CN-121996405-A - Multi-time-scale computing power load management method and system

CN121996405ACN 121996405 ACN121996405 ACN 121996405ACN-121996405-A

Abstract

The invention relates to the technical field of electric power systems and discloses a multi-time-scale calculation load management method and system, wherein the method comprises the steps of obtaining historical operation data of each type of data center, sequentially analyzing calculation load characteristics and corresponding processing adaptation rules of each type of data center under the multi-time scale based on the historical operation data, responding to a load optimization signal, carrying out integrated processing on the obtained real-time operation data of a target data center based on the calculation load characteristics and the corresponding processing adaptation rules to obtain multi-source real-time data under the multi-time scale, inputting the multi-source real-time data into a calculation load dynamic prediction model to obtain calculation load prediction results of the target data center under different time scales, and carrying out migration scheduling on time-adjustable tasks of the target data center based on the calculation load prediction results so as to realize load management optimization of the target data center. The method improves the utilization efficiency of the calculation load.

Inventors

WANG DONG
Shang Jiaping
PAN XIUKUI
BAI DESHENG
WANG JUNSHENG
GUO QINGLEI
YANG KE
ZHANG XUESEN
DU ZHE
Hao Jiahan
LIU BEN

Assignees

国网数字科技控股有限公司
国网区块链科技(北京)有限公司

Dates

Publication Date: 20260508
Application Date: 20251205

Claims (10)

1. A multi-time scale computational load management method, comprising: Acquiring historical operation data of each type of data center, wherein the historical operation data comprise server performance indexes, power consumption data and task processing capacity under a long time span; sequentially analyzing the calculation force load characteristics and the corresponding processing adaptation rules of each type of data center under multiple time scales based on the historical operation data; Responding to a load optimization signal, and based on the computational power load characteristics and the processing adaptation rules, integrating the acquired real-time server performance index, real-time power consumption data and real-time task processing capacity of the target data center to obtain multi-source real-time data under a multi-time scale; Respectively inputting the multisource real-time data into a pre-trained computational load dynamic prediction model to obtain computational load prediction results of the target data center under different time scales, wherein the computational load dynamic prediction model is configured into a multi-branch network structure and is used for dynamically adjusting the time granularity of the input data to adapt to the load rules of different time scales; And performing cross-scale migration scheduling on the time-adjustable task of the target data center based on the calculation force load prediction result so as to realize load management optimization of the target data center.
2. The multi-time scale computing power load management method of claim 1, wherein the data center comprises a general purpose data center, a mental data center, and a super-mental data center; The sequentially analyzing the computing power load characteristics and the corresponding processing adaptation rules of each type of data center under multiple time scales based on the historical operation data comprises the following steps: preprocessing the historical operation data to obtain a standardized historical load data set; based on the type division of the data center, respectively carrying out classification statistics on the server performance index, the power consumption data and the task processing capacity in the historical load data set to obtain basic load characteristic data of various types of data centers; And carrying out hierarchical analysis on the load change data under multiple time scales in the basic load characteristic data, extracting load peak-valley time periods, fluctuation amplitude and periodic characteristics under each time scale, obtaining the computational power load characteristics of various types of data centers under the multiple time scales, and generating corresponding processing adaptation rules based on the computational power load characteristics of the various types of data centers.
3. The multi-time scale computing power load management method according to claim 1, wherein the integrating the acquired real-time server performance index, real-time power consumption data and real-time task throughput of the target data center based on the computing power load characteristics and the processing adaptation rule to obtain multi-source real-time data under the multi-time scale comprises: Acquiring real-time operation data of a target data center, wherein the real-time operation data comprises a real-time server performance index, real-time power consumption data and real-time task processing capacity; performing outlier detection and filtering on the real-time server performance index, the real-time power consumption data and the real-time task processing capacity based on the processing adaptation rule of the target data center to obtain real-time filtering data; Sequentially performing time granularity alignment and dynamic correction on the real-time filtering data based on the calculation power load characteristics of the target data center to obtain corrected real-time subentry data, wherein the dynamic correction is designed to enable the change trend of the real-time filtering data to be matched with the historical load characteristics; And carrying out association integration on the real-time sub-item data according to a preset format to obtain standardized multi-source real-time data under a multi-time scale.
4. The multi-time scale computing power load management method according to claim 1, wherein the step of inputting each multi-source real-time data into a pre-trained computing power load dynamic prediction model to obtain computing power load prediction results of the target data center under different time scales comprises the following steps: Performing feature extraction on each multi-source real-time data, and constructing a model input feature set under each time scale according to a feature extraction result, wherein the model input feature set comprises real-time performance features, energy consumption features and task load features related to calculation force load; and respectively inputting the model input feature sets under each time scale into the computational load dynamic prediction model to perform load prediction of multiple time scales, so as to obtain computational load prediction results of the target data center under different time scales.
5. The multi-time scale computing power load management method of claim 1, wherein the performing cross-scale migration scheduling on the time-tunable tasks of the target data center based on the computing power load prediction result comprises: Analyzing the time migration characteristics of the target data center computing tasks based on the calculation force load prediction results, and identifying time-adjustable tasks in the target data center computing tasks; constructing a calculation task cross-time scale scheduling model adapting to new energy output based on the characteristics of the time-adjustable task and the green running requirement of the target data center; And flexibly transferring and scheduling the time-adjustable task under different time scales according to the scheduling model so as to match the calculation load prediction result with the fluctuation of the new energy power generation curve.
6. A multi-time scale computing power load management system, comprising: The system comprises an acquisition module, a processing module and a processing module, wherein the acquisition module is used for acquiring historical operation data of each type of data center, wherein the historical operation data comprise server performance indexes, power consumption data and task processing capacity under a long time span; The analysis module is used for sequentially analyzing the calculation force load characteristics of each type of data center under multiple time scales and the corresponding processing adaptation rules based on the historical operation data; The integration module is used for responding to the load optimization signal, and integrating the acquired real-time server performance index, the real-time power consumption data and the real-time task throughput of the target data center based on the calculation power load characteristics and the processing adaptation rule to obtain multi-source real-time data under a multi-time scale; the prediction module is used for respectively inputting the multisource real-time data into a pre-trained computational load dynamic prediction model to obtain computational load prediction results of the target data center under different time scales, wherein the computational load dynamic prediction model is configured into a multi-branch network structure and is used for dynamically adjusting the time granularity of the input data to adapt to the load rules of different time scales; And the scheduling module is used for performing cross-scale migration scheduling on the time-adjustable task of the target data center based on the calculation power load prediction result so as to realize the load management optimization of the target data center.
7. The multi-time scale computing power load management system of claim 6, wherein the data center comprises a general purpose data center, a mental data center, and a super-mental data center; The analysis module is specifically configured to: preprocessing the historical operation data to obtain a standardized historical load data set; based on the type division of the data center, respectively carrying out classification statistics on the server performance index, the power consumption data and the task processing capacity in the historical load data set to obtain basic load characteristic data of various types of data centers; And carrying out hierarchical analysis on the load change data under multiple time scales in the basic load characteristic data, extracting load peak-valley time periods, fluctuation amplitude and periodic characteristics under each time scale, obtaining the computational power load characteristics of various types of data centers under the multiple time scales, and generating corresponding processing adaptation rules based on the computational power load characteristics of the various types of data centers.
8. The multi-time scale computing power load management system of claim 6, wherein the integration module is specifically configured to: Acquiring real-time operation data of a target data center, wherein the real-time operation data comprises a real-time server performance index, real-time power consumption data and real-time task processing capacity; performing outlier detection and filtering on the real-time server performance index, the real-time power consumption data and the real-time task processing capacity based on the processing adaptation rule of the target data center to obtain real-time filtering data; Sequentially performing time granularity alignment and dynamic correction on the real-time filtering data based on the calculation power load characteristics of the target data center to obtain corrected real-time subentry data, wherein the dynamic correction is designed to enable the change trend of the real-time filtering data to be matched with the historical load characteristics; And carrying out association integration on the real-time sub-item data according to a preset format to obtain standardized multi-source real-time data under a multi-time scale.
9. The multi-time scale computational power load management system of claim 6 wherein the prediction module is specifically configured to: Performing feature extraction on each multi-source real-time data, and constructing a model input feature set according to a feature extraction result, wherein the model input feature set comprises real-time performance features, energy consumption features and task load features related to calculation force load; and respectively inputting the model input feature sets under each time scale into a computational load dynamic prediction model to perform load prediction of multiple time scales, so as to obtain computational load prediction results of the target data center under different time scales.
10. The multi-time scale computing power load management system of claim 6, wherein the scheduling module is specifically configured to: Analyzing the time migration characteristics of the target data center computing tasks based on the calculation force load prediction results, and identifying time-adjustable tasks in the target data center computing tasks; constructing a calculation task cross-time scale scheduling model adapting to new energy output based on the characteristics of the time-adjustable task and the green running requirement of the target data center; And flexibly transferring and scheduling the time-adjustable task under different time scales according to the scheduling model so as to match the calculation load prediction result with the fluctuation of the new energy power generation curve.

Description

Multi-time-scale computing power load management method and system Technical Field The invention relates to the technical field of data center energy efficiency management, in particular to a multi-time-scale computing power load management method and system. Background With the rapid development of digital economies, data centers are becoming the core of computing infrastructure, with the scale and number of data centers continually increasing, and the resulting high energy consumption problem is becoming more pronounced. The computational load management of the existing data center depends on empirical scheduling or single time scale monitoring, complex and changeable task loads and power supply fluctuation are difficult to deal with, and particularly under the background of improving the permeability of new energy, the existing method often ignores load characteristic differences of different types of data centers and load fluctuation rules under the multiple time scales, so that load peak-valley differences are large, the energy utilization efficiency is reduced, and the development demands of greenization and high efficiency of the data center are difficult to meet. Disclosure of Invention The invention provides a multi-time-scale computing power load management method and a multi-time-scale computing power load management system, which are used for solving the technical problem of how to improve the existing multi-time-scale computing power load management method and achieving the effect of improving the energy utilization efficiency of a data center. In order to solve the above technical problems, an aspect of the present invention provides a multi-time scale computing power load management method, including: Acquiring historical operation data of each type of data center, wherein the historical operation data comprise server performance indexes, power consumption data and task processing capacity under a long time span; sequentially analyzing the calculation force load characteristics and the corresponding processing adaptation rules of each type of data center under multiple time scales based on the historical operation data; Responding to a load optimization signal, and based on the computational power load characteristics and the processing adaptation rules, integrating the acquired real-time server performance index, real-time power consumption data and real-time task processing capacity of the target data center to obtain multi-source real-time data under a multi-time scale; Respectively inputting the multisource real-time data into a pre-trained computational load dynamic prediction model to obtain computational load prediction results of the target data center under different time scales, wherein the computational load dynamic prediction model is configured into a multi-branch network structure and is used for dynamically adjusting the time granularity of the input data to adapt to the load rules of different time scales; And performing cross-scale migration scheduling on the time-adjustable task of the target data center based on the calculation force load prediction result so as to realize load management optimization of the target data center. As one preferable scheme, the data center comprises a general data center, a intelligent data center and a super-arithmetic data center; The sequentially analyzing the computing power load characteristics and the corresponding processing adaptation rules of each type of data center under multiple time scales based on the historical operation data comprises the following steps: preprocessing the historical operation data to obtain a standardized historical load data set; based on the type division of the data center, respectively carrying out classification statistics on the server performance index, the power consumption data and the task processing capacity in the historical load data set to obtain basic load characteristic data of various types of data centers; And carrying out hierarchical analysis on the load change data under multiple time scales in the basic load characteristic data, extracting load peak-valley time periods, fluctuation amplitude and periodic characteristics under each time scale, obtaining the computational power load characteristics of various types of data centers under the multiple time scales, and generating corresponding processing adaptation rules based on the computational power load characteristics of the various types of data centers. As one preferable solution, the integrating processing is performed on the acquired real-time server performance index, real-time power consumption data and real-time task throughput of the target data center based on the computing power load feature and the processing adaptation rule, to obtain multi-source real-time data under a multi-time scale, including: Acquiring real-time operation data of a target data center, wherein the real-time operation data comprises a real-time server perf