CN-121984858-A - Resource optimization method, device, equipment and medium

CN121984858ACN 121984858 ACN121984858 ACN 121984858ACN-121984858-A

Abstract

The application provides a resource optimization method, a device, equipment and a medium, which relate to the technical field of network operation and maintenance and aim to solve the problem of balancing the gain effect of a gain algorithm corresponding to a task and the additional expense brought by the gain algorithm by evaluating the comprehensive value of the gain algorithm. The resource optimization method comprises the steps of obtaining a gain performance index and a no-gain performance index corresponding to a target task, wherein the gain performance index is used for representing the performance index when the target task is executed by adopting a gain algorithm, the no-gain performance index is used for representing the performance index when the target task is not executed by adopting the gain algorithm, determining a performance comprehensive gain corresponding to the gain algorithm based on the gain performance index and the no-gain performance index, wherein the performance comprehensive gain is used for representing the gain effect of the gain algorithm on the target task, and carrying out light weight processing on the gain algorithm under the condition that the difference value between the performance comprehensive gain and a preset gain threshold is larger than or equal to the preset difference value threshold.

Inventors

ZHOU WEI
HUANG RONG
WANG YOUXIANG
WEI JINWU
FAN BIN
LIU SHAN

Assignees

中国联合网络通信集团有限公司

Dates

Publication Date: 20260505
Application Date: 20260105

Claims (10)

1. A method of resource optimization, comprising: The method comprises the steps of obtaining a gain performance index and a no-gain performance index corresponding to a target task, wherein the gain performance index is used for representing the performance index when the target task is executed by adopting a gain algorithm, and the no-gain performance index is used for representing the performance index when the target task is not executed by adopting the gain algorithm; Determining a performance comprehensive gain corresponding to the gain algorithm based on the gain performance index and the gain-free performance index, wherein the performance comprehensive gain is used for representing the gain effect of the gain algorithm on the target task; And carrying out light weight processing on the gain algorithm under the condition that the difference value between the performance comprehensive gain and the preset gain threshold value is larger than or equal to the preset difference value threshold value.
2. The method of claim 1, wherein the performance metrics include a gain metric and an overhead metric; The determining the performance comprehensive gain corresponding to the target task based on the gain performance index and the gain-free performance index comprises the following steps: Determining a performance gain corresponding to the target task based on the gain index in the gain performance indexes and the gain index in the gain-free performance indexes, wherein the performance gain is used for representing the gain on the performance of the gain algorithm during operation; Determining performance cost corresponding to the target task based on the cost index in the gain performance index and the cost index in the gain-free performance index, wherein the performance cost is used for representing cost required by the operation of the gain algorithm; And determining the performance comprehensive gain corresponding to the target task based on the performance gain and the performance overhead.
3. The method according to claim 2, wherein the gain indicator comprises a transmission rate and/or spectral efficiency; The determining the performance gain corresponding to the target task based on the gain index in the gain performance indexes and the gain index in the gain-free performance indexes includes: Determining a transmission rate gain based on the transmission rate in the gain performance index and the transmission rate in the no-gain performance index, wherein the transmission rate gain is used for representing the gain of the gain algorithm on the data transmission rate of the target task; Determining a spectral efficiency gain based on the spectral efficiency in the gain performance index and the spectral efficiency in the no-gain performance index, wherein the spectral efficiency gain is used for representing the gain of the gain algorithm on the spectral efficiency of the target task; The performance gain includes the transmission rate gain and the spectral efficiency gain.
4. The method of claim 2, wherein the overhead indicator comprises at least one of a time delay, a reliability, or a computational resource, the reliability being indicative of a cost required to ensure system reliability in performing the target task; The determining the performance overhead corresponding to the target task based on the overhead index in the gain performance index and the overhead index in the gain-free performance index includes: determining delay overhead based on the inferred delay in the gain performance index and the data transmission delay in the gain-free performance index; determining reliability overhead based on the inference accuracy in the gain performance index and the data transmission reliability in the gain-less performance index; and determining computing power resource expenditure based on the computing power resources in the gain performance index and the computing power resources in the gain-free performance index.
5. The method of claim 2, wherein the performance gains include transmission rate gain, spectral efficiency gain, and wherein the performance overhead includes latency overhead, reliability overhead, and computational power resource overhead; The determining, based on the performance gain and the performance overhead, a performance synthesis gain corresponding to the target task includes: the transmission rate gain, the spectrum efficiency gain, the time delay cost, the reliability cost and the computing power resource cost are weighted and summed to determine the performance comprehensive gain corresponding to the target task; The weight of the transmission rate gain, the weight of the spectrum efficiency gain, the weight of the time delay cost, the weight of the reliability cost and the weight of the computing power resource cost are determined through a task scene corresponding to the target task.
6. The method of claim 5, wherein the task scenario comprises at least one of holographic communication, autopilot, ambulatory medical service, and large model push; In the case where the task scenario is the holographic communication, the weight of the transmission rate gain and the weight of the spectral efficiency gain are higher than other weights; Under the condition that the task scene is the automatic driving, the weight of the time delay overhead and the weight of the reliability overhead are higher than other weights; in the case that the task scenario is the mobile medical service, the weight of the transmission rate gain and the weight of the reliability overhead are higher than other weights; And under the condition that the task scene is the large model reasoning, the weight of the computing power resource overhead is higher than other weights.
7. The method of claim 1, wherein the lightweight process comprises at least one of model pruning and model quantification.
8. The resource optimizing device is characterized by comprising an index obtaining unit, an evaluating unit and an optimizing unit; The target task performance detection unit is used for detecting the target task performance according to the gain performance index and the gain-free performance index, wherein the gain performance index is used for indicating the performance index when the target task is executed by adopting a gain algorithm, and the gain-free performance index is used for indicating the performance index when the target task is not executed by adopting the gain algorithm; The evaluation unit is used for determining a performance comprehensive gain corresponding to the gain algorithm based on the gain performance index and the gain-free performance index, wherein the performance comprehensive gain is used for representing the gain effect of the gain algorithm on the target task; the optimizing unit is used for carrying out light weight processing on the gain algorithm under the condition that the difference value between the performance comprehensive gain and the preset gain threshold value is larger than or equal to the preset difference value threshold value.
9. A resource optimizing device is characterized by comprising a processor and a memory; the memory is for storing one or more programs, the one or more programs comprising computer-executable instructions that, when executed by the resource optimization device, cause the resource optimization device to perform the method of any of claims 1-7.
10. A computer readable storage medium, characterized in that, when computer-executable instructions stored in the computer readable storage medium are executed by a processor of a resource optimization device, the resource optimization device is capable of performing the method of any of claims 1 to 7.

Description

Resource optimization method, device, equipment and medium Technical Field The present application relates to the field of network operation and maintenance technologies, and in particular, to a method, an apparatus, a device, and a medium for resource optimization. Background At present, the network needs to bear various tasks based on gain algorithm, such as high speed, low time delay, high reliability and the like. However, because the network resources and the computational power resources are limited, in order to meet the requirements of a plurality of tasks at the same time, the performance of the gain algorithm corresponding to each task cannot be improved without limitation. Therefore, it is necessary to balance the gain effect of the gain algorithm and the overhead of the gain algorithm for each task when the task demands are satisfied. Disclosure of Invention The embodiment of the disclosure provides a resource optimization method, a device, equipment and a medium, which aim to solve the problem of balancing the gain effect of a gain algorithm corresponding to a task and the additional expense brought by the gain algorithm by evaluating the comprehensive value of the gain algorithm. In order to achieve the above purpose, the application adopts the following technical scheme: The resource optimization method comprises the steps of obtaining a gain performance index and a gain-free performance index corresponding to a target task, wherein the gain performance index is used for representing the performance index when the target task is executed by adopting a gain algorithm, the gain-free performance index is used for representing the performance index when the target task is not executed by adopting the gain algorithm, determining a performance comprehensive gain corresponding to the gain algorithm based on the gain performance index and the gain-free performance index, wherein the performance comprehensive gain is used for representing the gain effect of the gain algorithm on the target task, and carrying out light weight processing on the gain algorithm under the condition that the difference value between the performance comprehensive gain and a preset gain threshold value is larger than or equal to the preset difference value threshold value. According to the method, the gain performance index when the gain algorithm is adopted and the gain performance index when the gain algorithm is not adopted are obtained, the performance comprehensive gain is determined according to the two performance indexes, comprehensive quantitative evaluation of the algorithm effect is achieved, the gain algorithm with the excessively high gain effect (namely, the algorithm force redundancy) is screened out through the performance comprehensive gain, the preset gain threshold value and the preset difference threshold value, and the gain algorithm is light-weighted, so that the complexity degree of the gain algorithm is reduced on the premise that the core gain effect of the algorithm is maintained, algorithm resources can be inclined to the task execution process, and the gain effect and the extra resource expenditure of the gain algorithm are balanced. In some embodiments, the performance indicators comprise gain indicators and overhead indicators, determining the performance comprehensive gain corresponding to the target task based on the gain performance indicators and the gain-free performance indicators comprises determining the performance gain corresponding to the target task based on the gain indicators in the gain performance indicators and the gain indicators in the gain-free performance indicators, wherein the performance gain is used for representing the gain on the performance of the gain algorithm when the gain algorithm is running, determining the performance overhead corresponding to the target task based on the overhead indicators in the gain performance indicators and the overhead indicators in the gain-free performance indicators, wherein the performance overhead is used for representing the overhead required by the gain algorithm when the gain algorithm is running, and determining the performance comprehensive gain corresponding to the target task based on the performance gain and the performance overhead. In some embodiments, the gain indicator comprises a transmission rate and/or spectral efficiency, determining a performance gain corresponding to the target task based on the gain indicator in the gain performance indicator and the gain indicator in the no gain performance indicator comprises determining a transmission rate gain based on the transmission rate in the gain performance indicator and the transmission rate in the no gain performance indicator, the transmission rate gain being used for representing the gain of the gain algorithm to the data transmission rate of the target task, determining a spectral efficiency gain based on the spectral efficiency in the gain performance indicator and