Dynamic computing power scheduling method and system based on multivariate features

Abstract

The invention belongs to the technical field of dynamic computing power scheduling, and particularly relates to a dynamic computing power scheduling method and system based on multivariate features, and the method comprises the steps: firstly determining the basic weight of each dimension in a task feature vector, a node feature vector and a network feature vector based on a formulated and selected system scheduling mode; and then calculating an expected comprehensive utility value of each candidate calculation node combination based on the acquired task feature vector, node feature vector and network feature vector and the basic weight of each dimension, comparing the comprehensive utility values of all candidate node combinations, and selecting the optimal candidate calculation node to execute the task to be scheduled. And then, based on the actual comprehensive utility value and the expected comprehensive utility value when the scheduling task is executed, executing abnormity is evaluated, early warning of a corresponding level is carried out according to an evaluation result, and finally, early warning information and related executing data can be fed back to a dynamic weight decision module to be used for optimizing subsequent scheduling decisions, and closed-loop optimization is formed.

Description

Technical Field

[0001] This invention belongs to the field of dynamic computing power scheduling technology, specifically relating to a dynamic computing power scheduling method and system based on multiple features. Background Technology

[0002] With the rapid development of cloud computing, edge computing and distributed computing systems, computing power scheduling, as a core key technology, directly affects the performance, energy efficiency, cost and reliability of the system. Traditional computing power scheduling methods are usually based on static strategies, such as round-robin, least load first or resource allocation based on fixed weight. Although these methods are simple and easy to implement, they are difficult to adapt to dynamically changing workloads and heterogeneous computing environments.

[0003] In practical applications, task execution involves multiple dimensions of characteristics, including the characteristics of the task itself, such as type and priority; the characteristics of computing nodes, such as CPU / GPU / NPU utilization, memory I / O, and power consumption; and network characteristics, such as latency and bandwidth. Existing scheduling methods often only focus on single or a few indicators such as CPU / memory utilization, lacking comprehensive consideration of multi-dimensional information such as task feature vectors, node feature vectors, and network feature vectors. This leads to one-sided scheduling decisions, which cannot achieve global optimization. Moreover, when the performance of computing nodes fluctuates during execution, timely adjustments cannot be made, which can easily lead to system bottlenecks, resource waste, or service quality degradation.

[0004] To address the aforementioned issues, this application presents a dynamic computing power scheduling method and system based on multiple features. Summary of the Invention

[0005] To address the shortcomings of the prior art mentioned in the background section, this application proposes a dynamic computing power scheduling method and system based on multiple features. By performing correlation analysis on the expected comprehensive utility value of each node combination through task feature vectors, node feature vectors, and network feature vectors, the optimal candidate computing node is selected to execute the scheduled task. Furthermore, the actual comprehensive utility value is used to evaluate abnormal execution results and provide real-time early warnings, thereby solving the problems in the background section.

[0006] Firstly, to achieve the above objectives, this application provides a dynamic computing power scheduling method based on multiple features, which includes the following specific steps:

[0007] S1. Real-time acquisition and construction of task feature vectors, node feature vectors, and network feature vectors;

[0008] S2. Based on the preset system optimization goals and feedback results, dynamically assign weights to each feature dimension in the above multivariate feature vector;

[0009] S3. Based on the multivariate feature vectors and the dynamic weights, calculate the expected comprehensive utility value of each candidate computing node combination. The formula for calculating the expected comprehensive utility value of a node combination is as follows: ,in Let be the expected overall utility value of task i executed on node j. For the task At the node The expected execution time is as follows. Let $\frac{i}{j}$ be the expected resource cost for task $i$ to be executed on node $j$. Let be the expected energy consumption for task i to be executed on node j. The expected network transmission overhead between task i and node j. These are the dynamic weight coefficients for the corresponding dimensions, and ;

[0010] S4. Compare the comprehensive utility values ​​of all candidate node combinations and select the optimal candidate computing node to execute the scheduled task.

[0011] S5. Continuously monitor the execution status of tasks and the status of node resources on selected nodes, and perform monitoring and early warning to form an optimization closed loop.

[0012] Based on the above scheme, the preferred embodiment includes: task feature vector including: task type, estimated execution time, and priority; node feature vector including: CPU / GPU / NPU utilization, memory / storage I / O rate, real-time power consumption, and resource unit price; and network feature vector including: network latency, available bandwidth, and packet loss rate.

[0013] In a preferred embodiment based on the above scheme, the step of dynamically assigning weights to each feature dimension of the multivariate feature vector according to a preset system optimization objective includes the following steps:

[0014] S21. Formulate a set of system strategies, specifically divided into four modes: energy efficiency priority mode, balanced mode, performance priority mode, and cost priority mode.

[0015] S22. For each different system strategy, a set of basic weight vectors is preset, which correspond to each dimension of the task feature vector, node feature vector, and network feature vector.

[0016] S23. Select the appropriate system strategy according to actual needs, determine which feature dimensions are related to the current strategy, and assign corresponding weights to these dimensions.

[0017] Based on the above scheme, the preferred method for continuously monitoring the execution status and resource status of selected nodes to form an optimized closed loop includes the following steps:

[0018] S51. Obtain the actual execution status data and actual node resource status data of the task at the selected node;

[0019] S52. Based on the acquired actual data, evaluate and calculate the actual comprehensive utility value of the task at the selected node. The formula for the actual comprehensive utility value is: ,in Let i be the actual overall utility value of task i executed on node j. For the task At the node The actual execution time on the platform. The actual resource cost of executing task i on node j. Let i be the actual energy consumption of task i executed on node j. The actual network transmission overhead between task i and node j;

[0020] S53. Based on the actual comprehensive utility value calculated in step S52 Compared with the expected comprehensive utility value calculated in step S3 Execution exceptions are evaluated using the following formula: ,in Let be the actual overall utility value of task i executed on node j at time t. and These represent the start and end times of the node's execution. Integral over time;

[0021] S54. Compare the execution anomaly assessment result with the preset anomaly threshold. If the execution anomaly result is greater than the set threshold, trigger an early warning and record the anomaly information.

[0022] S55. Feedback the early warning information and related execution data to the dynamic weight decision module to optimize subsequent scheduling decisions and form a closed-loop optimization.

[0023] Based on the above scheme, the warning threshold set in step S54 is divided into three levels according to the severity level: Level 1, Level 2, and Level 3. When the evaluated execution abnormality result exceeds the Level 1 threshold, a minor warning is triggered; when it exceeds the Level 2 threshold, a moderate warning is triggered; and when it exceeds the Level 3 threshold, a severe warning is triggered. In the case of minor and moderate warnings, feedback optimization is automatically triggered. In the case of severe warning, task migration suggestions are generated and active scheduling optimization is triggered.

[0024] Secondly, this application provides a dynamic computing power scheduling system based on multiple features, which specifically includes: a feature acquisition and management module: used to acquire and construct task feature vectors, node feature vectors and network feature vectors in real time;

[0025] Dynamic weight decision module: used to dynamically assign weights to each feature dimension in the above multivariate feature vector based on the preset system optimization goals and feedback results;

[0026] Comprehensive utility value calculation module: used to calculate the expected comprehensive utility value of each candidate computing node combination based on the multivariate feature vector and the dynamic weight;

[0027] Optimal scheduling decision module: used to compare the comprehensive utility value of all candidate node combinations and select the optimal candidate computing node to execute the scheduled task;

[0028] The scheduling execution early warning and feedback module is used to continuously monitor the execution status of tasks and the resource status of nodes at selected nodes, perform monitoring and early warning to form an optimization closed loop, and feed the monitoring data back to the dynamic weight decision module to form a closed-loop optimization.

[0029] Thirdly, this application provides an electronic device, including: a processor and a memory, wherein the memory stores a computer program that can be called by the processor;

[0030] The processor executes the aforementioned dynamic computing power scheduling method based on multiple features by calling the computer program stored in the memory.

[0031] Fourthly, this application provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the dynamic computing power scheduling method based on multiple features as described above.

[0032] Compared with the prior art, the beneficial effects of the present invention are:

[0033] This invention first determines the basic weights of each dimension in the task feature vector, node feature vector, and network feature vector based on the selected system scheduling mode. Then, based on the collected task feature vector, node feature vector, and network feature vector and the basic weights of each dimension, it calculates the expected comprehensive utility value of each candidate computing node combination. Next, it compares the comprehensive utility values ​​of all candidate node combinations and selects the optimal candidate computing node to execute the scheduled task. Finally, it calculates the actual comprehensive utility value during task execution. Compared with expected total utility value Anomalies are assessed, and warnings of corresponding levels are issued based on the assessment results. Finally, the warning information and related execution data can be fed back to the dynamic weight decision module to optimize subsequent scheduling decisions, forming a closed-loop optimization. By introducing three major categories of multi-dimensional features—tasks, nodes, and networks—it can achieve a leap from single-dimensional scheduling to multi-dimensional collaborative scheduling, which can significantly improve the overall resource utilization and task execution efficiency of the system. Furthermore, through the closed-loop feedback mechanism, the system can adjust according to changes in the external environment and internal strategies, exhibiting strong robustness and adaptability. Attached Figure Description

[0034] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0035] Figure 1 This is a schematic diagram of the overall process of the dynamic computing power scheduling method based on multiple features of the present invention;

[0036] Figure 2 This is a flowchart of step S5 in the dynamic computing power scheduling method based on multiple features of the present invention.

[0037] Figure 3 This is a system block diagram of the dynamic computing power scheduling system based on multiple features of the present invention. Detailed Implementation

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0039] Example 1

[0040] To address the technical problems raised in the background art, this application provides a preferred embodiment: such as Figures 1-3 As shown, the dynamic computing power scheduling method based on multiple features includes the following specific steps:

[0041] S1. Real-time acquisition and construction of task feature vectors, node feature vectors, and network feature vectors;

[0042] S2. Based on the preset system optimization goals and feedback results, dynamically assign weights to each feature dimension in the above multivariate feature vector;

[0043] S3. Based on the multivariate feature vectors and the dynamic weights, calculate the expected comprehensive utility value of each candidate computing node combination. The formula for calculating the expected comprehensive utility value of a node combination is as follows: ,in Let be the expected overall utility value of task i executed on node j. For the task At the node The expected execution time is as follows. Let $\frac{i}{j}$ be the expected resource cost for task $i$ to be executed on node $j$. Let be the expected energy consumption for task i to be executed on node j. The expected network transmission overhead between task i and node j. These are the dynamic weight coefficients for the corresponding dimensions, and ;

[0044] S4. Compare the comprehensive utility values ​​of all candidate node combinations and select the optimal candidate computing node to execute the scheduled task.

[0045] S5. Continuously monitor the execution status of tasks and the status of node resources on selected nodes, and perform monitoring and early warning to form an optimization closed loop.

[0046] The advantages of this embodiment compared to existing technologies are as follows: The expected comprehensive utility value of each candidate computing node combination can be calculated using the collected task feature vector, node feature vector, network feature vector, and basic weights of each dimension. By comparing the comprehensive utility values ​​of all candidate node combinations, the optimal candidate computing node can be selected to execute the scheduled task. Then, the actual comprehensive utility value during task execution can be used as the basis for the final execution. Compared with expected total utility value Execution anomalies are evaluated, and corresponding warnings are issued based on the evaluation results. The warning information and related execution data can be fed back to the dynamic weight decision module to optimize subsequent scheduling decisions, forming a closed-loop optimization. This enables a leap from single-dimensional scheduling to multi-dimensional collaborative scheduling, which can significantly improve the overall resource utilization and task execution efficiency of the system. The system can adjust according to changes in the external environment and internal strategies, and has strong robustness and adaptability.

[0047] Furthermore:

[0048] In an optional embodiment, the task feature vector includes: task type, estimated execution time, and priority; the node feature vector includes: CPU / GPU / NPU utilization, memory / storage I / O rate, real-time power consumption, and resource unit price; the network feature vector includes: network latency, available bandwidth, and packet loss rate.

[0049] It should be noted that: in actual operation, the task feature vector includes at least one of task type, estimated execution time, and priority; the node feature vector includes at least one of CPU / GPU / NPU utilization, memory / storage I / O rate, real-time power consumption, and resource unit price; the network feature vector includes at least one of network latency, available bandwidth, and packet loss rate. This solution uses the formula... When calculating the actual comprehensive utility value, only four feature dimensions were selected: execution time, resource cost, energy consumption, and network transmission overhead. The selected feature dimensions can be added, removed, or replaced according to needs. In actual operation, the corresponding representative feature dimensions can be selected according to actual needs to calculate the actual comprehensive utility value.

[0050] In an optional embodiment, dynamically assigning weights to each feature dimension in the aforementioned multivariate feature vector according to a preset system optimization objective includes the following steps:

[0051] S21. Develop a set of system strategies, specifically divided into four modes: energy efficiency priority mode, balanced mode, performance priority mode, and cost priority mode. In actual operation, the basic weights of time, energy consumption, cost, and network in the energy efficiency priority mode can be preset to 0.2, 0.5, 0.2, and 0.1, respectively. In the balanced mode, the basic weights of time, energy consumption, cost, and network can be preset to 0.25, 0.25, 0.25, and 0.25, respectively. In the performance priority mode, the basic weights of time, energy consumption, cost, and network can be preset to 0.5, 0.2, 0.2, and 0.1, respectively. In the cost priority mode, the basic weights of time, energy consumption, cost, and network can be preset to 0.2, 0.2, 0.5, and 0.1, respectively. These preset basic weights can be fine-tuned in real time based on feedback and historical data without changing the selected system strategy mode.

[0052] S22. For each different system strategy, a set of basic weight vectors is preset, which correspond to each dimension of the task feature vector, node feature vector, and network feature vector.

[0053] S23. Select the appropriate system strategy according to actual needs, determine which feature dimensions are related to the current strategy, and assign corresponding weights to these dimensions.

[0054] It should be noted that the dynamic weight decision module integrates an online learning mechanism. The actual weight parameters can be fine-tuned in real time based on feedback and historical data, on the basis of the preset base weights and without changing the selected system strategy mode. Historical data can be obtained by querying the historical task database, finding similar tasks by task type, resource requirements, etc., and obtaining historical execution time as a reference for estimation. The online learning mechanism can establish a feedback data warehouse to store historical scheduling decisions, actual execution data, and anomaly records. These data can be used periodically or triggered to retrain utility value prediction models such as linear regression, neural networks, and weight adjustment models to achieve closed-loop optimization. This is existing technology and will not be described in detail here.

[0055] Furthermore:

[0056] In an optional embodiment, the continuous monitoring of the execution status and resource status of the selected node to form an optimization closed loop includes the following steps:

[0057] S51. Obtain the actual execution status data and actual node resource status data of the task at the selected node;

[0058] S52. Based on the acquired actual data, evaluate and calculate the actual comprehensive utility value of the task at the selected node. The formula for the actual comprehensive utility value is: ,in Let i be the actual overall utility value of task i executed on node j. For the task At the node The actual execution time on the platform. The actual resource cost of executing task i on node j. Let i be the actual energy consumption of task i executed on node j. The actual network transmission overhead between task i and node j;

[0059] S53. Based on the actual comprehensive utility value calculated in step S52 Compared with the expected comprehensive utility value calculated in step S3 Execution exceptions are evaluated using the following formula: ,in Let be the actual overall utility value of task i executed on node j at time t. and These represent the start and end times of the node's execution. Integral over time;

[0060] S54. Compare the execution anomaly assessment result with the preset anomaly threshold. If the execution anomaly result is greater than the set threshold, trigger an early warning and record the anomaly information.

[0061] S55. Feedback the early warning information and related execution data to the dynamic weight decision module to optimize subsequent scheduling decisions and form a closed-loop optimization.

[0062] In an optional embodiment, the warning thresholds set in step S54 are divided into three levels according to severity: Level 1, Level 2, and Level 3. When the evaluated execution anomaly result exceeds the Level 1 threshold, a minor warning is triggered; when it exceeds the Level 2 threshold, a moderate warning is triggered; and when it exceeds the Level 3 threshold, a severe warning is triggered. Feedback optimization is automatically triggered in the case of minor and moderate warnings. In the case of severe warnings, task migration suggestions are generated and active scheduling optimization is triggered.

[0063] It should be noted that the formula for calculating the actual comprehensive utility value is as follows: For example, the formula for calculating the actual comprehensive utility value can also be: Final choice The reason is Let $\mathbf$ be the actual network transmission overhead between task $i$ and node $j$, such as latency and packet loss rate. This $\mathbf$ is a non-negative value, and the smaller the network overhead, the better the network conditions for the task to execute on node $j$. To avoid $\mathbf$, we need to consider the network conditions between task $i$ and node $j$. When approaching 0, The value tends towards infinity, leading to numerical instability. Therefore, a formula is used. This can avoid the impact of extreme values ​​and improve computational stability.

[0064] Example 2

[0065] Based on the same inventive concept as in Embodiment 1, such as Figure 1 As shown, this embodiment provides a dynamic computing power scheduling system based on multiple features, which specifically includes: a feature acquisition and management module: used to acquire and construct task feature vectors, node feature vectors and network feature vectors in real time;

[0066] Dynamic weight decision module: used to dynamically assign weights to each feature dimension in the above multivariate feature vector based on the preset system optimization goals and feedback results;

[0067] Comprehensive utility value calculation module: used to calculate the expected comprehensive utility value of each candidate computing node combination based on the multivariate feature vector and the dynamic weight;

[0068] Optimal scheduling decision module: used to compare the comprehensive utility value of all candidate node combinations and select the optimal candidate computing node to execute the scheduled task;

[0069] The scheduling execution early warning and feedback module is used to continuously monitor the execution status of tasks and the resource status of nodes at selected nodes, perform monitoring and early warning to form an optimization closed loop, and feed the monitoring data back to the dynamic weight decision module to form a closed-loop optimization.

[0070] The parameters and steps of each unit module in the dynamic computing power scheduling system based on multiple features of the present invention described above can be referred to the parameters and steps in the embodiments of the dynamic computing power scheduling method based on multiple features mentioned above, and will not be repeated here.

[0071] Example 3

[0072] Based on the same inventive concept as Embodiment 1, this embodiment provides an electronic device, including: a processor and a memory, wherein the memory stores a computer program that can be called by the processor;

[0073] The processor executes the aforementioned dynamic computing power scheduling method based on multiple features by calling the computer program stored in the memory.

[0074] It should be noted that all computer programs for the dynamic computing power scheduling method based on multiple features are implemented in C language. The feature acquisition and management module, dynamic weight decision module, comprehensive utility value calculation module, optimal scheduling decision module, and scheduling execution early warning and feedback module are all controlled by a remote server.

[0075] Example 4

[0076] Based on the same inventive concept as in Embodiment 1, this embodiment proposes a computer-readable storage medium having an erasable and rewritable computer program stored thereon.

[0077] When a computer program runs on a computer device, it causes the computer device to execute the dynamic computing power scheduling method based on multiple features described above.

[0078] For example, computer-readable storage media can be read-only memory, random access memory, read-only optical disc, magnetic tape, floppy disk, and optical data storage devices.

[0079] The various embodiments in this invention are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on its differences from other embodiments. In particular, the embodiments for IoT devices and media are relatively simple in description because they are fundamentally similar to the method embodiments; relevant parts can be referred to the descriptions in the method embodiments.

[0080] The systems, media, and methods provided in the embodiments of the present invention are in one-to-one correspondence. Therefore, the systems and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the systems and media will not be repeated here.

[0081] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0082] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0083] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0084] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0085] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0086] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0087] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0088] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. The above are merely embodiments of the present invention and are not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of the present invention should be included within the scope of the claims of the present invention.

Claims

1. A dynamic computing power scheduling method based on multiple features, characterized in that, Includes the following steps: S1. Real-time acquisition and construction of task feature vectors, node feature vectors, and network feature vectors; S2. Based on the preset system optimization goals and feedback results, dynamically assign weights to each feature dimension in the above multivariate feature vector; S3. Based on the multivariate feature vectors and the dynamic weights, calculate the expected comprehensive utility value of each candidate computing node combination. The formula for calculating the expected comprehensive utility value of a node combination is as follows: ,in Let be the expected overall utility value of task i executed on node j. For the task At the node The expected execution time is as follows. Let $\frac{i}{j}$ be the expected resource cost for task $i$ to be executed on node $j$. Let be the expected energy consumption for task i to be executed on node j. The expected network transmission overhead between task i and node j. These are the dynamic weight coefficients for the corresponding dimensions, and ; S4. Compare the comprehensive utility values ​​of all candidate node combinations and select the optimal candidate computing node to execute the scheduled task. S5. Continuously monitor the execution status of tasks and the status of node resources on selected nodes, and perform monitoring and early warning to form an optimization closed loop.

2. The dynamic computing power scheduling method based on multiple features according to claim 1, characterized in that: The task feature vector includes: task type, estimated execution time, and priority; the node feature vector includes: CPU / GPU / NPU utilization, memory / storage I / O rate, real-time power consumption, and resource unit price; the network feature vector includes: network latency, available bandwidth, and packet loss rate.

3. The dynamic computing power scheduling method based on multiple features according to claim 2, characterized in that: The step of dynamically assigning weights to each feature dimension in the above-mentioned multivariate feature vector according to the preset system optimization objective includes the following steps: S21. Formulate a set of system strategies, specifically divided into four modes: energy efficiency priority mode, balanced mode, performance priority mode, and cost priority mode. S22. For each different system strategy, a set of basic weight vectors is preset, which correspond to each dimension of the task feature vector, node feature vector, and network feature vector. S23. Select the appropriate system strategy according to actual needs, determine which feature dimensions are related to the current strategy, and assign corresponding weights to these dimensions.

4. The dynamic computing power scheduling method based on multiple features according to claim 3, characterized in that: The continuous monitoring task monitors and alerts on the execution status and resource status of selected nodes to form an optimization closed loop, including the following steps: S51. Obtain the actual execution status data and actual node resource status data of the task at the selected node; S52. Based on the acquired actual data, evaluate and calculate the actual comprehensive utility value of the task at the selected node. The formula for the actual comprehensive utility value is: ,in Let i be the actual overall utility value of task i executed on node j. For the task At the node The actual execution time on the platform. The actual resource cost of executing task i on node j. Let i be the actual energy consumption of task i executed on node j. The actual network transmission overhead between task i and node j; S53. Based on the actual comprehensive utility value calculated in step S52 Compared with the expected comprehensive utility value calculated in step S3 Execution exceptions are evaluated using the following formula: ,in Let be the actual overall utility value of task i executed on node j at time t. and These represent the start and end times of the node's execution. Integral over time; S54. Compare the execution anomaly assessment result with the preset anomaly threshold. If the execution anomaly result is greater than the set threshold, trigger an early warning and record the anomaly information. S55. Feedback the early warning information and related execution data to the dynamic weight decision module to optimize subsequent scheduling decisions and form a closed-loop optimization.

5. The dynamic computing power scheduling method based on multiple features according to claim 4, characterized in that: The warning thresholds set in step S54 are divided into three levels according to severity: Level 1, Level 2, and Level 3. When the evaluated execution anomaly result exceeds the Level 1 threshold, a minor warning is triggered; when it exceeds the Level 2 threshold, a moderate warning is triggered; and when it exceeds the Level 3 threshold, a severe warning is triggered. Feedback optimization is automatically triggered in the case of minor and moderate warnings. In the case of severe warnings, task migration suggestions are generated and active scheduling optimization is triggered.

6. A dynamic computing power scheduling system based on multiple features, implemented based on the dynamic computing power scheduling method based on multiple features as described in any one of claims 1-5, characterized in that, Specifically, it includes: a feature acquisition and management module: used to acquire and construct task feature vectors, node feature vectors, and network feature vectors in real time; Dynamic weight decision module: used to dynamically assign weights to each feature dimension in the above multivariate feature vector based on the preset system optimization goals and feedback results; Comprehensive utility value calculation module: used to calculate the expected comprehensive utility value of each candidate computing node combination based on the multivariate feature vector and the dynamic weight; Optimal scheduling decision module: used to compare the comprehensive utility value of all candidate node combinations and select the optimal candidate computing node to execute the scheduled task; The scheduling execution early warning and feedback module is used to continuously monitor the execution status of tasks and the resource status of nodes at selected nodes, perform monitoring and early warning to form an optimization closed loop, and feed the monitoring data back to the dynamic weight decision module to form a closed-loop optimization.

7. An electronic device, comprising: A processor and a memory, wherein the memory stores a computer program that can be called by the processor, characterized in that: the processor executes the dynamic computing power scheduling method based on multiple features as described in any one of claims 1-5 by calling the computer program stored in the memory.

8. A computer-readable storage medium, characterized in that: The system stores instructions that, when executed on a computer, cause the computer to perform the dynamic computing power scheduling method based on multiple features as described in any one of claims 1-5.

Images