A Cloud-Edge Collaborative Intelligent Computing Cluster Task Allocation Method and System

Abstract

This invention discloses a method and system for task allocation in an intelligent computing power cluster based on cloud-edge collaboration, relating to the field of computing power task allocation technology. The method includes: extracting a first task from the tasks to be allocated and collecting first multi-dimensional feature parameters of the first task to obtain a first feature curve; performing cluster analysis on the tasks to be allocated to obtain clustering results; determining whether the distributed edge computing power in the cloud-edge computing power cluster meets the first resource requirement; if it does, comparing and analyzing to obtain the first fitness of the first cluster; sorting the computing power nodes in descending order to obtain the target computing power node; and processing the tasks of the first cluster. This invention solves the technical problems of low task allocation efficiency and poor task-computing node compatibility in the prior art under cloud-edge collaborative environments, achieving efficient task allocation in intelligent computing power clusters under cloud-edge collaborative environments and improving the adaptability and efficiency of task processing.

Description

Technical Field

[0001] This invention relates to the field of computing power task allocation technology, specifically to a method and system for intelligent computing power cluster task allocation based on cloud-edge collaboration. Background Technology

[0002] In today's digital age, the demand for computing power from various applications is exploding, giving rise to cloud-edge collaborative intelligent computing clusters. However, existing task allocation methods have many drawbacks. On the one hand, traditional methods struggle to accurately extract the multidimensional features of tasks, making it impossible to perform reasonable clustering based on task characteristics. Task allocation is often arbitrary, as the inability to effectively distinguish task characteristics frequently leads to tasks with vastly different levels of complexity being incorrectly grouped together and assigned to mismatched computing nodes, resulting in resource waste and inefficiency. On the other hand, the assessment and scheduling of edge computing resources lack dynamism and precision. When the load on nodes in the edge computing cluster is uneven or resources fluctuate, existing methods cannot promptly and accurately determine and meet the resource requirements of tasks, causing task processing delays and impacting overall system performance.

[0003] Existing technologies suffer from low task allocation efficiency and poor compatibility between tasks and computing nodes in cloud-edge collaborative environments. Summary of the Invention

[0004] This application provides a method and system for task allocation in intelligent computing power clusters based on cloud-edge collaboration, which is used to address the technical problems of low task allocation efficiency and poor adaptability between tasks and computing power nodes in the existing cloud-edge collaboration environment.

[0005] In view of the above problems, this application provides a method and system for task allocation of intelligent computing power clusters based on cloud-edge collaboration.

[0006] The first aspect of this application provides a method for task allocation in an intelligent computing cluster based on cloud-edge collaboration, the method comprising:

[0007] Extract the first task from the tasks to be assigned, and collect the first multidimensional feature parameters of the first task to obtain the first feature curve; perform cluster analysis on the tasks to be assigned based on the first feature curve to obtain the clustering result, wherein the clustering result includes a first cluster; analyze the first resource requirement of the first cluster and determine whether the distributed edge computing power in the cloud-edge computing power cluster meets the first resource requirement; if it does, compare and analyze the first fitness of the first computing power node in the distributed edge computing power to the first cluster; sort the computing power nodes in descending order based on the first fitness to obtain the target computing power node corresponding to the highest fitness; process the tasks of the first cluster through the target computing power node.

[0008] A second aspect of this application provides a cloud-edge collaborative intelligent computing cluster task allocation system, the system comprising:

[0009] A first feature curve acquisition module is used to extract a first task from the tasks to be assigned and collect the first multidimensional feature parameters of the first task to obtain a first feature curve; a clustering result acquisition module is used to perform clustering analysis on the tasks to be assigned based on the first feature curve to obtain clustering results, wherein the clustering results include a first cluster; a first resource requirement analysis module is used to analyze and obtain the first resource requirement of the first cluster and determine whether the distributed edge computing power in the cloud-edge computing power cluster meets the first resource requirement; a first fitness acquisition module is used to compare and analyze to obtain the first fitness of the first computing power node in the distributed edge computing power to the first cluster if it meets the requirement; a target computing power node acquisition module is used to sort the computing power nodes in descending order based on the first fitness to obtain the target computing power node corresponding to the highest fitness; and a task processing module is used to process the tasks of the first cluster through the target computing power node.

[0010] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0011] The process involves extracting the first task from the tasks to be assigned and collecting its first multidimensional feature parameters to obtain a first feature curve. Cluster analysis is then performed on the tasks to be assigned to obtain clustering results. The first resource requirement of the first cluster is analyzed, and it is determined whether the distributed edge computing power in the cloud-edge computing power cluster meets the first resource requirement. If it does, a comparative analysis is conducted to obtain the first fitness of the first computing power node in the distributed edge computing power to the first cluster. The computing power nodes are then sorted in descending order based on the first fitness to obtain the target computing power node corresponding to the highest fitness. The task processing for the first cluster is then performed through the target computing power node. This achieves efficient task allocation in an intelligent computing power cluster under a cloud-edge collaborative environment, improving the adaptability and efficiency of task processing. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in 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.

[0013] Figure 1 A schematic diagram of the task allocation method for intelligent computing power clusters based on cloud-edge collaboration provided in this application embodiment;

[0014] Figure 2 This is a schematic diagram of the structure of a cloud-edge collaborative intelligent computing cluster task allocation system provided in an embodiment of this application.

[0015] Explanation of reference numerals in the attached figures: First feature curve acquisition module 10, clustering result acquisition module 20, first resource demand analysis module 30, first fitness acquisition module 40, target computing power node acquisition module 50, task processing module 60. Detailed Implementation

[0016] This application provides a task allocation method and system for intelligent computing power clusters based on cloud-edge collaboration, which is used to address the technical problems of low task allocation efficiency and poor adaptability between tasks and computing power nodes in the existing cloud-edge collaborative environment.

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] Example 1, as Figure 1 As shown, this application provides a method for task allocation in an intelligent computing cluster based on cloud-edge collaboration, the method comprising:

[0019] Step S100: Extract the first task from the tasks to be assigned, and collect the first multidimensional feature parameters of the first task to obtain the first feature curve.

[0020] Specifically, after extracting the first task from the tasks to be assigned, information is collected on the first task based on predetermined task indicators (including task type, resource requirements, data scale, and expected execution time) to obtain the first multidimensional feature parameters. Then, a first scatter plot is drawn based on these first multidimensional feature parameters, and the scatter plot is randomly sampled to obtain the first scatter sample. The sample is subjected to polynomial fitting analysis to obtain the first fitting curve. The original scatter plot is then compared with the sample to obtain the first verification scatter point. The first spatial distance from the verification scatter point to the first fitting curve is calculated. If the distance is within a predetermined distance threshold, the first fitting curve is determined as the first feature curve. If it is not within the threshold, the above sampling, fitting, and verification process is repeated until the first feature curve is obtained.

[0021] In one possible implementation, step S100 further includes:

[0022] Step S110: Retrieve predetermined task indicators to collect the first multidimensional feature parameters for the first task; wherein, the predetermined task indicators include task type, resource requirements, data scale, and expected execution time.

[0023] Specifically, when collecting the first multidimensional feature parameters of the first task, information collection for the first task is carried out by retrieving a pre-defined task indicator system. This pre-defined task indicator system clearly includes four core dimensions: task type (such as different task attributes like data processing and computational analysis), resource requirements (covering the hardware resource specifications such as computing power, storage, and memory required to complete the task), data scale (the size of data dimensions such as the amount of raw data and intermediate processing data involved in the task), and expected execution time (the upper limit or target duration of the task completion time required by the task initiator). By comprehensively extracting and organizing the specific information of the first task in these four dimensions, the first multidimensional feature parameters that can fully characterize the features of the task are finally formed, providing basic data support for the subsequent construction of the first feature curve.

[0024] Step S200: Perform cluster analysis on the task to be assigned based on the first feature curve to obtain cluster results, wherein the cluster results include a first cluster.

[0025] Specifically, when performing cluster analysis on the tasks to be assigned based on the first feature curve, the second task is first extracted from the tasks to be assigned, its second multidimensional feature parameters are collected and a second scatter plot is drawn, the first scatter sample is obtained by random sampling and then a first fitting curve is obtained by polynomial fitting, and the second scatter plot is compared with the first scatter sample to obtain the first verification scatter point, and the first spatial distance from it to the first fitting curve is calculated. If it is within a predetermined distance threshold, the curve is used as the second feature curve (if it is not within the threshold, the sampling to fitting process is repeated until the second feature curve is obtained); then the first feature curve and the second feature curve are translated. When the translation result meets the predetermined constraints, the target distance value between the two is calculated. If the value is less than the predetermined distance limit, the first task and the second task are determined to belong to the same cluster. This process is repeated to complete the clustering of all tasks to be assigned, and finally a clustering result containing the first cluster is formed.

[0026] In one possible implementation, step S200 further includes:

[0027] Step S210: Extract the second task from the tasks to be assigned and obtain the second characteristic curve of the second task.

[0028] Step S220: Translate the first feature curve and the second feature curve to obtain the translation result.

[0029] Step S230: When the translation result meets the predetermined constraints, calculate the target distance value between the first feature curve and the second feature curve.

[0030] Step S240: If the target distance value is less than the predetermined distance limit, then the first task and the second task belong to the same cluster, forming the clustering result.

[0031] Specifically, after extracting the second task from the tasks to be assigned, the second multidimensional feature parameters of the second task are first collected, and then a second scatter plot is drawn based on these parameters. Subsequently, the second scatter plot is randomly sampled to obtain the first scatter sample, and a polynomial fitting analysis is performed on the sample to obtain the first fitting curve. Then, the second scatter plot is compared with the first scatter sample to obtain the first verification scatter point, and the first spatial distance from the first verification scatter point to the first fitting curve is calculated. If the spatial distance is within a predetermined distance threshold, the first fitting curve is used as the second feature curve. If it is not within the threshold, the process of random sampling, polynomial fitting, obtaining verification scatter points and calculating spatial distance is repeated until the second feature curve is obtained.

[0032] When translating the first and second characteristic curves, the positions of the two curves are adjusted along the coordinate axes (such as the time axis, feature parameter axis, etc.) within a preset coordinate space. This ensures that the starting feature points, key inflection points, or trend baselines of the curves are aligned with a preset alignment reference system, thus obtaining the relative spatial distribution of the two curves after position calibration, i.e., the translation result. This process aims to eliminate non-essential positional offsets caused by differences in the initial data acquisition baseline, laying the foundation for subsequent judgments on whether the translation result meets predetermined constraints (such as consistency of characteristic trend direction, overlap of key intervals, etc.).

[0033] When the translation results of the first and second feature curves meet the predetermined constraints (i.e., the overlap between the two curves reaches its maximum), the specific method for calculating the target distance value is as follows: In the unified coordinate reference system formed after translation processing, all corresponding feature points of the two curves (including the starting point, inflection point, ending point of the curve, and key parameter mutation points, etc.) are traversed, and the spatial distance between each corresponding point is calculated respectively; the distance value with the largest value is selected from these spatial distances, and this maximum distance value is the target distance value between the first and second feature curves. This quantifies the maximum feature difference between the two curves in the state of maximum overlap, providing an accurate numerical basis for subsequent judgment on whether the first task and the second task belong to the same cluster.

[0034] After completing the translation process of the first and second feature curves and ensuring that the translation results meet predetermined constraints (such as maximizing the overlap between the two curves), the target distance value between the two curves (such as the spatial distance at the maximum distance) is calculated. If the target distance value is less than a preset distance limit, it indicates that the differences in multidimensional feature parameters between the first and second tasks represented by the two curves are within an acceptable range, and they belong to similar task types. Therefore, the first and second tasks are determined to belong to the same cluster. By comparing and classifying similar tasks among all tasks to be assigned, a clustering result including this cluster is finally formed.

[0035] In one possible implementation, step S210 further includes:

[0036] Step S211: Collect the second multidimensional feature parameters of the second task.

[0037] Step S212: Draw a second scatter plot based on the second multidimensional feature parameters; perform curve fitting analysis on the second scatter plot to obtain the second feature curve.

[0038] This includes:

[0039] a: Randomly sample the second scatter plot to obtain the first scatter sample.

[0040] b: Perform polynomial fitting analysis on the first scatter sample to obtain the first fitting curve.

[0041] c: Compare the second scatter plot with the first scatter sample to obtain the first verification scatter plot.

[0042] d: Calculate the first spatial distance from the first verification scatter point to the first fitted curve.

[0043] e: Determine whether the first spatial distance is within a predetermined distance threshold.

[0044] f: If it is in the range, then the first fitted curve is used as the second feature curve.

[0045] Specifically, the collection of the second multidimensional feature parameters of the second task is achieved by retrieving predetermined task indicators to collect information about the second task. These predetermined task indicators explicitly include several core dimensions: task type, resource requirements, data scale, and expected execution time. By comprehensively extracting and organizing the specific information of the second task across these four dimensions, a second multidimensional feature parameter that can fully characterize the task's features is ultimately formed.

[0046] Based on the collected second multidimensional feature parameters, a second scatter plot is plotted in the corresponding coordinate system, with each scatter point corresponding to a set of feature parameter values. Then, curve fitting analysis is performed on this second scatter plot to obtain the second feature curve. The specific process is as follows:

[0047] a: When randomly sampling the second scatter plot to obtain the first scatter plot sample, a portion of the scatter plots are randomly selected from all the scatter plots contained in the second scatter plot. These selected scatter plots together form the first scatter plot sample. The number of samples can be determined according to the size of the scatter plot and the needs of subsequent fitting analysis, so as to provide basic data for subsequent polynomial fitting analysis.

[0048] b: The specific implementation method for performing polynomial fitting analysis on the first scattered sample to obtain the first fitting curve is as follows: First, determine the order of the polynomial used for fitting, which can be selected according to the distribution complexity of the first scattered sample; then, substitute the coordinate values ​​of each scattered point in the first scattered sample into the preset polynomial equation, and calculate the coefficients of each term of the polynomial using the least squares method, so that the sum of squares of the deviations between the curve corresponding to the polynomial and each scattered point in the first scattered sample is minimized; the final polynomial curve obtained is the first fitting curve, which can approximately reflect the overall distribution trend of the first scattered sample.

[0049] c: Compare all the scatter points in the first scatter sample with the scatter points in the second scatter plot one by one, and filter out the scatter points in the second scatter plot that are not included in the first scatter sample. These unselected scatter points are the first verification scatter points. In this way, the independent data points used to verify the accuracy of the fitted curve can be clearly identified.

[0050] d: When calculating the first spatial distance from the first verification scatter point to the first fitted curve, for each first verification scatter point, its specific coordinate data in the coordinate system is used, combined with the polynomial equation of the first fitted curve, and the distance from point to curve is calculated using the formula to obtain the shortest spatial distance from each verification scatter point to the fitted curve. This distance is the first spatial distance, which is used to measure the degree of deviation between the verification scatter point and the fitted curve.

[0051] e: When determining whether the first spatial distance is within the predetermined distance threshold, the first spatial distance corresponding to each first verification scatter point is calculated and compared with the predetermined distance threshold. The spatial distances are checked to see if they are within the range defined by the threshold. This is to verify whether the fitting effect of the first fitting curve on the second scatter plot meets the requirements.

[0052] f: When it is determined that the first spatial distance is within the predetermined distance threshold, it indicates that the first fitting curve can better fit the overall distribution characteristics of the second scatter plot and can accurately reflect the changing trend of the second multidimensional feature parameters of the second task. Therefore, the first fitting curve is determined as the second feature curve of the second task.

[0053] In one possible implementation, step S210 further includes:

[0054] After step e, if the condition is not met, steps a to f are repeated until the second characteristic curve is obtained.

[0055] Specifically, after determining in step e that the first spatial distance is not within the predetermined distance threshold, it indicates that the current first fitted curve fails to adequately match the overall distribution characteristics of the second scatter plot and cannot accurately reflect the changing trend of the second multidimensional feature parameters of the second task. At this point, it is necessary to restart from step a: randomly sample the second scatter plot again to obtain new first scatter samples; perform polynomial fitting analysis based on the new samples to obtain a new first fitted curve; then compare the second scatter plot with the new samples to obtain new first verification scatter points; calculate the first spatial distance from the new verification scatter points to the new fitted curve and determine whether it is within the predetermined distance threshold; if it is still not within the predetermined distance threshold, continue repeating steps a to f until the obtained first spatial distance is within the predetermined distance threshold, and then determine the first fitted curve at this point as the second feature curve.

[0056] Step S300: Analyze and obtain the first resource requirement of the first cluster, and determine whether the distributed edge computing power in the cloud-edge computing power cluster meets the first resource requirement.

[0057] Specifically, the multidimensional feature parameters of all tasks within the first cluster are traversed to extract indicators such as resource requirements and data scale. The total amount of computing resources (such as the total number of CPU cores required), storage resources (such as the total data scale), and network resources (such as the total data transmission bandwidth requirements) of the cluster are calculated using a summation algorithm, and this is used as the first resource requirement. Then, the resource query algorithm of the distributed edge computing power is called to collect the real-time available resource data of each edge node (including the number of remaining CPU cores, remaining storage space, and current available bandwidth). The total available resources of the distributed edge computing power are obtained through a summarization algorithm. Finally, a comparison algorithm is used to compare the first resource requirement with the total available resources of the distributed edge computing power item by item according to the corresponding dimensions. If the available resources in all dimensions are greater than or equal to the required resources, it is determined that the requirement is met; otherwise, it is determined that the requirement is not met. If the requirement is not met, the cloud computing power call algorithm is triggered for remote processing.

[0058] In one possible implementation, step S300 further includes:

[0059] Step S310: After determining whether the distributed edge computing power in the cloud-edge computing power cluster meets the first resource requirement, if it does not meet the requirement, remote processing of the first cluster is performed on the cloud computing power in the cloud-edge computing power cluster.

[0060] Specifically, after determining that the distributed edge computing power in the cloud-edge computing power cluster cannot meet the first resource requirements of the first cluster, the cloud computing power invocation mechanism is activated. The task information of the first cluster (including task type, data scale, resource requirements, etc.) is packaged and transmitted to the cloud computing power nodes in the cloud-edge computing power cluster. After receiving the task information, the cloud computing power nodes schedule and process the tasks for the first cluster according to their own computing power resource allocation. The tasks are executed through remote data transmission and computation interaction, and the processing results are fed back to the corresponding edge nodes or request sources, thereby achieving effective processing of tasks that the edge computing power cannot handle.

[0061] Step S400: If satisfied, compare and analyze to obtain the first fitness of the first computing node in the distributed edge computing power to the first cluster.

[0062] Specifically, firstly, a simulation environment is constructed for the first computing node to process the first cluster. The task information of the first cluster (such as task type, resource requirements, data scale, etc.) is input into this environment for simulation processing, generating first simulation information including processing time, resource utilization, network transmission efficiency, etc. Then, predetermined processing indicators including real-time load and network bandwidth are retrieved, and a weighted scoring method is used to evaluate the first simulation information. For example, corresponding weights are assigned according to the redundancy of real-time load and the utilization rate of network bandwidth, and a comprehensive score is calculated. This score is the first fitness of the first computing node to the first cluster, thereby quantifying the degree of adaptability of the node to the task of processing the cluster.

[0063] In one possible implementation, step S400 further includes:

[0064] Step S410: Perform the processing simulation of the first cluster on the first computing node to obtain the first simulation information.

[0065] Step S420: Retrieve predetermined processing indicators to perform a traversal analysis on the first simulation information to obtain the first fitness.

[0066] Step S430: Wherein, the predetermined processing indicators include real-time load and network bandwidth.

[0067] Specifically, a simulation platform is built to simulate the operating environment of the first computing power node. The task data of the first cluster (including information such as task type, resource requirements, and data scale) is input into the platform to simulate the entire process of the first computing power node receiving tasks, allocating computing resources (such as CPU and memory), performing data processing operations, and conducting network data interaction. During the simulation, key data such as task processing time, resource utilization rate at each moment, data transmission rate, network latency, and task processing success rate are recorded in real time through built-in monitoring tools. These data are summarized to form the first simulation information, providing raw data support for subsequent evaluation of the adaptability of the first computing power node to the first cluster.

[0068] The system retrieves predetermined processing metrics, including real-time load and network bandwidth. Then, a traversal algorithm is used to check each data item in the first simulation information. For the real-time load metric, the peak load, average load, and load fluctuation of the first computing node during the simulation are analyzed. For the network bandwidth metric, the bandwidth utilization, peak bandwidth, and bandwidth stability during data transmission are evaluated. By comparing these analysis results with the standard values ​​of the predetermined processing metrics, a comprehensive score is obtained using a weighted calculation method. This score is the first fitness, which quantifies the degree of adaptation of the first computing node to the processing of the first cluster.

[0069] The predetermined processing metrics used to analyze the first fitness of the first computing node for the first cluster specifically include real-time load and network bandwidth. Real-time load refers to the workload undertaken by the first computing node per unit time during the processing of the first cluster task, reflecting the node's busyness and resource constraints. Network bandwidth is the amount of data that the first computing node can transmit per unit time when interacting with other nodes, reflecting the data transmission capability and efficiency between nodes. These two metrics together serve as key criteria for evaluating the fitness of the first computing node.

[0070] Step S500: Based on the first fitness, sort the computing power nodes in descending order to obtain the target computing power node corresponding to the highest fitness.

[0071] Specifically, after obtaining the first fitness of each first computing power node in the distributed edge computing power to the first cluster, all computing power nodes participating in the evaluation are sorted in descending order based on these first fitness values, that is, in order of fitness from high to low. The computing power node with the highest fitness at the top is the target computing power node that is most suitable for the first cluster, and this node will be responsible for the task processing of the first cluster.

[0072] Step S600: Perform task processing for the first cluster through the target computing power node.

[0073] Specifically, after determining the target computing power node corresponding to the highest fitness, the task information of the first cluster (including task type, data size, resource requirements, etc.) is transmitted to the target computing power node, which then performs specific operations according to the preset task processing flow. The target computing power node will reasonably schedule and execute the tasks within the first cluster based on its own resource configuration and processing capabilities to ensure that the tasks are completed efficiently.

[0074] In one possible implementation, step S600 further includes:

[0075] Step S610: Perform continuous task processing monitoring on the first cluster to obtain the first monitoring record.

[0076] Step S620: Analyze the first monitoring record to obtain the first processing progress time sequence of the first cluster.

[0077] Step S630: Perform predictive analysis on the first processing progress time sequence to obtain the predicted time when the progress is completed.

[0078] Step S640: Obtain the first prediction delay of the first cluster based on the prediction time.

[0079] Step S650: If the first predicted delay does not meet the predetermined delay threshold, then the computing power node migration adjustment is performed on the first cluster.

[0080] Specifically, during the task processing of the first cluster by the target computing power node, a continuous monitoring mechanism is activated to continuously track various dynamic data of task processing, including but not limited to the real-time processing status of each subtask, the consumed computing resources (such as CPU utilization and memory usage), the real-time data transmission rate, the phased results of task processing, and whether abnormal interruptions occur. These real-time collected information are organized and recorded in chronological order to form a first monitoring record that comprehensively reflects the task processing process of the first cluster.

[0081] The first monitoring record is systematically analyzed to extract key data related to the processing progress of the first cluster task, such as the amount of tasks completed at different time points and changes in processing rate. Then, these data are organized and arranged in chronological order to form a first processing progress time sequence that clearly reflects the processing progress of the first cluster task from the beginning to the current moment. This time sequence can intuitively show the rhythm and completion trend of task processing.

[0082] A Long Short-Term Memory (LSTM) network is used to predict and analyze the time series of the first processing progress. First, the time series of the first processing progress is divided into training samples and test samples according to the time step. The sample data is normalized to eliminate the influence of units. Then, an LSTM model is constructed, and appropriate numbers of hidden layer neurons, time steps, and iterations are set. The model is trained using training samples, and the model parameters are optimized through the backpropagation algorithm so that the model can learn the pattern of processing progress changes over time. Then, the prediction accuracy of the model is verified using test samples. If the accuracy meets the standard, the model is used to predict the future processing progress. Based on the predicted progress trend, the predicted time when all tasks of the first clustering cluster are completed is calculated.

[0083] Using the start time of the first cluster task as a benchmark, the difference between the predicted completion time and the benchmark time is calculated. The result is the first predicted delay of the first cluster. This delay directly reflects the length of time required from the start of the task to its expected completion, providing a quantitative basis for subsequent judgment on whether the task can be completed within the specified time.

[0084] When the first predicted latency does not meet the predetermined latency threshold, the computing power node migration and adjustment mechanism will be activated. First, the processing bottleneck of the current target computing power node and the resource status and adaptability of other surrounding computing power nodes will be assessed. Then, the task data and processing status information of the first cluster will be migrated to a more suitable computing power node (such as a node with higher adaptability, lower load, or better network conditions). At the same time, the task processing association information will be updated to ensure the integrity of the data and the continuity of task processing during the migration process, so as to shorten the task completion latency and make it meet the predetermined latency threshold requirements.

[0085] When the first predicted latency does not meet the predetermined latency threshold, the current target computing power node's task processing for the first cluster is first paused, and the current processing status and completed data of the task are recorded. Then, the real-time load, network bandwidth, and fitness of other nodes in the distributed edge computing power are re-evaluated, and candidate computing power nodes with high fitness and meeting the latency requirements are selected. Then, the task data and processing status information of the first cluster are transmitted to the selected candidate node through the data migration protocol to complete the data synchronization. Finally, the candidate node takes over the task processing and continues to execute the remaining tasks of the first cluster to ensure that the task can be completed within the predetermined latency threshold.

[0086] In one possible implementation, step S650 further includes:

[0087] Step S651: If the first predicted delay meets the predetermined delay threshold, a task processing evaluation strategy is introduced to evaluate the first monitoring record to obtain the first processing fitness.

[0088] Step S652: Take the average value of the first processing fitness as the comprehensive fitness, wherein the comprehensive fitness is used to characterize the overall allocation effect of the task to be assigned.

[0089] Specifically, when the first predicted latency meets the predetermined latency threshold, a task processing evaluation strategy is introduced that includes task execution efficiency, resource utilization, and energy consumption. Data such as task processing time and completion amount are extracted from the first monitoring record to calculate task execution efficiency. The usage ratio of resources such as CPU and memory is statistically analyzed to evaluate resource utilization. Energy consumption data during the processing is recorded. Weights are assigned to these three indicators, and the evaluation results of each indicator are weighted and calculated. The resulting comprehensive score is the first processing fitness, which quantifies the overall performance of the current task processing.

[0090] After obtaining the first processing fitness, all data points of that fitness are counted, and their arithmetic mean is calculated by summing them and dividing by the number of data points. This average value is defined as the comprehensive fitness. The comprehensive fitness integrates the multi-dimensional evaluation results in the task processing process, and can comprehensively reflect the degree of adaptation and processing effect of the task to be assigned in the entire process from feature extraction, cluster analysis to computing node allocation and processing, providing a quantitative reference for the optimization of subsequent task allocation strategies.

[0091] Example 2 is based on the same inventive concept as the cloud-edge collaborative intelligent computing cluster task allocation method in the previous examples, such as... Figure 2 As shown, this application provides a cloud-edge collaborative intelligent computing cluster task allocation system. The system and method embodiments in this application are based on the same inventive concept. The system includes:

[0092] The first feature curve acquisition module 10 is used to extract the first task from the tasks to be assigned, and collect the first multidimensional feature parameters of the first task to obtain the first feature curve.

[0093] The clustering result acquisition module 20 is used to perform clustering analysis on the task to be assigned based on the first feature curve to obtain clustering results, wherein the clustering results include a first cluster.

[0094] The first resource requirement analysis module 30 is used to analyze and obtain the first resource requirement of the first cluster and determine whether the distributed edge computing power in the cloud-edge computing power cluster meets the first resource requirement.

[0095] The first fitness acquisition module 40 is used to obtain the first fitness of the first computing node in the distributed edge computing power to the first cluster if the condition is met.

[0096] The target computing node acquisition module 50 is used to sort the computing nodes in descending order based on the first fitness to obtain the target computing node corresponding to the highest fitness.

[0097] The task processing module 60 is used to perform task processing for the first cluster through the target computing power node.

[0098] Furthermore, the system is also used to implement the following functions:

[0099] The first multidimensional feature parameters are obtained by retrieving predetermined task indicators to collect data on the first task; wherein, the predetermined task indicators include task type, resource requirements, data scale, and expected execution time.

[0100] Furthermore, the system is also used to implement the following functions:

[0101] Extract the second task from the tasks to be assigned and obtain the second feature curve of the second task; perform translation processing on the first feature curve and the second feature curve to obtain the translation result; when the translation result meets the predetermined constraints, calculate the target distance value between the first feature curve and the second feature curve; if the target distance value is less than the predetermined distance value limit, then the first task and the second task belong to the same cluster, forming the clustering result.

[0102] Furthermore, the system is also used to implement the following functions:

[0103] The process involves: collecting second multidimensional feature parameters for the second task; plotting a second scatter plot based on the second multidimensional feature parameters; performing curve fitting analysis on the second scatter plot to obtain the second feature curve; including: a) randomly sampling the second scatter plot to obtain a first scatter sample; b) performing polynomial fitting analysis on the first scatter sample to obtain a first fitting curve; c) comparing the second scatter plot with the first scatter sample to obtain a first verification scatter point; d) calculating the first spatial distance from the first verification scatter point to the first fitting curve; e) determining whether the first spatial distance is within a predetermined distance threshold; f) if it is, using the first fitting curve as the second feature curve.

[0104] Furthermore, the system is also used to implement the following functions:

[0105] After step e, if the condition is not met, steps a to f are repeated until the second characteristic curve is obtained.

[0106] Furthermore, the system is also used to implement the following functions:

[0107] After determining whether the distributed edge computing power in the cloud-edge computing power cluster meets the first resource requirement, if it does not, the first cluster is remotely processed by the cloud computing power in the cloud-edge computing power cluster.

[0108] Furthermore, the system is also used to implement the following functions:

[0109] The first computing node is subjected to the first cluster processing simulation to obtain the first simulation information; the first simulation information is traversed and analyzed by retrieving predetermined processing indicators to obtain the first fitness; wherein, the predetermined processing indicators include real-time load and network bandwidth.

[0110] Furthermore, the system is also used to implement the following functions:

[0111] The first cluster is continuously monitored for task processing to obtain a first monitoring record; the first monitoring record is analyzed to obtain a first processing progress time sequence of the first cluster; the first processing progress time sequence is predicted and analyzed to obtain a predicted time when the progress is completed; a first predicted delay of the first cluster is obtained based on the predicted time; if the first predicted delay does not meet a predetermined delay threshold, the computing power node migration adjustment is performed on the first cluster.

[0112] Furthermore, the system is also used to implement the following functions:

[0113] If the first predicted delay meets the predetermined delay threshold, a task processing evaluation strategy is introduced to evaluate the first monitoring record to obtain a first processing fitness; the average value of the first processing fitness is taken as the comprehensive fitness, wherein the comprehensive fitness is used to characterize the overall allocation effect of the task to be assigned.

[0114] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0115] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0116] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A task allocation method for intelligent computing power clusters based on cloud-edge collaboration, characterized in that, include: Extract the first task from the tasks to be assigned, and collect the first multidimensional feature parameters of the first task to obtain the first feature curve; Cluster analysis is performed on the tasks to be assigned based on the first feature curve to obtain cluster results, wherein the cluster results include a first cluster. The analysis yields the first resource requirement of the first cluster, and it is determined whether the distributed edge computing power in the cloud-edge computing power cluster meets the first resource requirement. If satisfied, a comparative analysis is performed to obtain the first fitness of the first computing power node in the distributed edge computing power to the first cluster. The computing power nodes are sorted in descending order based on the first fitness to obtain the target computing power node corresponding to the highest fitness. The first cluster is processed through the target computing power node; The clustering analysis of the tasks to be assigned based on the first feature curve yields the following clustering results: Extract the second task from the tasks to be assigned, and obtain the second characteristic curve of the second task; The first feature curve and the second feature curve are translated to obtain the translation result. When the translation result meets the predetermined constraints, the target distance value between the first feature curve and the second feature curve is calculated; If the target distance value is less than the predetermined distance limit, then the first task and the second task belong to the same cluster, forming the clustering result; The comparative analysis yields the first fitness of the first computing node in the distributed edge computing power to the first cluster, including: The first computing node is processed and simulated to obtain the first cluster information. The first simulation information is analyzed by retrieving predetermined processing indicators to obtain the first fitness. The predetermined processing metrics include real-time load and network bandwidth.

2. The intelligent computing power cluster task allocation method based on cloud-edge collaboration as described in claim 1, characterized in that, Extract the first task from the tasks to be assigned, and collect the first multidimensional feature parameters of the first task, including: The first multidimensional feature parameters are obtained by retrieving predetermined task indicators to collect data for the first task. The predetermined task indicators include task type, resource requirements, data scale, and expected execution time.

3. The intelligent computing power cluster task allocation method based on cloud-edge collaboration as described in claim 1, characterized in that, Extract the second task from the tasks to be assigned, and obtain the second feature curve of the second task, including: The second multidimensional feature parameters of the second task are collected; A second scatter plot is drawn based on the second multidimensional feature parameters; The second characteristic curve is obtained by performing curve fitting analysis on the second scatter plot; This includes: a: Randomly sample the second scatter plot to obtain the first scatter plot sample; b: Perform polynomial fitting analysis on the first scatter sample to obtain the first fitting curve; c: Compare the second scatter plot with the first scatter sample to obtain the first verification scatter point; d: Calculate the first spatial distance from the first verification scatter point to the first fitted curve; e: Determine whether the first spatial distance is within a predetermined distance threshold; f: If it is in the range, then the first fitted curve is used as the second feature curve.

4. The intelligent computing power cluster task allocation method based on cloud-edge collaboration as described in claim 3, characterized in that, After step e, if the condition is not met, steps a to f are repeated until the second characteristic curve is obtained.

5. The intelligent computing power cluster task allocation method based on cloud-edge collaboration as described in claim 1, characterized in that, After determining whether the distributed edge computing power in the cloud-edge computing power cluster meets the first resource requirement, if it does not, the first cluster is remotely processed by the cloud computing power in the cloud-edge computing power cluster.

6. The intelligent computing power cluster task allocation method based on cloud-edge collaboration as described in claim 1, characterized in that, After processing the first cluster task through the target computing node, the process further includes: The first cluster is continuously monitored and processed to obtain the first monitoring record; The first processing progress timeline of the first cluster is obtained by analyzing the first monitoring record; Perform predictive analysis on the first processing progress timeline to obtain the predicted time when the progress is completed; The first prediction delay of the first cluster is obtained based on the prediction time. If the first predicted latency does not meet the predetermined latency threshold, then the computing power nodes of the first cluster are adjusted by migration.

7. The intelligent computing power cluster task allocation method based on cloud-edge collaboration as described in claim 6, characterized in that, Also includes: If the first predicted delay meets the predetermined delay threshold, a task processing evaluation strategy is introduced to evaluate the first monitoring record to obtain the first processing fitness. The average value of the first processing fitness is denoted as the comprehensive fitness, wherein the comprehensive fitness is used to characterize the overall allocation effect of the task to be assigned.

8. A cloud-edge collaborative intelligent computing cluster task allocation system, characterized in that, The system is used to implement the cloud-edge collaborative intelligent computing cluster task allocation method according to any one of claims 1-7, and the system includes: The first feature curve acquisition module is used to extract the first task from the tasks to be assigned, and collect the first multidimensional feature parameters of the first task to obtain the first feature curve. The clustering result acquisition module is used to perform clustering analysis on the task to be assigned based on the first feature curve to obtain clustering results, wherein the clustering results include a first cluster; The first resource demand analysis module is used to analyze and obtain the first resource demand of the first cluster and determine whether the distributed edge computing power in the cloud edge computing power cluster meets the first resource demand. The first fitness acquisition module is used to obtain the first fitness of the first computing power node in the distributed edge computing power to the first cluster if the condition is met; The target computing power node acquisition module is used to sort the computing power nodes in descending order based on the first fitness to obtain the target computing power node corresponding to the highest fitness; the task processing module is used to perform task processing for the first cluster through the target computing power node.

Images