A computing power operator-oriented computing power and electricity collaborative regulation potential capacity method

Abstract

The application provides a kind of computing power operator-oriented computing power and electricity coordination regulation potential method, comprising: collecting the computing power task load parameter of data center cluster;The computing power task load parameter is decomposed according to task type to obtain load decomposition model;Combining load decomposition model, the potential index of computing power task load parameter in multiple dimensions is calculated;The comprehensive potential grade is calculated using multiple dimensions of potential index;According to the comprehensive potential grade, whether the target data center cluster meets the scheduling requirement of computing power operator is judged.The application can identify the adjustable space of computing power load under different time, different region and different business constraints through multiple dimensions of potential index, and give the quantitative evaluation result that can be used for demand response, auxiliary service, new energy consumption and computing power facility planning.

Description

Technical Field

[0001] This invention relates to the field of data center load management technology, and more specifically, to a method for quantifying the potential of computing power coordinated regulation for computing power operators. Background Technology

[0002] As AI training, cloud computing services, and edge computing continue to expand, the proportion of data center load in the power system is constantly increasing. Computing operators typically manage multiple geographically dispersed data center nodes, and their loads are not of a single type: they include long-duration, high-power training tasks, latency-sensitive inference services, general-purpose computing, and deferred storage backup tasks. These tasks differ significantly in terms of real-time performance, latency tolerance, computational intensity, and data dependencies, and therefore, their corresponding power regulation capabilities also differ.

[0003] Existing data center demand response or load control solutions often treat the data center as a single load entity, primarily relying on peak shaving, peak shifting, or power curtailment based on the total power curve. This approach fails to adequately differentiate between task types, task queue status, cross-regional migration conditions, service quality requirements, and cooling system thermal inertia, easily leading to biases in adjustable capacity assessments. This, in turn, affects judgments regarding adjustable capacity, adjustable capacity duration, and viable market segments. For computing loads, the characteristics of task queuing, resource reallocation, and cross-regional business acceptance lack a unified assessment framework. A method for assessing the computing-power synergy potential is needed that can quantify and uniformly calculate computing task characteristics, data center spatial distribution, power grid regulation requirements, and market revenue. Summary of the Invention

[0004] To address the aforementioned problems, the present invention aims to provide a method for quantifying the potential of computing power coordinated regulation for computing power operators.

[0005] A method for quantifying the potential of computing power coordinated regulation for computing power operators includes: Step 1: Collect computing power and task load parameters of the data center cluster; Step 2: Decompose the computing power task load parameters according to task type to obtain the load decomposition model; Step 3: Calculate the potential indicators of computing power task load parameters in multiple dimensions by combining the load decomposition model; Step 4: Calculate the overall potential level using potential indicators across multiple dimensions; Step 5: Determine whether the target data center cluster meets the scheduling requirements of the computing power operator based on the comprehensive potential level.

[0006] Preferably, in step 2, the load decomposition model is: in, This represents the AI ​​training load of the m-th data center at time t. Indicates AI inference load, Indicates general computing load. Indicates storage / backup load. This represents the residual term caused by monitoring errors, unidentified tasks, or auxiliary equipment power.

[0007] Preferably, in step 3, the potential indicators include time-adjustable potential, spatial aggregation potential, and flexible boundary potential.

[0008] Preferably, in step 3, the time-adjustable potential index is: in, The maximum time-adjustable potential for the c-th type of task in the m-th data center. For a deferred execution time window, To enable batch processing and merging of windows, For windows that can be interrupted or paused, , , This is the response coefficient.

[0009] Preferably, in step 3, the spatial aggregation potential index is: in, and These represent the potential for upward and downward adjustments after aggregation by computing power operators, respectively. and These represent the potential for upward and downward adjustments for individual data centers, respectively. and These represent the correlation correction coefficients for the upscaling and downscaling potential of multiple data centers, respectively.

[0010] Preferably, in step 3, the elastic boundary potential index is: in, and To adjust the execution efficiency upwards and downwards, and To adjust the execution efficiency coefficient upwards and downwards, For power supply line capacity, Let m be the total power of data center m at time t. To meet the minimum load boundary required for the operation of the target business, and To adjust the proportional coefficient.

[0011] This invention also provides a system for quantifying the potential of computing power coordinated regulation for computing power operators, comprising: The parameter acquisition module is used to collect computing power and task load parameters of the data center cluster. The decomposition module is used to decompose the computing power task load parameters according to the task type to obtain the load decomposition model; The potential index calculation module is used to calculate the potential index of computing power task load parameters in multiple dimensions by combining the load decomposition model. The potential level determination module is used to calculate the comprehensive potential level using potential indicators across multiple dimensions. The judgment module is used to determine whether the target data center cluster meets the scheduling requirements of the computing power operator based on the comprehensive potential level.

[0012] The present invention also provides an electronic device, including a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the transceiver, the memory, and the processor are connected via the bus, characterized in that the computer program, when executed by the processor, implements the steps in the above-described method for quantifying the potential of computing power coordinated regulation for computing power operators.

[0013] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps in the above-described method for quantifying the potential of computing power coordinated regulation for computing power operators.

[0014] The beneficial effects of the proposed method for quantifying the computing power coordinated regulation potential for computing power operators are as follows: Compared with the prior art, the present invention can identify the adjustable space of computing power load under different times, different regions and different business constraints through potential indicators of multiple dimensions, and provide quantitative evaluation results that can be used for demand response, ancillary services, new energy consumption and computing power facility planning.

[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the 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.

[0017] Figure 1This invention provides a flowchart of a method for quantifying the potential of computing power coordinated regulation for computing power operators, according to an embodiment of the present invention. Figure 2 The diagram illustrates the principle of a method for quantifying the potential of computing power coordinated regulation for computing power operators, as provided in an embodiment of the present invention. Detailed Implementation

[0018] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0019] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0020] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0021] Please see Figure 1 A method for quantifying the potential of computing power coordinated regulation for computing power operators, comprising: Step 1: Collect computing power and task load parameters of the data center cluster; Step 2: Decompose the computing power task load parameters according to task type to obtain the load decomposition model; Step 3: Calculate the potential indicators of computing power task load parameters in multiple dimensions by combining the load decomposition model; Step 4: Calculate the overall potential level using potential indicators across multiple dimensions; Step 5: Determine whether the target data center cluster meets the scheduling requirements of the computing power operator based on the comprehensive potential level.

[0022] The process of the present invention will be further described below with reference to specific embodiments: This invention collects operational data from computing power operators and data centers, electricity market data, and regional power grid data. It establishes a computing power task classification system and a task characteristic load decomposition model. It calculates collaborative potential indicators from five aspects: time, space, elastic boundary, economic benefits, and system support. It comprehensively evaluates the computing and power collaboration potential and calculates the potential level.

[0023] Please see Figure 2 A method for quantifying the potential of computing power coordinated regulation for computing power operators involves evaluating technical adjustability along with electricity market revenue and system support value. This facilitates computing power operators and grid dispatching agencies using the same results for resource identification, market application, and subsequent dispatching. The method includes the following steps: Step 1: Based on the data center cluster computing and power collaboration scenario, collect relevant data and construct parameters; Step 2: Considering the characteristics of computing power tasks, construct a load decomposition model and a basic load curve; Step 3: Consider the potential characteristics in spatial and temporal dimensions, and perform multi-dimensional collaborative potential calculation; Step 4: Based on the potential indicators of each dimension, calculate the comprehensive evaluation value of computing and power synergy potential and determine the comprehensive potential level.

[0024] In step one, based on the data center cluster computing and power collaboration scenario, relevant data is collected and parameters are constructed, including: Within the evaluation period T, assume the computing power operator manages M data centers, each containing multiple types of computing power tasks. For the m-th data center, its total power at time t is collected. IT power Cooling power Task queue Task type set C, computational resource utilization rate Task latency tolerance Data migration constraints Network bandwidth constraints and power supply line capacity .

[0025] The task type set C includes AI training tasks, AI inference tasks, general computing tasks, and storage backup tasks. On the power system side, regional grid load curves, day-ahead or real-time electricity prices, renewable energy output curves, demand response instructions, and ancillary service demands are collected synchronously; on the market side, parameters such as demand response subsidy standards, frequency regulation capacity prices, frequency regulation mileage prices, peak-valley price differences, and capacity value are collected.

[0026] Table 1. Classification of computing tasks and load characteristics In step two, considering the characteristics of the computing task, a load decomposition model and a basic load curve are constructed, including: The total load for each data center is decomposed by task type, yielding the base load curves for each task type. The decomposition model is as follows: (1) In the formula: This represents the AI ​​training load of the m-th data center at time t. Indicates AI inference load, Indicates general computing load. Indicates storage / backup load. This represents the residual term caused by monitoring errors, unidentified tasks, or auxiliary equipment power.

[0027] For each type of task C, its base load is determined based on the task arrival rate, task execution duration, resource utilization rate, and historical power curve. Delayable window Batch processing window Interruptible window and service quality constraints Through this process, the total power curve is no longer treated as a single black-box input, but is transformed into an evaluable load curve with task attributes.

[0028] In step three, considering the potential characteristics of spatial and temporal dimensions, a multi-dimensional collaborative potential calculation is performed, including: The time-dimensional adjustable potential represents the ability of various tasks to adjust power through delays, batching, or interruptions within the evaluation period, and its expression is: (2) In the formula: The maximum time-adjustable potential for the c-th type of task in the m-th data center. For a deferred execution time window, To enable batch processing and merging of windows, . is a window that can be interrupted or paused. , , The response coefficient is related to the task type.

[0029] Spatial aggregation potential represents the load complementarity capability formed by multiple data centers due to differences in geographical location, business load, and power structure. Its expression is: (3) In the formula: and These represent the potential for upward and downward adjustments after aggregation by computing power operators, respectively. and These represent the potential for upward and downward adjustments for individual data centers, respectively. and These represent the correlation correction coefficients for the upscaling and downscaling potential of multiple data centers, respectively. The lower the correlation, the greater the complementary space after aggregation.

[0030] The elastic boundary potential is used to define the adjustable capacity of each data center without violating quality of service, equipment capacity, and electrical constraints. Its expression is: (4) In the formula: and To adjust the execution efficiency upwards and downwards, and To adjust the execution efficiency coefficient upwards and downwards, For power supply line capacity, Let m be the total power of data center m at time t. To meet the minimum load boundary required for core business operations, and This is the adjustment ratio coefficient corresponding to the task type.

[0031] Step 4: Based on the potential indicators of each dimension, calculate the comprehensive evaluation value of the computing-electricity synergy potential and determine the comprehensive potential level, including: After obtaining the time-adjustable potential, spatial aggregation potential, and elastic boundary potential, we continue to calculate the economic return potential. and system support potential Economic benefit potential includes demand response benefits. Ancillary service revenue Spot peak-valley arbitrage profits and revenue from green electricity consumption The system's support potential includes contributions to peak shaving and valley filling. Contribution of new energy consumption Capacity support contribution and frequency support contribution .

[0032] (5) In the formula: To assess the total electricity costs of the data center during the evaluation period (i.e., the benchmark revenue). For the set of time periods during which demand response events occur, The actual load capacity reduction at time t. Let t be the demand response subsidy price. For time step, Costs are incurred to compensate for demand response capacity. This is an indicator variable for whether capacity compensation is obtained (takes 0 or 1); The frequency modulation capacity at time t is the bid-winning capacity. For frequency modulation capacity price, The actual frequency modulation mileage at time t. Price based on FM mileage; To shift load capacity from peak to off-peak, and Electricity prices are calculated based on peak and off-peak hours. Let t be the capacity of new energy power consumed at time t. and For grid electricity price and green electricity agreement price, For the value of carbon emission reduction, This represents the average carbon emission factor of the power grid.

[0033] (6) In the formula: The weighting coefficients for each supporting component satisfy the following conditions: , To determine the standard deviation of the net load curve after adjustment, The standard deviation of the original load curve. This represents the average load. Let t be the amount of renewable energy curtailed. The capacity that can be increased for computing load at time t; This is the set of peak load periods for the system. The capacity to be reduced for computing load at time t. For system peak load, The number of data centers in the region; Let be the system frequency deviation at time t. Let t be the computing power capacity participating in frequency modulation, and k be the frequency modulation response sensitivity coefficient.

[0034] Based on this, the original indicator values ​​for the five dimensions were calculated, and the extreme value method was used to normalize the indicators for each dimension. The weighted average was then used to obtain the comprehensive potential assessment value. (7) In the formula: This is the comprehensive evaluation value of the computing and power synergy potential of the m-th data center. For spatially normalized time dimension indicators, As a spatial aggregation index, As an elastic boundary indicator, As an indicator of economic returns, As system support indicators, to The corresponding weights are determined using methods such as entropy weighting, expert weighting, or calibration based on historical participation effects.

[0035] Based on the comprehensive assessment value Set potential levels. When Greater than or equal to the first threshold At that time, it was determined to be a high-potential resource; when Less than And greater than or equal to the second threshold At that time, it was determined to be a medium-potential resource; when Less than At that time, it is judged as a low-potential resource. The threshold can be configured according to the regional power grid dispatching needs, market access conditions, or the risk preferences of computing power operators.

[0036] Table 2 Evaluation Indicators for Computing-Electronic Collaboration Potential This method outputs the adjustable capacity curve, adjustable capacity curve, adjustable duration, task type contribution ratio, eligible market segments, recommended participation periods, regional collaboration priority, and overall potential level for each data center within the evaluation period. These results can be used by computing power operators to formulate market application strategies and by power grid dispatching agencies to identify new demand-side resources that can be aggregated.

[0037] Taking a 100MW data center managed by a computing power operator as an example, data such as task queues, IT load, cooling system load, electricity price, demand response subsidies, and renewable energy output are collected during the evaluation period. The task structure of this data center is as follows: AI training tasks account for 45%, AI inference tasks account for 25%, general computing tasks account for 15%, and storage / backup tasks account for 15%. The adjustable proportion of each task type is determined by historical operating data and business service quality requirements.

[0038] Table 3 Parameter Settings for the Case Study According to this method, the total load of the data center is first decomposed into four types of task loads, and then the time adjustability potential, upward adjustment boundary, downward adjustment boundary and revenue potential of each task type are calculated. Finally, the results of the synergistic potential of the data center during the evaluation period are summarized.

[0039] Table 4. Evaluation Results of the Potential for Computer-Electronic Collaboration The assessment results indicate that the 100MW data center has room for adjustment in power coordination. During periods of high peak demand, approximately 25.9MW of downscaling capacity can be achieved by delaying storage backup tasks, adjusting general computing queues, and using inference batch processing. During periods of high renewable energy output or low electricity prices, approximately 11.5MW of upscaling capacity can be achieved by executing deferred tasks in advance, migrating some training tasks, and improving resource utilization.

[0040] When computing power operators manage multiple geographically dispersed data centers, electricity prices, renewable energy output, user request distribution, and task queue status often differ across regions. The synergistic potential of multi-center aggregation may be higher than the results of independent assessments of a single center. This method quantifies this aggregation effect using a correlation correction coefficient, outputting aggregated upward and downward capacity adjustments that are more suitable for market application and grid dispatching.

[0041] By classifying tasks, decomposing loads, calculating multi-dimensional potential, and conducting comprehensive evaluations, this method can quantify the ability of computing power operators to participate in power coordination. Compared to evaluation methods based solely on the total data center load curve, this method more easily identifies adjustable resource sources, adjustable capacity boundaries, and adapts to market scenarios, making it easier to implement and use on the engineering side.

[0042] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for quantifying the potential of computing power coordinated regulation for computing power operators, characterized in that, include: Step 1: Collect computing power and task load parameters of the data center cluster; Step 2: Decompose the computing power task load parameters according to task type to obtain the load decomposition model; Step 3: Calculate the potential indicators of computing power task load parameters in multiple dimensions by combining the load decomposition model; Step 4: Calculate the overall potential level using potential indicators across multiple dimensions; Step 5: Determine whether the target data center cluster meets the scheduling requirements of the computing power operator based on the comprehensive potential level.

2. The method for quantifying the potential of computing power coordinated regulation for computing power operators according to claim 1, characterized in that, In step 2, the load decomposition model is as follows: in, This represents the AI ​​training load of the m-th data center at time t. Indicates AI inference load, Indicates general computing load. Indicates storage / backup load. This represents the residual term caused by monitoring errors, unidentified tasks, or auxiliary equipment power.

3. The method for quantifying the potential of computing power coordinated regulation for computing power operators according to claim 2, characterized in that, In step 3, the potential indicators include time-adjustable potential, spatial aggregation potential, and flexible boundary potential.

4. The method for quantifying the potential of computing power coordinated regulation for computing power operators according to claim 3, characterized in that, In step 3, the time-adjustable potential index is: in, The maximum time-adjustable potential for the c-th type of task in the m-th data center. For a deferred execution time window, To enable batch processing and merging of windows, For windows that can be interrupted or paused, , , This is the response coefficient.

5. The method for quantifying the potential of computing power coordinated regulation for computing power operators according to claim 4, characterized in that, In step 3, the spatial aggregation potential index is: in, and These represent the potential for upward and downward adjustments after aggregation by computing power operators, respectively. and These represent the potential for upward and downward adjustments for individual data centers, respectively. and These represent the correlation correction coefficients for the upscaling and downscaling potential of multiple data centers, respectively.

6. The method for quantifying the potential of computing power coordinated regulation for computing power operators according to claim 5, characterized in that, In step 3, the elastic boundary potential index is: in, and To adjust the execution efficiency upwards and downwards, and To adjust the execution efficiency coefficient upwards and downwards, For power supply line capacity, Let m be the total power of data center m at time t. To meet the minimum load boundary required for the operation of the target business, and To adjust the proportional coefficient.

7. A system for quantifying the potential of computing power coordinated regulation for computing power operators, characterized in that, include: The parameter acquisition module is used to collect computing power and task load parameters of the data center cluster. The decomposition module is used to decompose the computing power task load parameters according to the task type to obtain the load decomposition model; The potential index calculation module is used to calculate the potential index of computing power task load parameters in multiple dimensions by combining the load decomposition model. The potential level determination module is used to calculate the comprehensive potential level using potential indicators across multiple dimensions. The judgment module is used to determine whether the target data center cluster meets the scheduling requirements of the computing power operator based on the comprehensive potential level.

8. An electronic device comprising a bus, a transceiver, a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the transceiver, the memory, and the processor are connected via the bus, characterized in that, When the computer program is executed by the processor, it implements the steps in the method for quantifying the potential of computing power coordinated regulation for computing power operators as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps in the method for quantifying the potential of computing power coordinated regulation for computing power operators as described in any one of claims 1-6.

Images