latency evaluation index is 260ms. If the target latency is 200ms, then the latency deviation index E=1.3. When the current granularity adjustment factor is 0.4 and the first correction coefficient is 0.1, then the updated granularity adjustment factor is: g_new=0.4+0.1×(1.3−1)=0.43, which means that the task granularity is adjusted towards coarser granularity to reduce tail latency.

[0054] It should be noted that this application uses the tail delay evaluation index to provide feedback on the task execution results. When the maximum completion time of a subtask is detected to deviate from the target delay, the task granularity adjustment factor is adaptively updated, thereby forming a closed-loop optimization mechanism that enables the granularity adjustment to continuously approach the optimal scheduling state.

[0055] Furthermore, in another aspect of this application, in some embodiments, this application provides a wireless network scheduling optimization system for a cloud-edge collaborative network, the system including a communication allocation unit, referencing... Figure 3 The figure is a schematic diagram of the exemplary hardware and / or software structure of a communication distribution unit according to some embodiments of this application. The communication distribution unit 200 includes: a data acquisition module 201, a data analysis module 202, a task execution module 203, and a feedback adjustment module 204, which are described below: The data acquisition module 201 is used to acquire the communication requests and corresponding communication task data of the user equipment, construct task granularity representation parameters based on the communication task data, and generate the initial granularity range of the communication task data based on the task granularity representation parameters. Data analysis module 202 is used to collect network behavior data of communicable edge nodes within the initial granularity range within a preset historical communication period, and to construct a node behavior pattern matrix. The data analysis module 202 is also used to extract features based on the node behavior pattern matrix to obtain the link feature vector of the communicable edge node, perform feature learning on the link feature vector, and determine the corresponding task granularity adjustment factor within the current communication cycle. The task execution module 203 is used to dynamically adjust the initial granularity range of the communication task data according to the task granularity adjustment factor to obtain multiple communication sub-tasks, and execute the corresponding communication sub-task allocation strategy according to the network priority of the communication edge node. The feedback adjustment module 204 is used to construct a tail delay evaluation index based on the maximum completion time of each communication subtask, and adaptively update the task granularity adjustment factor based on the tail delay evaluation index. The foregoing has detailed an example of a wireless network scheduling optimization method and system for a cloud-edge collaborative network provided by embodiments of this application. It is understood that the corresponding device, in order to achieve the above functions, includes hardware structures and / or software modules corresponding to the execution of each function.

[0056] Those skilled in the art should readily recognize that, based on the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a certain function in the application is executed in a manner that drives hardware or computer software depends on the specific application and design constraints of the technical solution. Therefore, those skilled in the art can use different methods to implement the described function for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0057] In addition, this application also provides a computer terminal device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described wireless network scheduling optimization method for a cloud-edge collaborative network.

[0058] In some embodiments, reference Figure 4 The figure is a schematic diagram of the structure of a computer terminal device implementing a wireless network scheduling optimization method for a cloud-edge collaborative network according to some embodiments of this application. The wireless network scheduling optimization method for a cloud-edge collaborative network in the above embodiments can... Figure 4 The computer terminal device 300 shown is used to implement this, and the computer terminal device 300 includes at least one communication bus 301, communication interface 302, processor 303 and memory 304.

[0059] The processor 303 can be a general-purpose central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more devices used to control the execution of a wireless network scheduling optimization method for a cloud-edge collaborative network as described in this application.

[0060] The communication bus 301 may include a path for transmitting information between the aforementioned components.

[0061] Memory 304 may be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 304 may exist independently and be connected to processor 303 via communication bus 301. Memory 304 may also be integrated with processor 303.

[0062] The memory 304 stores program code for executing the scheme of this application, and its execution is controlled by the processor 303. The processor 303 executes the program code stored in the memory 304. The program code may include one or more software modules. In the above embodiments, the determination of the task granularity adjustment factor can be achieved by the processor 303 and one or more software modules in the program code in the memory 304.

[0063] Communication interface 302 uses any transceiver-like device for communicating with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area network (WLAN), etc.

[0064] Optionally, the computer terminal device 300 may also include a power supply 305 for providing power to various devices or circuits in the real-time computer terminal device.

[0065] In a specific implementation, as one example, a computer terminal device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0066] The aforementioned computer terminal device can be a general-purpose computer terminal device or a dedicated computer terminal device. In specific implementations, the computer terminal device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer terminal device.

[0067] In addition, other aspects of this application provide a computer-readable storage medium storing at least one computer program, which is loaded and executed by a processor to implement the operations performed by the above-described wireless network scheduling optimization method for a cloud-edge collaborative network.

[0068] In summary, the wireless network scheduling optimization method and system for cloud-edge collaborative networks disclosed in this application involves: acquiring communication requests and corresponding communication task data from user equipment; constructing task granularity representation parameters based on the communication task data; generating an initial granularity interval for the communication task data based on the task granularity representation parameters; collecting network behavior data of communicable edge nodes within the initial granularity interval to construct a node behavior pattern matrix; extracting features based on the node behavior pattern matrix to obtain link feature vectors of communicable edge nodes; performing feature learning on the link feature vectors to determine the corresponding task granularity adjustment factor within the current communication cycle; dynamically adjusting the initial granularity interval of the communication task data according to the task granularity adjustment factor to obtain multiple communication sub-tasks; and executing the corresponding communication sub-task allocation strategy according to the network priority of the communication edge nodes; constructing a tail latency evaluation index based on the maximum completion time of each communication sub-task; and adaptively updating the task granularity adjustment factor based on the tail latency evaluation index. This allows for adaptive granularity adjustment of communication tasks based on the node behavior patterns of the current edge nodes within the initial granularity interval, improving the accuracy and stability of wireless network scheduling.

[0069] The above descriptions are merely embodiments of this application, and common knowledge such as specific technical solutions or characteristics in the solutions are not described in detail here. It should be noted that those skilled in the art can make several modifications and improvements without departing from the technical solutions of this application, and these should also be considered within the scope of protection of this application, without affecting the effectiveness of the implementation of this application or the practicality of the patent.

[0070] The scope of protection claimed in this application shall be determined by the content of its claims. The specific embodiments described in the specification can be used to interpret the content of the claims. Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of the invention. Therefore, if these modifications and variations of this application fall within the scope of the claims of this application and their equivalents, this application also intends to include these modifications and variations.

Claims

1. A wireless network scheduling optimization method for cloud-edge collaborative networks, characterized in that, include: Obtain communication requests and corresponding communication task data from user devices, construct task granularity representation parameters based on the communication task data, and generate an initial granularity range for the communication task data based on the task granularity representation parameters. Within a preset historical communication period, network behavior data of communicable edge nodes are collected within the initial granularity range to construct a node behavior pattern matrix. Based on the node behavior pattern matrix, feature extraction is performed to obtain the link feature vector of the communicable edge node. Feature learning is then performed on the link feature vector to determine the corresponding task granularity adjustment factor within the current communication cycle. The initial granularity range of the communication task data is dynamically adjusted according to the task granularity adjustment factor to obtain multiple communication sub-tasks, and the corresponding communication sub-task allocation strategy is executed according to the network priority of the communication edge node. A tail delay evaluation index is constructed based on the maximum completion time of each communication subtask, and the task granularity adjustment factor is adaptively updated based on the tail delay evaluation index.

2. The method as described in claim 1, characterized in that, Collecting network behavior data of communicable edge nodes within the initial granularity range and constructing a node behavior pattern matrix specifically includes: Multiple candidate granularities are determined based on the initial granularity range. Within a preset historical communication period, network behavior data of each edge node at different time windows corresponding to the candidate granularities are collected. The network behavior data includes wireless link status information and resource status information of each edge node. The network behavior data is arranged according to the node dimension and the time dimension to construct a node behavior pattern matrix.

3. The method as described in claim 1, characterized in that, Based on the node behavior pattern matrix, feature extraction is performed to obtain the link feature vector of the communicable edge node, specifically including: Statistical analysis is performed on the node behavior pattern matrix in the time dimension to extract the behavioral statistical features corresponding to each candidate granularity, and the corresponding granularity statistical time series is formed according to the time sequence. Node statistical features within different time windows are obtained, and the overall behavior sequence of nodes is formed according to the window time sequence. Correlation analysis is performed between the granular statistical time series under different candidate granularity conditions and the overall behavior sequence of nodes to determine the behavior correlation degree under different candidate granularity conditions. Link feature vectors are formed based on the behavioral correlation under different candidate granularity conditions.

4. The method as described in claim 1, characterized in that, The specific steps of performing feature learning on the link feature vector to determine the corresponding task granularity adjustment factor within the current communication cycle include: using a pre-trained convolutional neural network model as the link state discrimination model, classifying the link feature vector, and determining the corresponding task granularity adjustment factor based on the classification results.

5. The method as described in claim 1, characterized in that, The initial granularity range for generating the communication task data based on the task granularity characterization parameters specifically includes: After normalizing the task granularity representation parameters, the granularity benchmark value of the communication task data is generated by parameter weighting. The granularity reference value is expanded according to the preset interval length to obtain the initial granularity interval of the communication task.

6. The method as described in claim 1, characterized in that, Constructing task granularity representation parameters based on the communication task data specifically includes: extracting features from the communication task data to obtain task granularity representation parameters used to represent task characteristics. The task granularity representation parameters include at least data volume parameters, computational complexity parameters, latency sensitivity parameters, and service quality level parameters.

7. The method as described in claim 1, characterized in that, During the process of obtaining communication requests and corresponding communication task data from user equipment, the communication requests and corresponding communication task data are derived from reports submitted by the user equipment.

8. A wireless network scheduling and optimization system for a cloud-edge collaborative network, comprising a communication allocation unit, wherein the communication allocation unit is used to execute the intelligent switching control method for a home energy storage system as described in any one of claims 1 to 7, characterized in that, The communication allocation unit includes: The data acquisition module is used to acquire communication requests and corresponding communication task data of user devices, construct task granularity representation parameters based on the communication task data, and generate an initial granularity range of the communication task data based on the task granularity representation parameters. The data analysis module is used to collect network behavior data of communicable edge nodes within the initial granularity range within a preset historical communication period, and to construct a node behavior pattern matrix. The data analysis module is also used to extract features based on the node behavior pattern matrix to obtain the link feature vector of the communicable edge node, perform feature learning on the link feature vector, and determine the corresponding task granularity adjustment factor within the current communication cycle. The task execution module is used to dynamically adjust the initial granularity range of the communication task data according to the task granularity adjustment factor to obtain multiple communication sub-tasks, and execute the corresponding communication sub-task allocation strategy according to the network priority of the communication edge node. The feedback adjustment module is used to construct a tail delay evaluation index based on the maximum completion time of each communication subtask, and to adaptively update the task granularity adjustment factor based on the tail delay evaluation index.

9. A computer terminal device, characterized in that, The computer terminal device includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute a wireless network scheduling optimization method for a cloud-edge collaborative network as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing at least one computer program, characterized in that, The computer program is loaded and executed by a processor to perform the operations described in any one of claims 1 to 7 for the wireless network scheduling optimization method of a cloud-edge collaborative network.

Images