Method for evaluating adjustable capability of computing power center cluster and related device

Abstract

This invention belongs to the field of power computing power collaboration and discloses a method and related apparatus for assessing the adjustability of a computing power center cluster. The method includes acquiring day-ahead computing power demand data for the computing power center cluster, and obtaining day-ahead adjustability data based on multi-dimensional coupling constraints using this data; acquiring day-ahead correction data for the computing power center cluster, and correcting the day-ahead adjustability data based on this data to obtain the adjustability assessment result of the computing power center cluster. By comprehensively considering the coupling of multi-dimensional constraints such as service quality constraints, computing power economic constraints, green environment constraints, and safe operation constraints, bidirectional collaborative optimization at the day-ahead and real-time levels is achieved. Through overall planning at the day-ahead level, the comprehensiveness of the assessment results is ensured; through fine-grained real-time sensing at the real-time level, the assessment results no longer solely rely on offline static data but can dynamically reflect random fluctuations in actual operation, ensuring the accuracy of the assessment results.

Description

Technical Field

[0001] This invention belongs to the field of power computing collaboration, and relates to a method and related apparatus for evaluating the adjustability of a computing center cluster. Background Technology

[0002] With the rapid development of the digital economy, computing centers have become core infrastructure supporting emerging industries such as artificial intelligence, cloud computing, and big data. Their scale and number are experiencing explosive growth. The high energy consumption and continuous operation characteristics of computing centers put significant pressure on the stable operation of the power grid. As a collaborative combination of multiple individual computing centers, computing center clusters not only undertake massive data processing and high-performance computing tasks, but also play an important "adjustable load" role in the integrated power generation, grid, load, and storage system. If the adjustability of computing center clusters can be effectively explored and their participation in grid interaction can be promoted, the pressure on grid balance can be effectively alleviated.

[0003] Currently, methods for assessing the adjustability of computing center clusters have many limitations and fail to meet practical application needs. First, existing assessment methods often focus on the local energy efficiency optimization of a single subsystem, neglecting the coupling relationship between multiple constraints such as latency, energy efficiency, and carbon emissions. This leads to assessment results that are disconnected from actual operating scenarios and cannot provide comprehensive support for cluster adjustment strategy formulation. Second, the time scale of these assessment methods is relatively singular, with most only able to achieve day-ahead or real-time single-scale assessments. They lack multi-time-scale coordination mechanisms, making it difficult to adapt to the assessment needs of real-world scenarios and resulting in assessment results that are out of sync with reality. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and related apparatus for evaluating the adjustability of a computing center cluster.

[0005] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, this invention provides a method for assessing the adjustability of a computing center cluster, comprising: acquiring day-ahead computing power demand data of the computing center cluster, and obtaining day-ahead adjustability data of the computing center cluster based on multi-dimensional coupling constraints according to the day-ahead computing power demand data; wherein the multi-dimensional coupling constraints include service quality constraints, computing power economy constraints, green environment constraints, and safe operation constraints; acquiring day-ahead corrected data of the computing center cluster, and correcting the day-ahead adjustability data according to the day-ahead corrected data to obtain an assessment result of the adjustability of the computing center cluster; wherein the day-ahead corrected data is obtained by: obtaining real-time scheduling data for historical periods based on real-time grid frequency and real-time computing power demand data of the computing center cluster, based on multi-dimensional coupling constraints and real-time optimization scheduling objectives; obtaining day-ahead scheduling data for historical periods based on day-ahead computing power demand data of the computing center cluster, based on multi-dimensional coupling constraints and day-ahead optimization scheduling objectives; and obtaining day-ahead corrected data based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustability data for historical periods.

[0006] Optionally, the day-ahead computing power requirement data includes the number of day-ahead tasks, the computing power requirement of each day-ahead task, and the data volume of each day-ahead task; the day-ahead adjustment capability data includes the day-ahead power adjustment range, the day-ahead power adjustment change rate range, and the duration of the day-ahead power adjustment limit; the day-ahead correction data includes one or more of the following: the deviation between real-time adjustment power and day-ahead adjustment power, the constraint boundary triggering frequency of real-time adjustment power, and the real-time change of computing power service quality.

[0007] Optionally, the quality of service constraint is:

[0008] in, The total latency of the computing center cluster. This represents the number of tasks completed in the previous day. For the first The priority weight of each task. For the first The transmission latency of each task , For the first The amount of data per task For the first Network bandwidth for each task For bandwidth utilization, For the first The computational processing latency of each task. , For the first The computing power requirements for each task. For the first The allocation of computing power in the computing center for each task. For computing power utilization, This represents the maximum tolerable latency for the computing center cluster.

[0009] The economic constraints on computing power are as follows:

[0010] in, For effective computing power energy efficiency ratio, The total computing power output of the computing center cluster. This represents the total energy consumption of the computing center cluster. For power efficiency, This serves as the baseline energy efficiency threshold.

[0011] The green environment constraints are:

[0012] in, The total carbon emissions of the computing center cluster. The direct carbon emissions of the computing power center cluster, For the indirect carbon emissions of computing center clusters, For the quantity of energy types, For the first Consumption of energy sources, For the first Carbon emission factors of energy-like substances The electricity purchased by the computing center cluster from the power grid. for The carbon emission factor of the power grid at any time For carbon emission allowances.

[0013] The safe operation constraints are as follows: ,

[0014] in, For the computing power center cluster Taiwan Computing Center Power at any moment and For the computing power center cluster The upper and lower power limits of a computing center. For the computing power center cluster Taiwan Computing Center Power at any moment For the computing power center cluster The maximum power change rate of the computing center.

[0015] Optionally, the real-time optimization scheduling objective is:

[0016] in, To optimize the target in real time, For response precision weights, For computing power center clusters Adjust power demand in real time. The number of computing centers. For the computing power center cluster Taiwan Computing Center Power at any moment For stability weights, For the computing power center cluster Taiwan Computing Center Power at any given moment.

[0017] The current daytime optimization scheduling objective is:

[0018] in, To optimize the target in advance, As a weighted average of energy costs, For the energy cost of computing center clusters, As carbon cost weight, The total carbon emissions of the computing center cluster. For the cost of computing power loss, , This represents the number of tasks completed in the previous day. For the first The transmission latency of each task For the first The computational processing latency of each task. This is the delay compensation coefficient.

[0019] Optionally, obtaining the day-ahead scheduling data for historical periods based on the day-ahead computing power demand data of the computing center cluster, based on multidimensional coupling constraints and day-ahead optimization scheduling objectives, includes: constructing a day-ahead optimization model based on multidimensional coupling constraints and day-ahead optimization scheduling objectives; modeling the uncertainty variables in the day-ahead optimization model using a moment-information-based bibliometric optimization framework to construct a fuzzy set composed of confidence interval constraints of first-order moments and second-order moments; performing multimodal processing on the fuzzy set based on task type based on the preset task priority conservatism coefficients for real-time and non-real-time tasks to obtain a multimodal uncertainty fuzzy set; transforming the day-ahead optimization model containing the multimodal uncertainty fuzzy set into a deterministic optimization model using duality theory and linear matrix inequality transformation methods; and solving the deterministic optimization model based on the day-ahead computing power demand data of the computing center cluster for historical periods to obtain the day-ahead scheduling data for historical periods.

[0020] Optionally, the solution of the deterministic optimization model includes: using a particle swarm optimization algorithm to solve the deterministic optimization model to obtain the current particle swarm optimal solution, and using a beetle whisker algorithm to perform a local search based on the current particle swarm optimal solution to obtain the current beetle whisker optimal solution; using the current beetle whisker optimal solution as the initial optimal solution of the particle swarm optimization algorithm and repeating the above steps until a preset iteration termination condition is reached, and using the current beetle whisker optimal solution as the final solution.

[0021] Optionally, obtaining the real-time scheduling data for historical periods based on the real-time power grid frequency and real-time computing power demand data of the computing center cluster, and based on multi-dimensional coupling constraints and real-time optimization scheduling objectives, includes: obtaining the real-time scheduling data for historical periods by calling a pre-trained real-time scheduling model based on the real-time power grid frequency and real-time computing power demand data of the computing center cluster; wherein, the real-time scheduling model is constructed based on a dual-deep Q-network framework and multi-dimensional coupling constraints and real-time optimization scheduling objectives; the state space of the real-time scheduling model is the real-time power grid frequency and real-time computing power demand data, the action space is the power adjustment level of each computing center in the computing center cluster, and the reward function is the negative value of the real-time optimization scheduling objective.

[0022] Optionally, the pre-trained real-time scheduling model employs a random-first experience replay method for experience replay during training; wherein, during experience replay, the experience... The priority is , For experience Time difference error, Preset as a positive number; experience The normalized importance weights are , The size of the experience pool. For experience The sampling probability, As a regulating factor, The maximum importance weight for each sampled experience; and the storage and sampling of experiences based on the SumTree structure.

[0023] Optionally, obtaining real-time scheduling data for historical periods based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and based on multi-dimensional coupling constraints and real-time optimization scheduling objectives, includes: obtaining real-time scheduling data for historical periods in the current scenario based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and based on multi-dimensional coupling constraints and real-time optimization scheduling objectives; the day-ahead adjustment capability data includes day-ahead adjustment capability data for each scenario; the day-ahead scheduling data for historical periods includes day-ahead scheduling data for historical periods in each scenario; obtaining day-ahead correction data based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data for historical periods includes: obtaining day-ahead correction data for the current scenario based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data for historical periods in the current scenario.

[0024] Optionally, it also includes: acquiring the scenario of the power grid connected to the computing center cluster in real time, and when the scenario is an extreme scenario, acquiring the computing power demand data of the computing center cluster before a preset time period, and obtaining the adjustment capability data of the computing center cluster before a preset time period based on the computing power demand data before the preset time period and multi-dimensional coupling constraints; acquiring the correction data of the computing center cluster before the preset time period, and correcting the adjustment capability data before the preset time period based on the day-ahead correction data, to obtain the emergency adjustability assessment result of the computing center cluster.

[0025] In a second aspect, the present invention provides a system for assessing the adjustability of a computing power center cluster, comprising: a day-ahead module for acquiring day-ahead computing power demand data of the computing power center cluster and, based on the day-ahead computing power demand data, obtaining day-ahead adjustability data of the computing power center cluster based on multi-dimensional coupling constraints; wherein the multi-dimensional coupling constraints include quality of service constraints, computing power economic constraints, green environment constraints, and safe operation constraints; and a correction module for acquiring day-ahead correction data of the computing power center cluster and correcting the day-ahead adjustability data based on the day-ahead correction data to obtain an assessment result of the adjustability of the computing power center cluster; wherein the day-ahead correction data is obtained through the following methods: obtaining real-time scheduling data for historical periods based on real-time grid frequency and real-time computing power demand data of the computing power center cluster, based on multi-dimensional coupling constraints and real-time optimization scheduling objectives; obtaining day-ahead scheduling data for historical periods based on day-ahead computing power demand data of the computing power center cluster, based on multi-dimensional coupling constraints and day-ahead optimization scheduling objectives; and obtaining day-ahead correction data based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustability data for historical periods.

[0026] Optionally, the day-ahead computing power requirement data includes the number of day-ahead tasks, the computing power requirement of each day-ahead task, and the data volume of each day-ahead task; the day-ahead adjustment capability data includes the day-ahead power adjustment range, the day-ahead power adjustment change rate range, and the duration of the day-ahead power adjustment limit; the day-ahead correction data includes one or more of the following: the deviation between real-time adjustment power and day-ahead adjustment power, the constraint boundary triggering frequency of real-time adjustment power, and the real-time change of computing power service quality.

[0027] Optionally, the quality of service constraint is:

[0028] in, The total latency of the computing center cluster. This represents the number of tasks completed in the previous day. For the first The priority weight of each task. For the first The transmission latency of each task , For the first The amount of data per task For the first Network bandwidth for each task For bandwidth utilization, For the first The computational processing latency of each task. , For the first The computing power requirements for each task. For the first The allocation of computing power in the computing center for each task. For computing power utilization, This represents the maximum tolerable latency for the computing center cluster.

[0029] The economic constraints on computing power are as follows:

[0030] in, For effective computing power energy efficiency ratio, The total computing power output of the computing center cluster. This represents the total energy consumption of the computing center cluster. For power efficiency, This serves as the baseline energy efficiency threshold.

[0031] The green environment constraints are:

[0032] in, The total carbon emissions of the computing center cluster. The direct carbon emissions of the computing power center cluster, For the indirect carbon emissions of computing center clusters, For the quantity of energy types, For the first Consumption of energy sources, For the first Carbon emission factors of energy-like substances The electricity purchased by the computing center cluster from the power grid. for The carbon emission factor of the power grid at any time For carbon emission allowances.

[0033] The safe operation constraints are as follows: ,

[0034] in, For the computing power center cluster Taiwan Computing Center Power at any moment and For the computing power center cluster The upper and lower power limits of a computing center. For the computing power center cluster Taiwan Computing Center Power at any moment For the computing power center cluster The maximum power change rate of the computing center.

[0035] Optionally, the real-time optimization scheduling objective is:

[0036] in, To optimize the target in real time, For response precision weights, For computing power center clusters Adjust power demand in real time. The number of computing centers. For the computing power center cluster Taiwan Computing Center Power at any moment For stability weights, For the computing power center cluster Taiwan Computing Center Power at any given moment.

[0037] The current daytime optimization scheduling objective is:

[0038] in, To optimize the target in advance, As a weighted average of energy costs, For the energy cost of computing center clusters, As carbon cost weight, The total carbon emissions of the computing center cluster. For the cost of computing power loss, , This represents the number of tasks completed in the previous day. For the first The transmission latency of each task For the first The computational processing latency of each task. This is the delay compensation coefficient.

[0039] Optionally, obtaining the day-ahead scheduling data for historical periods based on the day-ahead computing power demand data of the computing center cluster, based on multidimensional coupling constraints and day-ahead optimization scheduling objectives, includes: constructing a day-ahead optimization model based on multidimensional coupling constraints and day-ahead optimization scheduling objectives; modeling the uncertainty variables in the day-ahead optimization model using a moment-information-based bibliometric optimization framework to construct a fuzzy set composed of confidence interval constraints of first-order moments and second-order moments; performing multimodal processing on the fuzzy set based on task type based on the preset task priority conservatism coefficients for real-time and non-real-time tasks to obtain a multimodal uncertainty fuzzy set; transforming the day-ahead optimization model containing the multimodal uncertainty fuzzy set into a deterministic optimization model using duality theory and linear matrix inequality transformation methods; and solving the deterministic optimization model based on the day-ahead computing power demand data of the computing center cluster for historical periods to obtain the day-ahead scheduling data for historical periods.

[0040] Optionally, the solution of the deterministic optimization model includes: using a particle swarm optimization algorithm to solve the deterministic optimization model to obtain the current particle swarm optimal solution, and using a beetle whisker algorithm to perform a local search based on the current particle swarm optimal solution to obtain the current beetle whisker optimal solution; using the current beetle whisker optimal solution as the initial optimal solution of the particle swarm optimization algorithm and repeating the above steps until a preset iteration termination condition is reached, and using the current beetle whisker optimal solution as the final solution.

[0041] Optionally, obtaining the real-time scheduling data for historical periods based on the real-time power grid frequency and real-time computing power demand data of the computing center cluster, and based on multi-dimensional coupling constraints and real-time optimization scheduling objectives, includes: obtaining the real-time scheduling data for historical periods by calling a pre-trained real-time scheduling model based on the real-time power grid frequency and real-time computing power demand data of the computing center cluster; wherein, the real-time scheduling model is constructed based on a dual-deep Q-network framework and multi-dimensional coupling constraints and real-time optimization scheduling objectives; the state space of the real-time scheduling model is the real-time power grid frequency and real-time computing power demand data, the action space is the power adjustment level of each computing center in the computing center cluster, and the reward function is the negative value of the real-time optimization scheduling objective.

[0042] Optionally, the pre-trained real-time scheduling model employs a random-first experience replay method for experience replay during training; wherein, during experience replay, the experience... The priority is , For experience Time difference error, Preset as a positive number; experience The normalized importance weights are , The size of the experience pool. For experience The sampling probability, As a regulating factor, The maximum importance weight for each sampled experience; and the storage and sampling of experiences based on the SumTree structure.

[0043] Optionally, obtaining real-time scheduling data for historical periods based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and based on multi-dimensional coupling constraints and real-time optimization scheduling objectives, includes: obtaining real-time scheduling data for historical periods in the current scenario based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and based on multi-dimensional coupling constraints and real-time optimization scheduling objectives; the day-ahead adjustment capability data includes day-ahead adjustment capability data for each scenario; the day-ahead scheduling data for historical periods includes day-ahead scheduling data for historical periods in each scenario; obtaining day-ahead correction data based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data for historical periods includes: obtaining day-ahead correction data for the current scenario based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data for historical periods in the current scenario.

[0044] Optionally, an emergency assessment module is also included. The emergency assessment module is used to: acquire in real time the scenario of the power grid connected to the computing center cluster, and when the scenario is an extreme scenario, acquire computing power demand data of the computing center cluster before a preset time period, and obtain the adjustment capability data of the computing center cluster before a preset time period based on the computing power demand data before the preset time period and multi-dimensional coupling constraints; acquire the correction data of the computing center cluster before the preset time period, and correct the adjustment capability data before the preset time period based on the day-ahead correction data, so as to obtain the emergency adjustability assessment result of the computing center cluster.

[0045] In a third aspect, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method for evaluating the adjustable capability of a computing center cluster.

[0046] In a fourth aspect, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method for evaluating the adjustable capability of a computing center cluster.

[0047] Compared with the prior art, the present invention has the following beneficial effects: This invention presents a method for assessing the adjustability of computing center clusters. In the day-ahead phase, based on day-ahead computing demand data, it obtains day-ahead adjustability data of the computing center cluster based on multi-dimensional coupling constraints. This comprehensively considers the coupling of multiple constraints, including quality of service constraints, computing power economic constraints, green environment constraints, and safe operation constraints, ensuring the assessment results closely align with actual operating scenarios and providing comprehensive support for cluster adjustment strategy formulation. Simultaneously, it acquires day-ahead corrected data for the computing center cluster and corrects the day-ahead adjustability data accordingly, obtaining the overall adjustability assessment result. The day-ahead corrected data is obtained from historical real-time scheduling data, day-ahead scheduling data, and day-ahead adjustability data. This achieves bidirectional collaborative optimization at the day-ahead and real-time levels. Through comprehensive planning at the day-ahead level, the comprehensiveness of the assessment results is ensured. Through fine-grained real-time sensing at the real-time level, the assessment results no longer rely solely on offline static data but dynamically reflect random fluctuations in actual operation, ensuring the accuracy of the assessment results. This invention provides accurate and reliable adjustability assessments for computing center clusters participating in grid interaction, effectively supporting the efficient operation and optimized scheduling of computing center clusters. Attached Figure Description

[0048] Figure 1 This is a flowchart of the method for evaluating the adjustable capability of a computing center cluster according to an embodiment of the present invention.

[0049] Figure 2This is a block diagram of the adjustable capability assessment system for computing power center clusters according to an embodiment of the present invention. Detailed Implementation

[0050] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0051] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0052] The present invention will now be described in further detail with reference to the accompanying drawings: See Figure 1 In one embodiment of the present invention, a method for assessing the adjustability of a computing center cluster is provided. This method aims to overcome the technical defects of existing methods for assessing the adjustability of computing center clusters, such as insufficient consideration of multiple constraints, poor coordination of time scales, weak ability to cope with uncertainties, and low real-time assessment accuracy. This method can effectively improve the comprehensiveness, accuracy, and real-time performance of the assessment results of the adjustability of computing center clusters, and provide reliable technical support for computing center clusters to participate in grid interaction and operate efficiently on their own.

[0053] Specifically, the method for evaluating the adjustability of a computing center cluster according to the present invention includes the following steps: S1: Obtain the day-ahead computing power demand data of the computing power center cluster, and based on the day-ahead computing power demand data, obtain the day-ahead adjustment capability data of the computing power center cluster based on multi-dimensional coupling constraints. Among them, the multi-dimensional coupling constraints include service quality constraints, computing power economy constraints, green environment constraints, and safe operation constraints.

[0054] S2: Obtain the day-ahead correction data of the computing center cluster, and correct the day-ahead adjustment capability data based on the day-ahead correction data to obtain the adjustment capability assessment result of the computing center cluster.

[0055] The day-ahead correction data is obtained in the following ways: real-time scheduling data for historical periods is obtained based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and multi-dimensional coupling constraints and real-time optimization scheduling objectives; day-ahead scheduling data for historical periods is obtained based on day-ahead computing power demand data of the computing center cluster, and multi-dimensional coupling constraints and day-ahead optimization scheduling objectives; and day-ahead correction data is obtained based on real-time scheduling data, day-ahead scheduling data, and day-ahead regulation capability data for historical periods.

[0056] Explanatoryly, the term "day before" is used here primarily for ease of description and understanding. Essentially, it represents a period of time (usually 24 hours, hence defined as "day before"), meaning the evaluation cycle is based on a 24-hour period. Of course, setting this period to 48 or 12 hours also applies. For example, for the day-before adjustment capability data of a computing center cluster, a rolling verification window (generally set to 4 hours) can be set. This involves changing the time period represented by the day before, repeating the above process, and comparing it with existing data for verification.

[0057] Explanatoryly, the historical time period also represents a period of time (generally set to 500ms). It should be noted that the day-ahead correction data includes day-ahead correction data corresponding to multiple historical time periods. Generally, the day-ahead correction data calculated from all historical time periods of the previous day is used as the current day-ahead correction data. The purpose of this day-ahead correction data is to use the deviation between the day-ahead scheduling data of the previous day and the real-time scheduling data as the basis for analyzing and adjusting today's day-ahead adjustment capability data, thereby achieving rolling correction.

[0058] For example, the adjustable capability assessment method of the computing power center cluster of the present invention can be divided into a day-ahead layer and a real-time layer in specific applications, and the two layers achieve deep collaboration through data interaction.

[0059] Specifically, the day-ahead layer uses a 24-hour basic evaluation cycle and simultaneously sets a 4-hour rolling verification window. It acquires and, based on day-ahead computing power demand data and multi-dimensional coupling constraints, obtains day-ahead adjustment capability data for the computing center cluster. It then obtains day-ahead scheduling data based on the day-ahead computing power demand data, multi-dimensional coupling constraints, and day-ahead optimization scheduling objectives. This day-ahead adjustment capability data and day-ahead scheduling data are transmitted to the real-time layer via an encrypted interface. The real-time layer uses 500ms as the minimum evaluation unit. Based on the real-time grid frequency and real-time computing power demand data of the computing center cluster, it obtains real-time scheduling data based on multi-dimensional coupling constraints and real-time optimization scheduling objectives. It also obtains day-ahead correction data based on historical real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data. For example, the real-time layer packages and feeds back the day-ahead correction data every minute, providing a basis for the next rolling correction by the day-ahead layer.

[0060] This invention presents a method for assessing the adjustability of computing center clusters. In the day-ahead phase, based on day-ahead computing demand data, it obtains day-ahead adjustability data of the computing center cluster based on multi-dimensional coupling constraints. This comprehensively considers the coupling of multiple constraints, including quality of service constraints, computing power economic constraints, green environment constraints, and safe operation constraints, ensuring the assessment results closely align with actual operating scenarios and providing comprehensive support for cluster adjustment strategy formulation. Simultaneously, it acquires day-ahead corrected data for the computing center cluster and corrects the day-ahead adjustability data accordingly, obtaining the overall adjustability assessment result. The day-ahead corrected data is obtained from historical real-time scheduling data, day-ahead scheduling data, and day-ahead adjustability data. This achieves bidirectional collaborative optimization at the day-ahead and real-time levels. Through comprehensive planning at the day-ahead level, the comprehensiveness of the assessment results is ensured. Through fine-grained real-time sensing at the real-time level, the assessment results no longer rely solely on offline static data but dynamically reflect random fluctuations in actual operation, ensuring the accuracy of the assessment results. This invention provides accurate and reliable adjustability assessments for computing center clusters participating in grid interaction, effectively supporting the efficient operation and optimized scheduling of computing center clusters.

[0061] In one possible implementation, the day-ahead computing power requirement data includes the number of day-ahead tasks, the computing power requirement of each day-ahead task, and the data volume of each day-ahead task; the day-ahead adjustment capability data includes the day-ahead power adjustment range, the day-ahead power adjustment change rate range, and the day-ahead power adjustment limit duration; the day-ahead correction data includes one or more of the following: the deviation between real-time adjustment power and day-ahead adjustment power, the constraint boundary triggering frequency of real-time adjustment power, and the real-time change of computing power service quality.

[0062] Interpretively, real-time changes in computing power service quality generally refer to the latency of the computing power center cluster. The deviation between real-time adjusted power and day-ahead adjusted power directly reflects the degree of deviation between day-ahead predictions and actual operating conditions. A larger deviation indicates a less accurate baseline for day-ahead adjustment capability assessment, which can be corrected to reduce this prediction error. The frequency of triggering the constraint boundaries of real-time adjusted power reveals whether the adjustment range boundaries given in the day-ahead assessment are reasonable. Frequent triggering of the upper or lower limits indicates that the adjustable boundaries defined in the day-ahead assessment are too optimistic, and the boundaries need to be tightened in subsequent assessments to avoid the risk of exceeding limits. Real-time changes in computing power service quality quantify the actual impact of power adjustment on core computing power services. If service quality significantly declines, it indicates that the consideration of service quality constraints in the day-ahead assessment is too lenient, and the weight or strictness of these constraints needs to be increased.

[0063] In one possible implementation, the quality of service constraint is:

[0064] in, The total latency of the computing center cluster. This represents the number of tasks completed in the previous day. For the first The priority weight of each task. For the first The transmission latency of each task , For the first The amount of data per task For the first Network bandwidth for each task For bandwidth utilization, For the first The computational processing latency of each task. , For the first The computing power requirements for each task. For the first The allocation of computing power in the computing center for each task. For computing power utilization, This represents the maximum tolerable latency for the computing center cluster.

[0065] Interpretive, the quality of service constraints are mainly to ensure that power regulation does not sacrifice the quality of service of the service, while latency represents the quality of service of computing power, and is regarded as a constraint condition that must be met. Latency mainly considers transmission latency and processing latency, and takes into account task priority weights to achieve differentiated constraint control.

[0066] In one possible implementation, the computing power economy constraint is:

[0067] in, For effective computing power energy efficiency ratio, The total computing power output of the computing center cluster. This represents the total energy consumption of the computing center cluster. For power efficiency, This serves as the baseline energy efficiency threshold.

[0068] Explanatory, the economic efficiency of computing power is characterized by the computing power energy efficiency ratio, i.e. (Power efficiency) The corrected effective energy efficiency is the core indicator. The unit is TOPS / kWh. It reflects the proportion of energy consumption in systems such as cooling.

[0069] In one possible implementation, the green environment constraint is:

[0070] in, The total carbon emissions of the computing center cluster. The direct carbon emissions of the computing power center cluster, For the indirect carbon emissions of computing center clusters, For the quantity of energy types, For the first Consumption of energy sources, For the first Carbon emission factors of energy-like substances The electricity purchased by the computing center cluster from the power grid. for The carbon emission factor of the power grid at any time For carbon emission allowances.

[0071] Interpretive methods are employed, using carbon emission costs to characterize green environmental indicators and combining them with carbon trading prices to establish green environmental constraints, ensuring that the operation of computing center clusters meets green environmental requirements.

[0072] In one possible implementation, the safe operation constraint is: ,

[0073] in, For the computing power center cluster Taiwan Computing Center Power at any moment and For the computing power center cluster The upper and lower power limits of a computing center. For the computing power center cluster Taiwan Computing Center Power at any moment For the computing power center cluster The maximum power change rate of the computing center.

[0074] Explanatory, the constraints on the safe operation of computing center clusters mainly include power limits and the maximum power change rate of computing centers, in order to ensure the safe and stable operation of computing center clusters.

[0075] In one possible implementation, the real-time optimization scheduling objective is:

[0076] in, To optimize the target in real time, For response precision weights, For computing power center clusters Adjust power demand in real time. for, For the computing power center cluster Taiwan Computing Center Power at any moment For stability weights, For the computing power center cluster Taiwan Computing Center Power at any given moment.

[0077] The current daytime optimization scheduling objective is:

[0078] in, To optimize the target in advance, As a weighted average of energy costs, For the energy cost of computing center clusters, As carbon cost weight, The total carbon emissions of the computing center cluster. For the cost of computing power loss, , This represents the number of tasks completed in the previous day. For the first The transmission latency of each task For the first The computational processing latency of each task. This is the delay compensation coefficient. For example, , It can be dynamically adjusted according to needs. The delay compensation factor is the amount of compensation to be paid for each task for every second of delay, which can be determined according to the penalties in the Service Level Agreement (SLA).

[0079] Interpretive, and taking into account the characteristics of a layered architecture, a differentiated quantitative objective function is established, and a balance of multi-objective optimization is achieved through weighting coefficients. Specifically, the day-ahead layer's objective considers economic and environmental factors to achieve comprehensiveness, while the real-time layer's objective primarily considers response stability and accuracy to achieve precision.

[0080] In one possible implementation, obtaining the day-ahead scheduling data for historical periods based on the day-ahead computing power demand data of the computing center cluster, and based on multidimensional coupling constraints and day-ahead optimization scheduling objectives, includes: constructing a day-ahead optimization model based on multidimensional coupling constraints and day-ahead optimization scheduling objectives; modeling the uncertainty variables in the day-ahead optimization model using a moment-information-based bibliometric optimization framework to construct a fuzzy set composed of confidence interval constraints of first-order moments and second-order moments; performing multimodal processing on the fuzzy set based on task type, using preset task priority conservatism coefficients for real-time and non-real-time tasks, to obtain a multimodal uncertainty fuzzy set; transforming the day-ahead optimization model containing the multimodal uncertainty fuzzy set into a deterministic optimization model using duality theory and linear matrix inequality transformation methods; and solving the deterministic optimization model based on the day-ahead computing power demand data of the computing center cluster for historical periods to obtain the day-ahead scheduling data for historical periods.

[0081] Interpretive approaches are employed to address the different needs of the day-ahead and real-time layers, using differentiated solution methods. For the day-ahead layer, the core requirement is to address the uncertainties in parameters such as computing power requirements and energy prices, aiming to generate a reliable baseline power and adjustment range. Therefore, a sub-Bruker optimization and heuristic algorithm are used.

[0082] Specifically, the first step is data preparation, including: (1) day-ahead computing power demand data: computing power demand, task volume, task data volume and task priority, mainly using historical task log data, obtained through prediction (or obtained from existing data center management platforms). (2) Energy price information: grid electricity price and energy price, mainly obtained directly from the electricity market or energy market; (3) Carbon emission factors: energy carbon emission factors and grid carbon emission factors, from publicly available grid data and energy supplier reports; (4) Service level agreement: maximum tolerable latency and compensation coefficient, from user contracts. (5) Day-ahead correction data fed back from the real-time layer.

[0083] Secondly, uncertainty is modeled. For various uncertain parameters such as computing power demand and grid electricity prices, a partial Bruker optimization framework based on moment information is used for modeling, assuming an uncertainty vector. (including computing power load) 、 A fuzzy set of data (such as electricity prices and energy prices). P The fuzzy set is constrained by confidence intervals of the first moment (mean) and the second moment (covariance):

[0084] in, The set of all possible probability distributions. , This represents the upper bound of the covariance matrix, reflecting the range of fluctuations in uncertainty. For distribution P The expectations below.

[0085] Based on the traditional single fuzzy set description, a multimodal uncertainty modeling method based on computing task priority is proposed. The computing tasks are divided into real-time and non-real-time types, thus the upper bound of the weighted covariance matrix is ​​modified as follows: ,in, and These are the prediction covariance matrices for uncertainty parameters related to real-time and non-real-time tasks, respectively. and Let be the task priority conservatism coefficient, and satisfy 0 < 0. < The coefficient ≤1 is obtained by fitting the risk of task interruption loss from historical data, making the model more conservative for fluctuations in real-time tasks and more lenient for non-real-time tasks. Meanwhile, day-ahead optimization models containing multimodal uncertainty fuzzy sets are difficult to solve directly due to uncertain constraints. They can be transformed into efficiently solvable convex optimization problems (deterministic optimization models) using duality theory and linear matrix inequalities, thus reducing problem complexity.

[0086] In one possible implementation, solving the deterministic optimization model includes: using a particle swarm optimization algorithm to solve the deterministic optimization model to obtain the current particle swarm optimal solution, and using a beetle whisker algorithm to perform a local search based on the current particle swarm optimal solution to obtain the current beetle whisker optimal solution; using the current beetle whisker optimal solution as the initial optimal solution of the particle swarm optimization algorithm and repeating the above steps until a preset iteration termination condition is reached, and using the current beetle whisker optimal solution as the final solution.

[0087] Explained, while Particle Swarm Optimization (PSO) is a traditional heuristic algorithm with fast convergence speed, it is prone to getting trapped in local optima. Therefore, this invention combines PSO with the beetle whisker algorithm. On one hand, it utilizes the implicit parallel search strategy of PSO to perform a global search of the solution space; on the other hand, it leverages the beetle whisker algorithm to perform further local optimization on the results. This approach fully utilizes the global and local search capabilities of both algorithms, avoiding the weakness of traditional PSO in getting trapped in local optima, and comprehensively improving performance.

[0088] Specifically, the detailed implementation steps include: Step 1: Set the particle swarm size, determine parameters such as learning factor, inertia factor, initial velocity, and position, calculate the fitness function value for each particle based on the objective function, and calculate the optimal position of each individual particle at this point. Pi ( t and population optimal position Pg ( tSpecifically, each particle represents a set of scheduling data for the next 24 hours, i.e., generated by the first... k Taiwan Computing Center t Moment power Electricity purchased by the computing center from the power grid and the j Consumption of energy types The solution vector is formed. The fitness function is set as the day-ahead optimization scheduling objective.

[0089] Step 2: Update the particle velocity and position using the particle swarm optimization algorithm's update rules, and calculate the fitness function value corresponding to the new position; if this value is better than the initial individual optimal value, then update the individual's optimal position. Pi ( t If this value is better than the population optimum, then update the population optimum position. Pg ( t ).

[0090] Step 3: Use the beetle whisker algorithm to find the optimal solution for the particle swarm. Pg ( t Perform a local search, setting the initial solution for the longhorn beetle to be... Pg ( t and the initial optimal solution Pgbest = Pg ( t ).

[0091] Step 4: If the termination condition for the longhorn beetle search is met at this point, proceed to step 7; otherwise, proceed to step 5.

[0092] Step 5: Randomly generate the search direction of the longhorn beetle, and update the positions of the left and right whiskers according to the longhorn beetle whisker algorithm update rules. Pleft and Pright Calculate the corresponding fitness function values ​​to determine the next position of the longhorn beetle.

[0093] Step 6: If the updated solution satisfies the acceptance principle, then adopt it as the current solution, and replace the optimal solution with the current solution. Pgbest Then proceed to step 4; if not satisfied, replace the current solution with the optimal solution of the longhorn beetle, and then proceed to step 4.

[0094] Step 7: If the algorithm reaches the iteration limit at this point, it outputs the optimal solution and terminates the algorithm; otherwise, it jumps to step 2 to continue iterating.

[0095] In one possible implementation, obtaining real-time scheduling data for historical periods based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and based on multi-dimensional coupling constraints and real-time optimization scheduling objectives, includes: obtaining real-time scheduling data for historical periods by calling a pre-trained real-time scheduling model based on the real-time grid frequency and real-time computing power demand data of the computing center cluster for historical periods; wherein, the real-time scheduling model is constructed based on a dual-deep Q-network framework and multi-dimensional coupling constraints and real-time optimization scheduling objectives; the state space of the real-time scheduling model is the real-time grid frequency and real-time computing power demand data, the action space is the power adjustment level of each computing center in the computing center cluster, and the reward function is the negative value of the real-time optimization scheduling objective.

[0096] The core requirement of the interpretable real-time layer is to achieve rapid evaluation at the second / minute level. However, the inputs of the real-time layer (such as computing load and power grid frequency deviation) have strong randomness and rapid fluctuations, making it difficult for traditional optimization methods based on precise mathematical models to be established and solved quickly online.

[0097] The value-iterative DQN (Deep Q-Network) framework has extremely low forward propagation computational overhead, meeting the requirements for millisecond-level decision response. However, pure DQN suffers from Q-value overestimation and low sample utilization efficiency. Therefore, this implementation adopts DDQN (Dual Deep Q-Network).

[0098] Specifically, the core idea of ​​DQN is to use deep neural networks to approximate value functions. ,in Indicates state, It represents actions. It learns a value function to evaluate the expected reward of taking various actions in a given state, thereby optimizing action selection and policy. Unlike traditional Q-learning, DQN uses deep neural networks to represent the value function, allowing it to handle more complex state and action spaces. The update rule of Q-learning is as follows:

[0099] in, For instant rewards, As a discount factor, For learning rate, For the next state and the next action value, This is the action space.

[0100] Therefore, the objective function of DQN is:

[0101] in, For the first The loss function for the next iteration. For the expectation operator, These are the parameters of the current network. These are the parameters of the target network.

[0102] The ultimate goal of updating the DQN algorithm is to make... Approaching Since the Temporal Difference (TD) error target itself includes the output of the neural network, if the current Q-network is directly used to calculate the target value during training, the target value will constantly change as the network parameters are updated, leading to instability in the training process. To solve this problem, DQN uses the idea of ​​a target network, temporarily fixing the Q-network in the TD target while using the original training network. Used to calculate the original loss function In The terms are updated using the normal gradient descent method. Target network Used to calculate the original loss function Item, of which This represents the parameters in the target network.

[0103] There are two methods for updating the target network: one is called hard update, which updates every... One approach is to update the target network every step; another is the soft update proposed in DDQN, which ensures that the target network is updated in each iteration. This is achieved by using a convex combination of the current network parameters and the target network parameters to update the network, i.e., the parameters of the target network. Will use current network parameters Update as follows, where the soft-interval update coefficients are... :

[0104] When selecting actions, adopt Strategy: Balancing exploration and exploitation. Parameters It is determined whether, at each time step, the action is selected through exploration (randomly selecting an action with a certain probability) or through exploitation (selecting the action with the highest current Q value with a certain probability). The update formula is:

[0105] Where EPS_END represents The minimum value indicates that the later stages of learning focus more on utilizing known strategies; EPS_START represents The initial value of EPS_DECAY is the probability of randomly selecting an action at the beginning. It is generally set to a high value (such as 0.9 or 1.0) to encourage exploration in the early stages of learning. EPS_DECAY is the decay rate, which controls the... The rate at which the value decreases from the initial value to the minimum value; steps_done is the number of steps or decisions made so far.

[0106] The standard DQN algorithm usually leads to... Overestimation of the target value, especially in cases of large action spaces, can lead to training instability and performance degradation. Double DQN addresses this issue by introducing two independent neural networks. The core idea of ​​Double DQN is to... The estimation of the target value is separated from the neural network that selects the action. Specifically, at each update, the neural network that selects the action (called the action selection network) is used to choose the next action, while the neural network that estimates the target value is used to select the next action. A value-based neural network (called the target network) is used to evaluate the value of this action. Double DQN can provide more stable learning, reduce the volatility of the value function, and thus improve training stability. Its optimization objective is:

[0107] In one possible implementation, the pre-trained real-time scheduling model uses a random-first experience replay method to replay the experience during training.

[0108] Among them, during experience replay, experience The priority is , For experience Time difference error, Preset as a positive number; experience The normalized importance weights are , The size of the experience pool. For experience The sampling probability, As a regulating factor, The maximum importance weight for each sampled experience; and the storage and sampling of experiences based on the SumTree structure.

[0109] Interpretive, priority-based experience replay builds upon the original uniform distribution sampling method by assigning higher priority weights to experience values ​​that are more valuable for training. This increases their probability pairs when sampled, resulting in better performance. The TD error is used to measure the priority of experiences, and random-priority experience replay is employed. The formula for calculating the TD error is: Random-priority experience replay ensures that the probability of each experience being selected from the experience pool is monotonically related to its TD error value, but also guarantees that lower-priority experiences will still have a non-zero probability of being selected. For experience... The probability of being selected is calculated as follows: ,in Representing experience Priority, index This indicates the degree to which experience is replayed based on priority.

[0110] Explanatoryly, to correct for biases introduced by high-priority samples and prevent the model from overfitting due to using only partial experience, importance sampling is used, i.e., normalized importance weights are considered: This allows for model updates by reducing the weights of common samples. Suppose we need to draw from the experience pool... One piece of experience is to first define the cumulative priority range. Divided into The data structure is divided into several sequences, and then uniform random sampling is performed on each sequence. Finally, the corresponding experience is extracted from the data structure.

[0111] Explanatoryly, to improve sampling efficiency, a SumTree structure is used. The time complexity of sampling and updating is O(n log n). The SumTree structure is a special binary tree data structure primarily used for efficient probabilistic sampling. It consists of two parts: a data layer and a tree structure layer. The data layer stores the actual data elements, such as samples in the experience replay cache. Each node in the tree structure layer stores the cumulative sum of data from that node to all its child nodes, and the root node stores the sum of all data in the entire SumTree structure, which is essentially the priority value of the current experience.

[0112] By employing a random-first experience replay approach during training, we can improve the efficiency of sample utilization, accelerate learning convergence, pay more attention to rare events, and thus improve the robustness of the model.

[0113] In one possible implementation, obtaining real-time scheduling data for historical periods based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and based on multi-dimensional coupling constraints and real-time optimization scheduling objectives, includes: obtaining real-time scheduling data for historical periods under the current scenario based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and based on multi-dimensional coupling constraints and real-time optimization scheduling objectives; the day-ahead adjustment capability data includes day-ahead adjustment capability data under each scenario; the day-ahead scheduling data for historical periods includes day-ahead scheduling data for historical periods under each scenario; obtaining day-ahead correction data based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data for historical periods includes: obtaining day-ahead correction data for the current scenario based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data for historical periods under the current scenario.

[0114] Explanatory, the scenarios mentioned mainly refer to the operating scenarios of the power grid to which the computing center cluster is connected, generally including normal scenarios, peak shaving scenarios, frequency regulation scenarios, and emergency scenarios.

[0115] This scenario-based design avoids the evaluation distortion caused by mixing operational data from different operating conditions. For example, the high requirements for response speed in frequency regulation scenarios and the emphasis on regulation amplitude in peak shaving scenarios can be modeled and calibrated separately. Ultimately, the regulation capacity data for each scenario can be optimized independently, enabling the computing center cluster to provide more realistic and reliable regulation capacity support when facing diverse power grid regulation needs, significantly improving the adaptability and accuracy of the evaluation method.

[0116] For example, when the day-ahead layer distributes day-ahead adjustment capability data for each scenario, it distinguishes between scenarios by different values ​​of response accuracy weight and stability weight.

[0117] In one possible implementation, the method for assessing the adjustability of the computing power center cluster further includes: acquiring the scenario of the power grid connected to the computing power center cluster in real time; and when the scenario is an extreme scenario, acquiring computing power demand data of the computing power center cluster before a preset time period, and obtaining the adjustability data of the computing power center cluster before a preset time period based on the computing power demand data before the preset time period and multi-dimensional coupling constraints; acquiring the correction data of the computing power center cluster before the preset time period, and correcting the adjustability data before the preset time period based on the correction data before the day, to obtain the emergency adjustability assessment result of the computing power center cluster.

[0118] For example, the scenario is considered an extreme scenario when the power grid frequency fluctuation is ≥ ±0.5Hz, or the deviation between the real-time dispatch data and the day-ahead dispatch data exceeds 5%. The preset time period is generally set to 1 hour.

[0119] Explanatoryly, by introducing an emergency response mechanism for extreme scenarios, a rapid reassessment process with compressed timescales can be automatically triggered in emergency situations. When the power grid encounters sudden faults or severe disturbances, the day-ahead assessment results based on a 24-hour timescale are often severely outdated. By dynamically switching to a short-cycle emergency assessment mode, the system can provide timely and reliable adjustment capability boundaries under the current extreme scenarios, providing key decision-making basis for power grid dispatch in emergency situations and improving the reliability and timeliness of the computing center cluster's participation in power grid emergency response.

[0120] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0121] See Figure 2 In another embodiment of the present invention, a computing power center cluster adjustability assessment system is provided, which can be used to implement the above-mentioned computing power center cluster adjustability assessment method. Specifically, the computing power center cluster adjustability assessment system includes a day-ahead module and a correction module.

[0122] The day-ahead module is used to acquire day-ahead computing power demand data of the computing center cluster, and based on the day-ahead computing power demand data, obtains day-ahead adjustment capability data of the computing center cluster based on multi-dimensional coupling constraints. These multi-dimensional coupling constraints include quality of service constraints, computing power economic constraints, green environment constraints, and safe operation constraints. The correction module is used to acquire day-ahead correction data of the computing center cluster, and correct the day-ahead adjustment capability data based on the day-ahead correction data to obtain the adjustment capability assessment result of the computing center cluster. The day-ahead correction data is obtained in the following ways: based on the real-time grid frequency and real-time computing power demand data of the computing center cluster during historical periods, real-time scheduling data for historical periods is obtained based on multi-dimensional coupling constraints and real-time optimization scheduling objectives; day-ahead scheduling data for historical periods is obtained based on the day-ahead computing power demand data of the computing center cluster during historical periods, based on multi-dimensional coupling constraints and day-ahead optimization scheduling objectives; and day-ahead correction data is obtained based on the real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data for historical periods.

[0123] In one possible implementation, the day-ahead computing power requirement data includes the number of day-ahead tasks, the computing power requirement of each day-ahead task, and the data volume of each day-ahead task; the day-ahead adjustment capability data includes the day-ahead power adjustment range, the day-ahead power adjustment change rate range, and the day-ahead power adjustment limit duration; the day-ahead correction data includes one or more of the following: the deviation between real-time adjustment power and day-ahead adjustment power, the constraint boundary triggering frequency of real-time adjustment power, and the real-time change of computing power service quality.

[0124] In one possible implementation, the quality of service constraint is:

[0125] in, The total latency of the computing center cluster. This represents the number of tasks completed in the previous day. For the first The priority weight of each task. For the first The transmission latency of each task , For the first The amount of data per task For the first Network bandwidth for each task For bandwidth utilization, For the first The computational processing latency of each task. , For the first The computing power requirements for each task. For the first The allocation of computing power in the computing center for each task. For computing power utilization, This represents the maximum tolerable latency for the computing center cluster.

[0126] The economic constraints on computing power are as follows:

[0127] in, For effective computing power energy efficiency ratio, The total computing power output of the computing center cluster. This represents the total energy consumption of the computing center cluster. For power efficiency, This serves as the baseline energy efficiency threshold.

[0128] The green environment constraints are:

[0129] in, The total carbon emissions of the computing center cluster. The direct carbon emissions of the computing power center cluster, For the indirect carbon emissions of computing center clusters, For the quantity of energy types, For the first Consumption of energy sources, For the first Carbon emission factors of energy-like substances The electricity purchased by the computing center cluster from the power grid. for The carbon emission factor of the power grid at any time For carbon emission allowances.

[0130] The safe operation constraints are as follows: ,

[0131] in, For the computing power center cluster Taiwan Computing Center Power at any moment and For the computing power center cluster The upper and lower power limits of a computing center. For the computing power center cluster Taiwan Computing Center Power at any moment For the computing power center cluster The maximum power change rate of the computing center.

[0132] In one possible implementation, the real-time optimization scheduling objective is:

[0133] in, To optimize the target in real time, For response precision weights, For computing power center clusters Adjust power demand in real time. The number of computing centers. For the computing power center cluster Taiwan Computing Center Power at any moment For stability weights, For the computing power center cluster Taiwan Computing Center Power at any given moment.

[0134] The current daytime optimization scheduling objective is:

[0135] in, To optimize the target in advance, As a weighted average of energy costs, For the energy cost of computing center clusters, As carbon cost weight, The total carbon emissions of the computing center cluster. For the cost of computing power loss, , This represents the number of tasks completed in the previous day. For the first The transmission latency of each task For the first The computational processing latency of each task. This is the delay compensation coefficient.

[0136] In one possible implementation, obtaining the day-ahead scheduling data for historical periods based on the day-ahead computing power demand data of the computing center cluster, and based on multidimensional coupling constraints and day-ahead optimization scheduling objectives, includes: constructing a day-ahead optimization model based on multidimensional coupling constraints and day-ahead optimization scheduling objectives; modeling the uncertainty variables in the day-ahead optimization model using a moment-information-based bibliometric optimization framework to construct a fuzzy set composed of confidence interval constraints of first-order moments and second-order moments; performing multimodal processing on the fuzzy set based on task type, using preset task priority conservatism coefficients for real-time and non-real-time tasks, to obtain a multimodal uncertainty fuzzy set; transforming the day-ahead optimization model containing the multimodal uncertainty fuzzy set into a deterministic optimization model using duality theory and linear matrix inequality transformation methods; and solving the deterministic optimization model based on the day-ahead computing power demand data of the computing center cluster for historical periods to obtain the day-ahead scheduling data for historical periods.

[0137] In one possible implementation, solving the deterministic optimization model includes: using a particle swarm optimization algorithm to solve the deterministic optimization model to obtain the current particle swarm optimal solution, and using a beetle whisker algorithm to perform a local search based on the current particle swarm optimal solution to obtain the current beetle whisker optimal solution; using the current beetle whisker optimal solution as the initial optimal solution of the particle swarm optimization algorithm and repeating the above steps until a preset iteration termination condition is reached, and using the current beetle whisker optimal solution as the final solution.

[0138] In one possible implementation, obtaining real-time scheduling data for historical periods based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and based on multi-dimensional coupling constraints and real-time optimization scheduling objectives, includes: obtaining real-time scheduling data for historical periods by calling a pre-trained real-time scheduling model based on the real-time grid frequency and real-time computing power demand data of the computing center cluster for historical periods; wherein, the real-time scheduling model is constructed based on a dual-deep Q-network framework and multi-dimensional coupling constraints and real-time optimization scheduling objectives; the state space of the real-time scheduling model is the real-time grid frequency and real-time computing power demand data, the action space is the power adjustment level of each computing center in the computing center cluster, and the reward function is the negative value of the real-time optimization scheduling objective.

[0139] In one possible implementation, the pre-trained real-time scheduling model employs a random-first experience replay approach for experience replay during training; wherein, during experience replay, the experience... The priority is , For experience Time difference error, Preset as a positive number; experience The normalized importance weights are , The size of the experience pool. For experience The sampling probability, As a regulating factor, The maximum importance weight for each sampled experience; and the storage and sampling of experiences based on the SumTree structure.

[0140] In one possible implementation, obtaining real-time scheduling data for historical periods based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and based on multi-dimensional coupling constraints and real-time optimization scheduling objectives, includes: obtaining real-time scheduling data for historical periods under the current scenario based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and based on multi-dimensional coupling constraints and real-time optimization scheduling objectives; the day-ahead adjustment capability data includes day-ahead adjustment capability data under each scenario; the day-ahead scheduling data for historical periods includes day-ahead scheduling data for historical periods under each scenario; obtaining day-ahead correction data based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data for historical periods includes: obtaining day-ahead correction data for the current scenario based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data for historical periods under the current scenario.

[0141] In one possible implementation, the computing power center cluster adjustability assessment system further includes an emergency assessment module. This module is used to acquire real-time scenarios of the power grid connected to the computing power center cluster, and when the scenarios are extreme, to acquire computing power demand data for a preset time period prior to the computing power center cluster, and based on the computing power demand data for the preset time period prior to the preset time period, to obtain the adjustability data for the computing power center cluster for the preset time period prior to the preset time period based on multi-dimensional coupling constraints; to acquire the corrected data for the computing power center cluster for the preset time period prior to the preset time period, and to correct the adjustability data for the preset time period prior to the previous day's corrected data, thereby obtaining the emergency adjustability assessment result of the computing power center cluster.

[0142] All relevant content of each step involved in the aforementioned embodiments of the computing power center cluster adjustability assessment method can be referenced to the functional description of the corresponding functional module of the computing power center cluster adjustability assessment system in the embodiments of the present invention, and will not be repeated here.

[0143] The module division in this embodiment of the invention is illustrative and represents only one logical functional division. In actual implementation, other division methods may be used. Furthermore, the functional modules in the various embodiments of the invention can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0144] In another embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions from the computer storage medium to achieve a corresponding method flow or function. The processor described in this embodiment of the present invention can be used in the operation of a method for evaluating the adjustable capability of a computing center cluster.

[0145] In another embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the computing center cluster adjustability assessment method in the above embodiments.

[0146] 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.

[0147] 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, and 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 illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0148] 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.

[0149] 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.

[0150] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for evaluating adjustable capacity of a hash center cluster, characterized in that, include: Obtain the day-ahead computing power demand data of the computing power center cluster, and based on the day-ahead computing power demand data, obtain the day-ahead adjustment capability data of the computing power center cluster based on multi-dimensional coupling constraints; among which, multi-dimensional coupling constraints include service quality constraints, computing power economy constraints, green environment constraints, and safe operation constraints. Obtain the day-ahead correction data of the computing center cluster, and correct the day-ahead adjustment capability data based on the day-ahead correction data to obtain the adjustment capability assessment result of the computing center cluster. The day-ahead correction data is obtained in the following ways: real-time scheduling data for historical periods is obtained based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and multi-dimensional coupling constraints and real-time optimization scheduling objectives; day-ahead scheduling data for historical periods is obtained based on day-ahead computing power demand data of the computing center cluster, and multi-dimensional coupling constraints and day-ahead optimization scheduling objectives; and day-ahead correction data is obtained based on real-time scheduling data, day-ahead scheduling data, and day-ahead regulation capability data for historical periods.

2. The computing center cluster adjustable capability evaluation method according to claim 1, characterized in that, The daily computing power requirement data includes the number of daily tasks, the computing power requirement of each daily task, and the data volume of each daily task; the daily adjustment capability data includes the daily power adjustment range, the daily power adjustment rate range, and the daily power adjustment limit duration. The day-ahead correction data includes one or more of the following: the deviation between the real-time adjustment power and the day-ahead adjustment power, the constraint boundary triggering frequency of the real-time adjustment power, and the real-time changes in the computing power service quality. 3.The method of claim 1, wherein, The service quality constraints are as follows: in, The total latency of the computing center cluster. This represents the number of tasks completed in the previous day. For the first The priority weight of each task. For the first The transmission latency of each task , For the first The amount of data per task For the first Network bandwidth for each task For bandwidth utilization, For the first The computational processing latency of each task. , For the first The computing power requirements for each task. For the first The allocation of computing power in the computing center for each task. For computing power utilization, This represents the maximum tolerable latency for the computing center cluster. The economic constraints on computing power are as follows: in, For effective computing power energy efficiency ratio, The total computing power output of the computing center cluster. This represents the total energy consumption of the computing center cluster. For power efficiency, The baseline energy efficiency threshold; The green environment constraints are: in, The total carbon emissions of the computing center cluster. The direct carbon emissions of the computing power center cluster, For the indirect carbon emissions of computing center clusters, For the quantity of energy types, For the first Consumption of energy sources, For the first Carbon emission factors of energy-like substances The electricity purchased by the computing center cluster from the power grid. for The carbon emission factor of the power grid at any time For carbon emission quotas; The safe operation constraints are as follows: ， in, For the computing power center cluster Taiwan Computing Center Power at any moment and For the computing power center cluster The upper and lower power limits of a computing center. For the computing power center cluster Taiwan Computing Center Power at any moment For the computing power center cluster The maximum power change rate of the computing center.

4. The method for evaluating the adjustability of a computing center cluster according to claim 1, characterized in that, The real-time optimization scheduling objective is: in, To optimize the target in real time, For response precision weights, For computing power center clusters Adjust power demand in real time. The number of computing centers. For the computing power center cluster Taiwan Computing Center Power at any moment For stability weights, For the computing power center cluster Taiwan Computing Center Power at any given moment; The current daytime optimization scheduling objective is: in, To optimize the target in advance, As a weighted average of energy costs, For the energy cost of computing center clusters, As carbon cost weight, The total carbon emissions of the computing center cluster. For the cost of computing power loss, , This represents the number of tasks completed in the previous day. For the first The transmission latency of each task For the first The computational processing latency of each task. This is the delay compensation coefficient.

5. The method for evaluating the adjustability of a computing center cluster according to claim 1, characterized in that, The process of obtaining the day-ahead scheduling data for historical periods based on the day-ahead computing power demand data of the computing center cluster, and grounded in multidimensional coupling constraints and day-ahead optimization scheduling objectives, includes: A day-ahead optimization model is constructed based on multidimensional coupling constraints and day-ahead optimization scheduling objectives. Uncertain variables in the day-ahead optimization model are modeled for uncertainty using a split-Brow bar optimization framework based on moment information, and a fuzzy set consisting of confidence interval constraints of first-order moments and second-order moments is constructed. Based on the task priority conservatism coefficients preset for real-time and non-real-time tasks, the fuzzy set is subjected to multimodal processing based on task type to obtain a multimodal uncertain fuzzy set. The day-ahead optimization model containing multimodal uncertainty fuzzy sets is transformed into a deterministic optimization model using duality theory and linear matrix inequality transformation method. Based on the day-ahead computing power demand data of the computing center cluster in historical periods, the deterministic optimization model is solved to obtain the day-ahead scheduling data of historical periods.

6. The method for evaluating the adjustability of a computing center cluster according to claim 5, characterized in that, The solution to the deterministic optimization model includes: The particle swarm optimization algorithm is used to solve the deterministic optimization model to obtain the current particle swarm optimal solution, and the beetle whisker algorithm is used to perform local search based on the current particle swarm optimal solution to obtain the current beetle whisker optimal solution. The current optimal solution of the longhorn beetle is used as the initial optimal solution of the particle swarm optimization algorithm, and the above steps are repeated until the preset iteration termination condition is reached. The current optimal solution of the longhorn beetle is then used as the final solution.

7. The method for evaluating the adjustability of a computing center cluster according to claim 1, characterized in that, The process of obtaining real-time scheduling data for historical periods based on real-time power grid frequency and real-time computing power demand data from the computing center cluster, and grounded in multi-dimensional coupling constraints and real-time optimization scheduling objectives, includes: Based on the real-time power grid frequency and real-time computing power demand data of the computing center cluster during historical periods, the real-time scheduling data of historical periods is obtained by calling the pre-trained real-time scheduling model. The real-time scheduling model is based on a dual-deep Q-network framework and is constructed based on multi-dimensional coupling constraints and real-time optimization scheduling objectives. The state space of the real-time scheduling model is the real-time power grid frequency and real-time computing power demand data, the action space is the power adjustment level of each computing center in the computing center cluster, and the reward function is the negative value of the real-time optimization scheduling objective.

8. The method for evaluating the adjustability of a computing center cluster according to claim 7, characterized in that, The pre-trained real-time scheduling model uses a random-first experience replay method to replay the experience during training. Among them, during experience replay, experience The priority is , For experience Time difference error, Preset as a positive number; experience The normalized importance weights are , The size of the experience pool. For experience The sampling probability, As a regulating factor, The maximum importance weight for each sampled experience; and the storage and sampling of experiences based on the SumTree structure.

9. The method for evaluating the adjustability of a computing center cluster according to claim 1, characterized in that, The process of obtaining real-time scheduling data for historical periods based on real-time power grid frequency and real-time computing power demand data from the computing center cluster, and grounded in multi-dimensional coupling constraints and real-time optimization scheduling objectives, includes: Based on the real-time power grid frequency and real-time computing power demand data of the computing center cluster during historical periods, real-time scheduling data of historical periods under the current scenario is obtained based on multi-dimensional coupling constraints and real-time optimization scheduling objectives. The day-ahead adjustment capability data includes day-ahead adjustment capability data under various scenarios; the day-ahead scheduling data for historical periods includes day-ahead scheduling data for historical periods under various scenarios. The step of obtaining day-ahead correction data based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data from historical time periods includes obtaining day-ahead correction data for the current scenario based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data from historical time periods in the current scenario.

10. The method for evaluating the adjustability of a computing center cluster according to claim 1, characterized in that, Also includes: The system acquires real-time data on the power grid connected to the computing center cluster, and when the scenario is an extreme scenario, it acquires computing power demand data of the computing center cluster before a preset time period, and obtains the adjustment capability data of the computing center cluster before a preset time period based on the computing power demand data before the preset time period and multi-dimensional coupling constraints. Obtain the pre-set adjustment data of the computing center cluster, and adjust the pre-set adjustment capability data based on the pre-set adjustment data to obtain the emergency adjustability assessment result of the computing center cluster.

11. A system for evaluating the adjustable capability of a computing center cluster, characterized in that, include: The day-ahead module is used to obtain the day-ahead computing power demand data of the computing power center cluster, and based on the day-ahead computing power demand data, obtain the day-ahead adjustment capability data of the computing power center cluster based on multi-dimensional coupling constraints; among which, multi-dimensional coupling constraints include service quality constraints, computing power economy constraints, green environment constraints, and safe operation constraints. The correction module is used to obtain the day-ahead correction data of the computing center cluster, and correct the day-ahead adjustment capability data based on the day-ahead correction data to obtain the adjustment capability assessment result of the computing center cluster. The day-ahead correction data is obtained in the following ways: real-time scheduling data for historical periods is obtained based on real-time grid frequency and real-time computing power demand data of the computing center cluster, and multi-dimensional coupling constraints and real-time optimization scheduling objectives; day-ahead scheduling data for historical periods is obtained based on day-ahead computing power demand data of the computing center cluster, and multi-dimensional coupling constraints and day-ahead optimization scheduling objectives; and day-ahead correction data is obtained based on real-time scheduling data, day-ahead scheduling data, and day-ahead regulation capability data for historical periods.

12. The adjustable capability assessment system for computing center clusters according to claim 11, characterized in that, The daily computing power requirement data includes the number of daily tasks, the computing power requirement of each daily task, and the data volume of each daily task; the daily adjustment capability data includes the daily power adjustment range, the daily power adjustment rate range, and the daily power adjustment limit duration. The day-ahead correction data includes one or more of the following: the deviation between the real-time adjustment power and the day-ahead adjustment power, the constraint boundary triggering frequency of the real-time adjustment power, and the real-time changes in the computing power service quality.

13. The adjustable capability assessment system for computing center clusters according to claim 11, characterized in that, The service quality constraints are as follows: in, The total latency of the computing center cluster. This represents the number of tasks completed in the previous day. For the first The priority weight of each task. For the first The transmission latency of each task , For the first The amount of data per task For the first Network bandwidth for each task For bandwidth utilization, For the first The computational processing latency of each task. , For the first The computing power requirements for each task. For the first The allocation of computing power in the computing center for each task. For computing power utilization, This represents the maximum tolerable latency for the computing center cluster. The economic constraints on computing power are as follows: in, For effective computing power energy efficiency ratio, The total computing power output of the computing center cluster. This represents the total energy consumption of the computing center cluster. For power efficiency, The baseline energy efficiency threshold; The green environment constraints are: in, The total carbon emissions of the computing center cluster. The direct carbon emissions of the computing power center cluster, For the indirect carbon emissions of computing center clusters, For the quantity of energy types, For the first Consumption of energy sources, For the first Carbon emission factors of energy-like substances The electricity purchased by the computing center cluster from the power grid. for The carbon emission factor of the power grid at any time For carbon emission quotas; The safe operation constraints are as follows: ， in, For the computing power center cluster Taiwan Computing Center Power at any moment and For the computing power center cluster The upper and lower power limits of a computing center. For the computing power center cluster Taiwan Computing Center Power at any moment For the computing power center cluster The maximum power change rate of the computing center.

14. The adjustable capability assessment system for computing center clusters according to claim 11, characterized in that, The real-time optimization scheduling objective is: in, To optimize the target in real time, For response precision weights, For computing power center clusters Adjust power demand in real time. The number of computing centers. For the computing power center cluster Taiwan Computing Center Power at any moment For stability weights, For the computing power center cluster Taiwan Computing Center Power at any given moment; The current daytime optimization scheduling objective is: in, To optimize the target in advance, As a weighted average of energy costs, For the energy cost of computing center clusters, As carbon cost weight, The total carbon emissions of the computing center cluster. For the cost of computing power loss, , This represents the number of tasks completed in the previous day. For the first The transmission latency of each task For the first The computational processing latency of each task. This is the delay compensation coefficient.

15. The adjustable capability assessment system for computing center clusters according to claim 11, characterized in that, The process of obtaining the day-ahead scheduling data for historical periods based on the day-ahead computing power demand data of the computing center cluster, and grounded in multidimensional coupling constraints and day-ahead optimization scheduling objectives, includes: A day-ahead optimization model is constructed based on multidimensional coupling constraints and day-ahead optimization scheduling objectives. Uncertain variables in the day-ahead optimization model are modeled for uncertainty using a split-Brow bar optimization framework based on moment information, and a fuzzy set consisting of confidence interval constraints of first-order moments and second-order moments is constructed. Based on the task priority conservatism coefficients preset for real-time and non-real-time tasks, the fuzzy set is subjected to multimodal processing based on task type to obtain a multimodal uncertain fuzzy set. The day-ahead optimization model containing multimodal uncertainty fuzzy sets is transformed into a deterministic optimization model using duality theory and linear matrix inequality transformation method. Based on the day-ahead computing power demand data of the computing center cluster in historical periods, the deterministic optimization model is solved to obtain the day-ahead scheduling data of historical periods.

16. The adjustable capability assessment system for computing center clusters according to claim 15, characterized in that, The solution to the deterministic optimization model includes: The particle swarm optimization algorithm is used to solve the deterministic optimization model to obtain the current particle swarm optimal solution, and the beetle whisker algorithm is used to perform local search based on the current particle swarm optimal solution to obtain the current beetle whisker optimal solution. The current optimal solution of the longhorn beetle is used as the initial optimal solution of the particle swarm optimization algorithm, and the above steps are repeated until the preset iteration termination condition is reached. The current optimal solution of the longhorn beetle is then used as the final solution.

17. The adjustable capability assessment system for computing center clusters according to claim 11, characterized in that, The process of obtaining real-time scheduling data for historical periods based on real-time power grid frequency and real-time computing power demand data from the computing center cluster, and grounded in multi-dimensional coupling constraints and real-time optimization scheduling objectives, includes: Based on the real-time power grid frequency and real-time computing power demand data of the computing center cluster during historical periods, the real-time scheduling data of historical periods is obtained by calling the pre-trained real-time scheduling model. The real-time scheduling model is based on a dual-deep Q-network framework and is constructed based on multi-dimensional coupling constraints and real-time optimization scheduling objectives. The state space of the real-time scheduling model is the real-time power grid frequency and real-time computing power demand data, the action space is the power adjustment level of each computing center in the computing center cluster, and the reward function is the negative value of the real-time optimization scheduling objective.

18. The adjustable capability assessment system for computing center clusters according to claim 17, characterized in that, The pre-trained real-time scheduling model uses a random-first experience replay method to replay the experience during training. Among them, during experience replay, experience The priority is , For experience Time difference error, Preset as a positive number; experience The normalized importance weights are , The size of the experience pool. For experience The sampling probability, As a regulating factor, The maximum importance weight for each sampled experience; and the storage and sampling of experiences based on the SumTree structure.

19. The adjustable capability assessment system for computing center clusters according to claim 11, characterized in that, The process of obtaining real-time scheduling data for historical periods based on real-time power grid frequency and real-time computing power demand data from the computing center cluster, and grounded in multi-dimensional coupling constraints and real-time optimization scheduling objectives, includes: Based on the real-time power grid frequency and real-time computing power demand data of the computing center cluster during historical periods, real-time scheduling data of historical periods under the current scenario is obtained based on multi-dimensional coupling constraints and real-time optimization scheduling objectives. The day-ahead adjustment capability data includes day-ahead adjustment capability data under various scenarios; the day-ahead scheduling data for historical periods includes day-ahead scheduling data for historical periods under various scenarios. The step of obtaining day-ahead correction data based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data from historical time periods includes obtaining day-ahead correction data for the current scenario based on real-time scheduling data, day-ahead scheduling data, and day-ahead adjustment capability data from historical time periods in the current scenario.

20. The adjustable capability assessment system for computing center clusters according to claim 11, characterized in that, It also includes an emergency assessment module, which is used for: The system acquires real-time data on the power grid connected to the computing center cluster, and when the scenario is an extreme scenario, it acquires computing power demand data of the computing center cluster before a preset time period, and obtains the adjustment capability data of the computing center cluster before a preset time period based on the computing power demand data before the preset time period and multi-dimensional coupling constraints. Obtain the pre-set adjustment data of the computing center cluster, and adjust the pre-set adjustment capability data based on the pre-set adjustment data to obtain the emergency adjustability assessment result of the computing center cluster.

21. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for assessing the adjustable capability of a computing center cluster as described in any one of claims 1 to 10.

22. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for assessing the adjustability of a computing center cluster as described in any one of claims 1 to 10.

Images