Data center cluster scheduling methods, electronic devices and readable storage media

By incorporating carbon management accounting and operational data into the data center cluster scheduling model and using economic operating costs as the optimization objective, the disconnect between carbon management accounting and carbon emission management is resolved, thereby improving the overall operational efficiency of the data center cluster.

CN123117564APending Publication Date: 2026-11-17湖南工商大学
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611614707.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-19
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing data center cluster scheduling methods, the carbon management accounting mechanism is separated from the carbon emission management mechanism, which makes it impossible to optimize economic operating costs and carbon compliance costs in a coordinated manner, thus affecting overall operational efficiency.

Method used

A scheduling model is constructed, incorporating carbon management accounting data and operational data. The model is then solved using economic operating costs as the optimization objective, thereby achieving synergistic optimization of carbon management accounting data and economic operating costs and avoiding conflicts in scheduling decisions.

Benefits of technology

Accurately predict the economic operating costs and carbon management accounting volume under different scheduling strategies to achieve comprehensive optimization of data center clusters and improve overall operational efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123117564A_ABST
    Figure CN123117564A_ABST
Patent Text Reader

Abstract

This application relates to the field of energy planning technology, and discloses a data center cluster scheduling method, electronic device, and readable storage medium. The method includes: acquiring operational data of the data center cluster; constructing a scheduling model based on the operational data, wherein the decision variables of the scheduling model are the operational scheduling variables of the data center, and the scheduling model is used to describe the accounting relationship between carbon management accounting quantities, operational data, and operational scheduling variables; solving the scheduling model with at least economic operating cost as the optimization objective to obtain a scheduling decision, wherein the economic operating cost is determined by the operational scheduling variables; and configuring the operational scheduling variables of the data center based on the scheduling decision. The embodiments of this application can achieve synergistic optimization of carbon management accounting quantities and economic operating costs, avoid conflicts between economic costs and carbon compliance costs in scheduling decisions, and improve the overall operational efficiency of the data center cluster.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of energy planning technology, and in particular to a data center cluster scheduling method, electronic device, and readable storage medium. Background Technology

[0002] In current data center cluster scheduling practices, there is a significant disconnect between carbon management accounting mechanisms and carbon emission management mechanisms. Carbon management accounting encompasses key indicators such as actual emissions and green electricity carbon credits, but existing scheduling methods fail to dynamically link these indicators with operational data. The scheduling decision-making process typically handles economic operating cost optimization and carbon compliance cost control independently, resulting in a lack of coordinated adjustments between the two. This separation makes it difficult for scheduling models to accurately reflect the intrinsic relationship between carbon management accounting and operational scheduling variables, ultimately hindering the improvement of overall operational efficiency. Summary of the Invention

[0003] The purpose of this application is to provide a data center cluster scheduling method, electronic device, and readable storage medium that can achieve synergistic optimization of carbon management accounting and economic operating costs, avoid conflicts between economic costs and carbon compliance costs in scheduling decisions, and improve the overall operating efficiency of the data center cluster.

[0004] This application provides a data center cluster scheduling method, including: Obtain operational data from the data center cluster; Based on the operational data, a scheduling model is constructed; the decision variables of the scheduling model are the operational scheduling variables of the data center, and the scheduling model is used to describe the accounting relationship between carbon management accounting, the operational data, and the operational scheduling variables. The scheduling model is solved with at least economic operating cost as the optimization objective to obtain the scheduling decision; the economic operating cost is determined by the operation scheduling variables. Based on the scheduling decision, configure the operation scheduling variables of the data center.

[0005] In some embodiments, the operational data includes the node carbon intensity, carbon quota, and green electricity carbon offset coefficient of the data center; the operational scheduling variables include the data center's regular electricity purchases, green electricity usage, and carbon purchases; the carbon management accounting quantities include actual emissions and green electricity carbon offsets; and the accounting relationships include the accounting relationship between actual emissions, regular electricity purchases, and node carbon intensity, the accounting relationship between green electricity carbon offsets, green electricity usage, and green electricity carbon offset coefficient, and the accounting relationship between carbon purchases, actual emissions, green electricity carbon offsets, and carbon quotas.

[0006] In some embodiments, the operational data includes the online task load and offline batch task load of the data center, the operational scheduling variables include the online task processing volume and offline batch task processing volume of the data center, the sum of the online task processing volume within a preset time period equals the online task load, the cumulative amount of the offline batch task processing volume within a preset delay window equals the offline batch task load, and the accounting relationship includes the accounting relationship between the carbon management accounting volume, the online task load, and the offline batch task load.

[0007] In some embodiments, the operational data further includes the task migration load of the data center, and the operational scheduling variable further includes the task migration amount. Within a preset time period, the task migration amount does not exceed the task migration load. The accounting relationship includes the accounting relationship between the carbon management accounting amount, the online task load, the offline batch task load, and the task migration amount.

[0008] In some embodiments, the operational data includes the data center's energy storage capacity, initial power, charging power limit, discharging power limit, charging efficiency, and discharging efficiency. The operational scheduling variables include the data center's energy storage charging amount and energy storage discharging amount. The energy storage charging amount is determined by the energy storage capacity, the initial power, the charging power limit, and the charging efficiency. The energy storage discharging amount is determined by the energy storage capacity, the initial power, the discharging power limit, and the discharging efficiency. The accounting relationship includes the accounting relationship between the carbon management accounting amount, the energy storage charging amount, and the energy storage discharging amount.

[0009] In some embodiments, constructing a scheduling model based on the runtime data includes: Boundary constraints are applied to the running data to obtain constrained running data; A linear relationship is established between the constrained running data and the running schedule variables to obtain the accounting relationship; The scheduling model is obtained by linearly fitting each of the aforementioned accounting relationships.

[0010] In some embodiments, solving the scheduling model with at least economic operating cost as the optimization objective includes one of the following steps: The scheduling model is solved linearly with the goal of minimizing the economic operating cost. The scheduling model is solved linearly with the optimization objective of minimizing the linear fit between the economic operating cost and the carbon management accounting amount. Based on the constraints of the carbon management accounting amount, and with the minimization of the economic operating cost as the optimization objective, the scheduling model is solved linearly.

[0011] In some embodiments, configuring the operational scheduling variables of the data center based on the scheduling decision includes: Convert the operational scheduling variables in the scheduling decision into power trading instructions; The power trading instructions are sent to the data center via the automatic power generation control interface; Based on the actual emissions, green electricity carbon credits, and carbon purchases calculated by the scheduling decision, emission analysis reports under different scheduling modes and / or sensitivity analysis reports under different parameter configurations are generated.

[0012] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-described data center cluster scheduling method.

[0013] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described data center cluster scheduling method.

[0014] The beneficial effects of this application are as follows: By constructing an accurate scheduling model, the accounting relationship between carbon management accounting, operational data and operational scheduling variables is incorporated into the scheduling model, and the solution is performed with at least economic operating cost as the optimization objective. This enables accurate prediction of economic operating cost and carbon management accounting under different scheduling strategies, avoids the conflict between economic cost and carbon compliance cost in scheduling decisions, achieves comprehensive optimization of data center cluster operation, and improves the overall operational efficiency of data center cluster. Attached Figure Description

[0015] Figure 1 This is a diagram illustrating the application environment of the data center cluster scheduling method provided in the embodiments of this application.

[0016] Figure 2 This is a flowchart of the data center cluster scheduling method provided in the embodiments of this application.

[0017] Figure 3 This is a flowchart of the method for constructing a scheduling model provided in the embodiments of this application.

[0018] Figure 4 This is a flowchart of a method for configuring the operation scheduling variables of a data center, as provided in an embodiment of this application.

[0019] Figure 5 This is a flowchart of the secure rolling optimization method provided in the embodiments of this application.

[0020] Figure 6 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0022] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and drawings are used to distinguish similar objects and are not used to describe a specific order or sequence.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit the application. Furthermore, the information, data, and signals involved in the embodiments of this application are all authorized by relevant parties or have been fully authorized by the parties involved.

[0024] In existing data center cluster scheduling methods, there is a lack of effective coordination between carbon management accounting mechanisms and carbon emission management mechanisms, resulting in the inability to jointly optimize economic operating costs and carbon compliance costs. The carbon management accounting process and the scheduling decision-making process are independent of each other, which means that carbon emission constraints are not simultaneously incorporated into the optimization process, causing a disconnect between resource allocation and carbon emission targets, and affecting overall operational efficiency.

[0025] If the aforementioned issues are not addressed, the operation of data center clusters will face ongoing economic burdens and compliance risks. The separation of carbon management accounting from dispatching decisions may lead to irrational allocation of power resources, exacerbate the uncontrollability of energy consumption, and generate unnecessary compliance expenditures in the carbon trading market. Consequently, the sustainable operation capability of data centers will be constrained, making it difficult to meet long-term development goals under carbon emission regulatory requirements.

[0026] Based on this, embodiments of this application provide a data center cluster scheduling method, electronic device, and readable storage medium. By constructing a scheduling model that integrates carbon management accounting, operational data, and operational scheduling variables, and solving the model with at least economic operating cost as the optimization objective, the method achieves synergistic optimization of carbon management accounting and economic operating cost, avoids conflicts between economic cost and carbon compliance cost in scheduling decisions, and improves the overall operational efficiency of the data center cluster.

[0027] Figure 1 This is a diagram illustrating the application environment of the data center cluster scheduling method provided in the embodiments of this application. Figure 1As shown, this method is applied to a data center cluster scheduling system. The system includes a terminal 101 and a server 102. Terminal 101 and server 102 are connected via a network. Terminal 101 can be at least one of a mobile phone, tablet, laptop, or vehicle-mounted terminal. Server 102 can be a standalone server or a server cluster composed of several servers. Terminal 101 is used to send operational data of the data center cluster to server 102, and the intelligent agent interaction request includes user questions. Server 102 is used to obtain the operational data of the data center cluster, construct a scheduling model based on the operational data, solve the scheduling model with at least economic operating cost as the optimization objective, obtain scheduling decisions, and configure the operational scheduling variables of the data center based on the scheduling decisions. The decision variables of the scheduling model are the operational scheduling variables of the data center. The scheduling model describes the accounting relationship between carbon management accounting quantities, operational data, and operational scheduling variables. The economic operating cost is determined by the operational scheduling variables.

[0028] It should be understood that Figure 1 The application scenarios shown are merely examples. In practical applications, the data center cluster scheduling method provided in this application embodiment can also be applied to other scenarios. For example, the above-described data center cluster scheduling method can be directly applied to terminal 101. Terminal 101 is used to obtain the operating data of the data center cluster, construct a scheduling model based on the operating data, solve the scheduling model with at least economic operating cost as the optimization objective, obtain scheduling decisions, and configure the operating scheduling variables of the data center based on the scheduling decisions.

[0029] like Figure 2 As shown, in one embodiment, a data center cluster scheduling method is provided, wherein the execution subject of the method is the aforementioned terminal or server, and the method includes S201 to S204.

[0030] S201, obtain the operational data of the data center cluster.

[0031] Operational data refers to the various operational status and environmental information of a data center cluster within a specific time period, such as power consumption, workload, energy prices, and carbon emission factors.

[0032] Data center cluster operational data can be collected via sensors, obtained from power suppliers through interfaces, or manually entered. In one implementation, historical operational logs can be exported from the data center management system and used as operational data. In another implementation, operators can input the current operational data of the data center.

[0033] S202, based on runtime data, constructs a scheduling model.

[0034] The decision variables of the scheduling model are the operation scheduling variables of the data center. The scheduling model is used to describe the accounting relationship between carbon management accounting quantities, operation data and operation scheduling variables.

[0035] A scheduling model is a mathematical or logical framework used to describe the operating mechanism of a data center cluster and the interactions between various variables.

[0036] The decision variables in a scheduling model refer to the parameters that can be adjusted and controlled within the model, and their values ​​directly affect the operational status and optimization objectives of the data center cluster. Examples include the on / off status of devices, the amount of tasks assigned, and the amount of energy purchased.

[0037] Data center operation scheduling variables refer to parameters that can be adjusted during the operation of a data center, and their changes directly affect the data center's resource utilization, energy consumption, and carbon emissions.

[0038] Carbon management metrics refer to indicators used to measure the carbon emission performance of data center clusters, such as actual carbon emissions, carbon credits, and carbon footprint. These quantitative indicators are key to achieving carbon management goals.

[0039] Accounting relationships refer to the mathematical expressions or logical rules used in scheduling models to describe the quantitative relationships between different variables. For example, the relationship between electricity consumption and carbon emissions.

[0040] Constructing a scheduling model can be done by building a mathematical model where operational data serves as input parameters, scheduling variables are adjustable parameters, and carbon management accounting quantities and economic operating costs are outputs. One implementation approach uses a rule-based model, pre-setting a series of conditions and actions; when the operational data meets specific conditions, the corresponding scheduling action is triggered. Another approach involves constructing a linear programming model, where the relationships between variables are represented by a system of simple linear equations.

[0041] S203, with at least economic operating cost as the optimization objective, solve the scheduling model to obtain the scheduling decision.

[0042] The economic operating cost is determined by the operation scheduling variables.

[0043] Economic operating costs refer to the various economic expenditures incurred during the operation of a data center cluster, including but not limited to electricity procurement costs, equipment maintenance costs, and carbon emission compliance costs.

[0044] Optimization objectives refer to one or more quantitative goals that are desired to be achieved during the scheduling process, such as minimizing economic operating costs, minimizing carbon emissions, and maximizing system throughput.

[0045] Scheduling decisions refer to the specific operating schemes or parameter configurations obtained by solving the scheduling model.

[0046] After constructing the scheduling model, an objective function can be defined, representing the total economic operating cost of the data center under specific scheduling variables. Through iterative calculation or exhaustive search, the combination of scheduling variables that minimizes the objective function can be found, thus obtaining the scheduling decision. One implementation uses a greedy algorithm, selecting the seemingly optimal scheduling variable configuration at each step until a local optimum is reached. Another implementation evaluates all possible combinations of scheduling variables and selects the combination with the lowest economic operating cost as the scheduling decision.

[0047] S204, based on scheduling decisions, configures the data center's operational scheduling variables.

[0048] After receiving the scheduling decision, the data center's operational scheduling variables are configured based on that decision. The scheduling decision might instruct a data center to reduce its regular electricity purchases within a specific time period. In one implementation, administrators can adjust regular electricity purchases, green electricity usage, and carbon purchases individually based on the scheduling decision through the data center's interactive interface. In another implementation, a script can be written to write the parameter values ​​from the scheduling decision into the data center device's configuration file, and then the configuration can be restarted or reloaded.

[0049] Therefore, this application constructs a precise scheduling model, incorporates the accounting relationship between carbon management accounting, operational data, and operational scheduling variables into the scheduling model, and solves the problem with at least economic operating cost as the optimization objective. This enables accurate prediction of economic operating cost and carbon management accounting under different scheduling strategies, avoids conflicts between economic cost and carbon compliance cost in scheduling decisions, achieves comprehensive optimization of data center cluster operation, and improves the overall operational efficiency of data center clusters.

[0050] In some embodiments, operational data includes the node carbon intensity, carbon quota, and green electricity carbon offset coefficient of the data center; operational scheduling variables include the data center's regular electricity purchases, green electricity usage, and carbon purchases; carbon management accounting quantities include actual emissions and green electricity carbon offsets; and accounting relationships include the accounting relationship between actual emissions, regular electricity purchases, and node carbon intensity, the accounting relationship between green electricity carbon offsets, green electricity usage, and green electricity carbon offset coefficient, as well as the accounting relationship between carbon purchases, actual emissions, green electricity carbon offsets, and carbon quotas.

[0051] Node carbon intensity refers to the amount of carbon emissions generated by a data center consuming a unit of electricity per unit time. It can be set based on the average carbon emission intensity of the power grid in the area where the data center is located, or calculated as a weighted average based on the actual power supply structure connected to the data center and its respective carbon emission factors.

[0052] Carbon allowances refer to the upper limit of carbon emissions that data centers are allowed to emit within a specific period, allocated by relevant authorities. These allowances can be obtained based on carbon emission documents issued by the relevant authorities or by participating in the carbon emission trading market.

[0053] The carbon offset factor for green electricity refers to the amount of carbon emissions that can be offset by using a unit of green electricity. It can be set according to the carbon offset factor of the type of green electricity (such as wind power and photovoltaic) specified in national or industry standards, or calculated based on the carbon emission data of green electricity provided by green electricity suppliers and certified by a third party.

[0054] Ordinary electricity purchase refers to the amount of non-green electricity that a data center purchases from the traditional power grid. This amount can be determined through long-term power purchase contracts with power grid companies or through dynamic forecasting and adjustment based on the real-time load demand of the data center and market electricity prices.

[0055] Green electricity usage refers to the amount of green electricity that data centers directly or indirectly purchase and use. This can be purchased through green electricity trading platforms or obtained through renewable energy power generation facilities such as photovoltaic and wind power built by the data centers themselves.

[0056] Carbon purchases refer to the amount of carbon emission rights that data centers purchase from the carbon trading market to make up for emissions exceeding their carbon quotas. They can purchase these rights by participating in the carbon emission rights trading market or by cooperating with professional carbon asset management agencies.

[0057] Actual emissions refer to the total amount of greenhouse gases generated by data centers during a specific period due to activities such as electricity consumption. They can be calculated based on ordinary electricity purchases and nodal carbon intensity, or by real-time monitoring and data collection through the installation of carbon emission monitoring equipment in data centers.

[0058] Green electricity carbon credit refers to the amount of carbon emissions reduced by data centers due to the use of green electricity. It can be calculated based on the amount of green electricity used and the green electricity carbon credit coefficient, or based on the credit amount stated on the green electricity trading voucher or green electricity certificate.

[0059] After acquiring operational data from the data center cluster, the scheduling model can accurately construct accounting relationships between actual emissions and carbon credits, based on specific node carbon intensity, carbon quotas, green electricity carbon offset coefficients, and operational scheduling variables such as regular electricity purchases, green electricity usage, and carbon purchases. These accounting relationships enable the scheduling model to accurately quantify the data center's carbon footprint and carbon reduction contribution. Therefore, when solving the scheduling model with economic operating costs as the optimization objective, it can incorporate carbon emission costs (such as carbon purchase costs) and green electricity benefits (such as carbon offset benefits), achieving synergistic optimization of economic and environmental benefits. Ultimately, based on the obtained scheduling decisions, the data center's operational scheduling variables (such as regular electricity purchases, green electricity usage, and carbon purchases) are precisely configured, ensuring that the data center effectively manages carbon emissions while meeting operational needs.

[0060] Actual emissions, green electricity carbon credits, and carbon purchases can be calculated according to formula (1): , in, For regular electricity purchases, For green electricity usage, For carbon purchases, For node carbon strength, The carbon credit factor for green electricity. For carbon quotas, Indicates data center, Indicates discrete scheduling time periods. This represents the actual emissions. This is the amount of carbon credit for green electricity.

[0061] The following is a specific example to illustrate this. Assume that during a certain scheduling cycle, the node carbon intensity of a data center in its region is 0.5 tons of CO2 / MWh, the carbon allowance obtained is 1000 tons of CO2, and the green electricity carbon offset coefficient is 0.8 tons of CO2 / MWh. The scheduling model will use ordinary electricity purchase, green electricity usage, and carbon purchase as the operating scheduling variables. At this time, by substituting the node carbon intensity, green electricity carbon offset coefficient, and carbon allowance into equation (1), the actual emissions, green electricity carbon offset, and required carbon purchase for this scheduling cycle can be obtained. Taking into account the costs of ordinary electricity purchase, green electricity purchase, and carbon purchase, the total economic operating cost will be minimized by adjusting the ordinary electricity purchase, green electricity usage, and carbon purchase, while meeting the power demand of the data center. For example, when the price of green electricity is relatively low, the scheduling decision may increase the usage of green electricity to reduce the actual emissions and carbon purchase; conversely, when the carbon allowance is sufficient and the carbon price is high, ordinary electricity purchase may be given priority to reduce short-term costs.

[0062] In some embodiments, the running data includes the online task load and offline batch task load of the data center, the running scheduling variables include the online task processing volume and offline batch task processing volume of the data center, the sum of the online task processing volume within a preset time period equals the online task load, the cumulative amount of offline batch task processing volume within a preset delay window equals the offline batch task load, and the accounting relationship includes the accounting relationship between carbon management accounting volume, online task load and offline batch task load.

[0063] Online workload refers to the amount of tasks that a data center cluster needs to process within a specific time period, which have high requirements for response time, such as user interaction requests and real-time data queries.

[0064] Offline batch task workload refers to the amount of tasks that can be completed within a certain time window with relatively relaxed response time requirements, such as data analysis report generation, model training, and log processing.

[0065] Online task processing volume refers to the number of online tasks or computing resources actually completed by a data center cluster within a specific time period.

[0066] Offline batch task processing volume refers to the number of offline batch tasks or computing resource consumption actually completed by the data center cluster within a specific time period.

[0067] When constructing the scheduling model, the scheduling variables are specified as online task processing volume and offline batch task processing volume, enabling scheduling decisions to directly affect different types of task processing. To ensure service quality and task completion, this application introduces key constraints: within a preset time period, the sum of online task processing volumes is set to equal the online task load, ensuring timely response for online tasks with high real-time requirements and avoiding service interruptions or delays. Simultaneously, within a preset delay window, the cumulative amount of offline batch task processing is set to equal the offline batch task load, providing valuable flexibility for offline task scheduling. This allows, while meeting the total processing volume, offline tasks to be scheduled for periods with lower economic operating costs or lower carbon management accounting volume based on external environmental factors (such as electricity price fluctuations or changes in carbon emission intensity). Furthermore, by establishing the accounting relationship between carbon management accounting volume and online and offline batch task loads, the scheduling model can simultaneously consider the contribution of different task types to carbon emissions when optimizing economic operating costs. This integration enables the scheduling model to balance task performance, economic benefits, and environmental benefits within a unified framework during the solution process, thereby obtaining a scheduling decision that can both meet business needs and effectively reduce operating costs and carbon emissions.

[0068] To incorporate the real-time requirements of online tasks and the latency window requirements of offline batch tasks into the scheduling model, the service constraints of equation (2) can be adopted: , in, For data center collection, Let i be the number of online tasks processed by data center i during time period t. Let τ represent the offline batch task processing volume of data center i within the time period τ. This represents the total online task load that the data center cluster needs to handle during time period t. This represents the total offline batch task load of the data center cluster arriving in time period t. The number of scheduling periods included in the preset delay window for offline batch tasks. It is the set of discrete scheduling periods within a scheduling cycle. This represents the set of arrival time periods for offline batch tasks. τ is the index of the execution time period within the offline batch task delay window; the superscripts on and off identify online and offline batch tasks, respectively.

[0069] The following example illustrates this. Suppose a data center cluster needs to schedule resources within a 24-hour period, divided into 24 one-hour time slots. During a specific time slot, such as 10:00 AM, the cluster's operational data might show an online task load of 1000 requests per hour, while the total offline batch task load for the day is 10,000, and these offline batch tasks are allowed to complete within an 8-hour delay window. The scheduling model uses both online and offline batch task volumes as decision variables. Based on the constraint that the sum of online task volumes within a preset time slot equals the online task load, the model ensures that the online task volume at 10:00 AM is set at 1000 requests. Simultaneously, based on the constraint that the cumulative offline batch task volume within a preset delay window equals the offline batch task load, the model flexibly allocates offline batch task volume within this 8-hour window based on factors such as electricity prices and carbon emission factors. For example, if electricity prices and carbon emission factors are lower at night (e.g., 1 AM to 5 AM), the scheduling model may concentrate most of the offline batch task processing volume during these periods to reduce overall economic operating costs and carbon management accounting workload. During the daytime, when electricity prices and carbon emission factors are higher, the focus is primarily on ensuring the processing of online tasks, with offline batch tasks allocated appropriately based on remaining resources and optimization objectives. In this way, scheduling decisions can precisely configure the online and offline batch task processing volumes for each time period, thereby meeting the real-time requirements of online tasks while optimizing the execution timing of offline tasks, achieving a dual optimization of economic and environmental benefits.

[0070] In some embodiments, the runtime data also includes the data center's task migration load, and the runtime scheduling variables also include the task migration amount. Within a preset time period, the task migration amount does not exceed the task migration load. The accounting relationship includes the accounting relationship between carbon management accounting amount, online task load, offline batch task load, and task migration amount.

[0071] Task migration load refers to the total amount of task migrations that a data center can handle or is allowed to make within a specific time period. This can be an upper limit, representing the amount of task data or task instances that a data center can receive or send without affecting its normal operation or exceeding resource limitations such as network bandwidth and storage I / O. For example, task migration load can be dynamically calculated and set based on physical resource limitations such as network bandwidth between data centers, server computing power, and storage system I / O performance; alternatively, the task migration load can also be a fixed value preset based on management policies, such as setting a daily or hourly task migration limit to avoid system instability caused by excessive migration.

[0072] Task migration volume refers to the number of tasks or task load that are actually migrated from one data center to another or from other data centers to the current data center within a scheduling cycle. It is a decision variable in the scheduling model, representing the specific operation the scheduler decides to perform for task migration. For example, task migration volume can be expressed as the number of migrated task instances, such as migrating N virtual machines or containers from data center A to data center B; alternatively, task migration volume can also be expressed as the computing resources (such as the number of CPU cores, memory size) or data volume occupied by the migrated tasks.

[0073] When solving the scheduling model, in addition to considering the constraints that online task processing volume and offline batch task processing volume must meet their respective load and latency window requirements, a constraint is added that the task migration volume within a preset time period should not exceed the task migration load. This mechanism enables the scheduling model to make more comprehensive evaluations and decisions, optimizing not only local task processing but also balancing the load across the entire data center cluster through task migration and leveraging the resource advantages of different data centers (such as lower electricity prices or carbon intensity). This achieves overall optimization of economic operating costs and carbon management accounting while meeting task requirements. By incorporating task migration capabilities into scheduling considerations, the scheduling model can more flexibly balance the load across different data centers, effectively utilizing the overall resources of the cluster and avoiding local overload or resource idleness.

[0074] For task migration between different data centers, equation (3) can be used to limit the migration volume of each migration link in each time period: , in, The migration amount for tasks moving from data center i to data center j. This refers to the task migration load for this link during the corresponding scheduling period.

[0075] The following example illustrates this. Assume a data center cluster comprises two data centers, DC1 and DC2. DC1 currently faces high online and offline batch task loads, and its region has high electricity prices or carbon intensity. DC2, on the other hand, has relatively low loads, and its region has low electricity prices or carbon intensity. Furthermore, there is sufficient network bandwidth between DC2 and DC1, allowing for task migration. The operational data will then include the online and offline batch task loads of DC1 and DC2 respectively, as well as the task migration load between DC1 and DC2. For example, the task migration load from DC1 to DC2 might be 500 virtual machine instances per hour. The scheduling model will use the online and offline batch task processing volumes of DC1 and DC2, and the task migration volume from DC1 to DC2, as decision variables. When constructing the accounting relationships, the impact of task migration volume on carbon management accounting volume will be considered. For example, migrating a virtual machine instance may incur some network transmission energy consumption, but if migrating it to DC2 significantly reduces its carbon emissions during operation, overall carbon reduction may be achieved. The scheduling model aims to minimize economic operating costs during the solution process, constrained by a limit of no more than 500 virtual machine instances per hour for task migration from DC1 to DC2. Through the solution, the scheduling model might decide to migrate, for example, 100 virtual machine instances from DC1 to DC2 to reduce overall economic operating costs and carbon emissions. Ultimately, based on this scheduling decision, the online task processing volume, offline batch task processing volume, and task migration volume of DC1 and DC2 will be configured for execution.

[0076] In some embodiments, the operational data includes the data center's energy storage capacity, initial power, charging power limit, discharging power limit, charging efficiency, and discharging efficiency. The operational scheduling variables include the data center's energy storage charging amount and energy storage discharging amount. The energy storage charging amount is determined by the energy storage capacity, initial power, charging power limit, and charging efficiency. The energy storage discharging amount is determined by the energy storage capacity, initial power, discharging power limit, and discharging efficiency. The accounting relationships include the accounting relationships between carbon management accounting amounts, energy storage charging amounts, and energy storage discharging amounts.

[0077] Energy storage capacity refers to the maximum electrical energy that the energy storage equipment equipped in a data center can store. It is a key parameter for measuring the scale of an energy storage system. It can be a fixed value or an effective capacity that is dynamically updated according to the aging of the energy storage equipment.

[0078] Initial energy refers to the actual electrical energy stored in the energy storage device at the beginning of the scheduling cycle. This parameter provides the scheduling model with the initial state information of the energy storage system and affects subsequent charging and discharging decisions.

[0079] The charging power limit refers to the maximum electrical power that an energy storage device can absorb per unit time, which limits the charging speed of the energy storage system. This limit can be determined by the rated power of the energy storage device or can be dynamically adjusted according to the grid load or the health status of the device.

[0080] The discharge power limit refers to the maximum electrical power that an energy storage device can output per unit time, which limits the discharge rate of the energy storage system. This limit can also be determined by the rated power of the energy storage device, or dynamically adjusted according to system requirements and equipment status.

[0081] Charging efficiency refers to the ratio of input electrical energy to actual stored electrical energy during the charging process of an energy storage device. It reflects the energy loss during the charging process. For example, DC-DC conversion efficiency and AC-DC conversion efficiency can be used.

[0082] Discharge efficiency refers to the ratio of the actual output electrical energy to the stored electrical energy during the discharge process of an energy storage device. It reflects the energy loss during the discharge process. For example, DC-AC conversion efficiency and DC-DC conversion efficiency can be used.

[0083] Energy storage charging capacity refers to the electrical energy that data center energy storage devices obtain and store from the grid or other power sources during the scheduling cycle. This variable is an important decision variable in the scheduling model and directly affects the state of charge of energy storage devices and grid interaction.

[0084] Energy storage discharge refers to the electrical energy released by data center energy storage devices to the data center load or the power grid during the scheduling cycle. This variable is also an important decision variable in the scheduling model, used to meet the power demand of the data center or participate in grid ancillary services. The determination of energy storage charging capacity is limited by the physical characteristics and operational constraints of the energy storage devices. During the discrete scheduling period, the energy storage state of charge, charge / discharge capacity, and output power can be dynamically constrained according to equation (4): , in, For the energy storage state of charge, in equation (4), it is represented by the electrical energy stored in the data center i energy storage device at the beginning of time period t. For energy storage capacity; and These represent the amount of charge and the amount of discharge released from the energy storage side, respectively. and These represent charging efficiency and discharging efficiency, respectively. This indicates the effective output power for data center loads. This refers to the maximum permissible charging power measured on the charging input side of the data center i-energy storage device. This refers to the maximum permissible discharge power measured on the energy storage side of the data center i-energy storage device. This refers to the duration of a single discrete scheduling period. All electrical energy values ​​are expressed in kWh or MWh, and power values ​​are expressed in kW or MW accordingly. Power and The product of these is the upper limit of electrical energy for the corresponding time period.

[0085] When acquiring operational data from data center clusters, not only traditional load information is included, but also key operational parameters of the energy storage system, such as energy storage capacity, initial charge, charging power limits, discharging power limits, charging efficiency, and discharging efficiency, are further incorporated. These parameters provide the physical boundaries and operational characteristics of the energy storage system for the scheduling model. In constructing the scheduling model, in addition to traditional operational scheduling variables, energy storage charging and discharging quantities are added as decision variables. The determination of these energy storage-related operational scheduling variables strictly adheres to the physical constraints and efficiency characteristics defined by energy storage capacity, initial charge, charging power limits, discharging power limits, charging efficiency, and discharging efficiency. For example, energy storage charging and discharging quantities cannot exceed their respective power limits at any time, and their cumulative amount cannot cause the energy storage state of charge to exceed the energy storage capacity. Simultaneously, efficiency losses during charging and discharging are accurately included in the calculation, ensuring the model accurately reflects the actual physical processes. Based on this, the accounting relationship is further expanded, establishing a quantitative correlation between carbon management accounting quantities and energy storage charging and discharging quantities. This means that the charging and discharging behavior of energy storage systems is not merely a transfer of electrical energy, but is imbued with the attributes of carbon emissions or carbon reduction. For example, charging when the grid carbon intensity is low, or discharging instead of purchasing electricity when the grid carbon intensity is high, will directly impact the carbon management accounting of data centers. In this way, the scheduling model can comprehensively consider economic operating costs and carbon management accounting. When solving the scheduling model, it not only seeks the lowest economic operating cost but also effectively manages the carbon emissions of data centers by optimizing the charging and discharging strategies of energy storage. Ultimately, based on the obtained scheduling decisions, data centers can accurately configure their operational scheduling variables, including energy storage charging and discharging, thereby achieving a dual optimization of economic and carbon management benefits while meeting the operational needs of the data center. This method of deeply integrating energy storage systems into the scheduling model enables data centers to respond more flexibly to fluctuations in electricity and carbon prices, effectively shaving peak loads and reducing overall operating costs and carbon footprint.

[0086] The following is a concrete example to illustrate this. Assume a data center is equipped with a lithium-ion battery energy storage system with a storage capacity of 1000 kWh and an initial charge of 300 kWh at the start of a scheduling cycle. The charging power limit of this energy storage system is 500 kW, the discharging power limit is 400 kW, the charging efficiency is 90%, and the discharging efficiency is 95%. These parameters are input as operational data when building the scheduling model. The scheduling model optimizes the energy storage charging and discharging amounts as operational scheduling variables. For example, in a certain hour, if the electricity price is low and the grid carbon intensity is low, the scheduling model might decide to charge the energy storage system. The charging amount could be set to 450 kWh, but it must be ensured that the total charge after charging does not exceed 1000 kWh, and the charging power does not exceed 500 kW. With a 90% charging efficiency, the actual stored charge is 405 kWh. Conversely, if electricity prices are high or the grid carbon intensity is high, the scheduling model may decide to discharge the energy storage system. The energy storage discharge amount can be set at 380 kWh, but it must be ensured that the remaining energy after discharge is not lower than a safe threshold, and the discharge power does not exceed 400 kW. Based on a 95% discharge efficiency, the effective output energy to the data center load is 361 kWh. These energy storage charging and discharging amounts will directly affect the data center's electricity purchase and consumption structure, and thus influence carbon management accounting through accounting relationships. For example, electricity obtained from the grid during charging increases carbon emissions, while electricity purchased during discharging reduces carbon emissions. In this way, the scheduling model can dynamically adjust the energy storage charging and discharging strategy based on real-time electricity price and carbon intensity information to minimize economic operating costs and carbon management accounting.

[0087] like Figure 3 As shown, in one embodiment, the method for constructing a scheduling model includes S301 to S303.

[0088] S301, apply boundary constraints to the running data to obtain constrained running data.

[0089] S302, construct the linear relationship between the constrained running data and the running schedule variables to obtain the accounting relationship.

[0090] S303, perform linear fitting on each accounting relationship to obtain the scheduling model.

[0091] Boundary constraints on operational data can be implemented by setting reasonable upper and lower limits for various operational data points (such as server power consumption, energy storage capacity, and task load) based on the data center's technical specifications or historical operating experience. Data points exceeding these boundaries can be corrected to the boundary values, or identified as outliers and removed or smoothed to avoid negatively impacting model building. Another approach is to utilize statistical methods, such as quantile analysis based on historical data or outlier detection algorithms, to automatically identify and constrain data ranges, ensuring data reliability.

[0092] Constructing a linear relationship between constrained operational data and operational scheduling variables can be achieved by fitting the ratio between ordinary electricity purchases and actual carbon emissions using the least squares method, or by establishing a linear dependency between online task processing volume and server energy consumption. Alternatively, linear equations or inequalities between operational data and scheduling variables can be directly derived based on known physical laws or engineering principles; for example, the relationship between the charging and discharging power of an energy storage system and changes in electricity volume.

[0093] In some embodiments, equation (5) can be used to first apply boundary constraints to any running data, and then the constrained data can be used to participate in the construction of linear relationships: , in, This is the original running data. The running data after boundary constraints, and These are its lower and upper limits, respectively. This is a set of indexes for the running data.

[0094] Linear fitting of the various accounting relationships can be used as constraints on the model, combined with a pre-defined optimization objective (such as minimizing economic operating costs), to form a linear programming (LP) or mixed-integer linear programming (MILP) problem. Linear fitting ensures that the entire scheduling model maintains a linear structure, which is crucial for solving the problem using efficient optimization algorithms (such as the simplex method or interior-point method), significantly improving solution speed and stability.

[0095] In some embodiments, the relationship between task processing volume, task migration volume, energy storage charging and discharging volume, and purchased electricity volume can be expressed as Equation (6) to maintain the linear structure of the model: , in, This represents the total power consumption of data center i on the load side during time period t, including power consumption for online task execution, offline batch task execution, task migration, and basic operations. The power consumption coefficient corresponding to the number of online tasks per unit processed by data center i. The power consumption coefficient corresponding to the offline batch task volume per unit of data center i. The power consumption coefficient corresponding to the unit task volume of data center i migration. Let t be the amount of tasks that need to be migrated from data center i to data center j within time period t. This represents the total number of tasks migrated from data center i during this period, excluding its own migration items when summing. This refers to the basic power consumption of data center i during a single scheduling period, independent of the aforementioned task processing and migration volumes; it is the constant term in the linear power consumption model. The aforementioned power consumption coefficients and basic power consumption are determined based on historical operating data fitting or equipment parameters. The two rows of equation (6)... They have the same meaning, both referring to electrical energy, not power.

[0096] The proposed solution effectively improves data quality and reliability by imposing boundary constraints on operational data, preventing interference from anomalous data in model construction. Furthermore, by explicitly establishing linear accounting relationships between operational data and scheduling variables, and then linearly fitting these relationships, the mathematical rigor and computational efficiency of the scheduling model are ensured. This systematic model construction method allows for the accurate and concise expression of complex carbon management accounting quantities, operational data, and scheduling variables, thus laying a solid foundation for subsequent optimization solutions.

[0097] The following is a specific example to illustrate this. When constructing a data center cluster scheduling model, firstly, boundary constraints are applied to the acquired operational data, such as the real-time power consumption, electricity price, and carbon intensity of the data center. For example, the server power consumption data is limited to its rated minimum power consumption (e.g., 100W) and maximum power consumption (e.g., 500W), and any readings exceeding this range will be corrected to the corresponding boundary values. Secondly, based on these constrained data, a linear relationship is constructed between ordinary electricity purchase and actual emissions, and between green electricity usage and green electricity carbon offset, according to equation (1). Finally, these linear relationships and other related linear constraints (e.g., task load balancing, energy storage charging and discharging limits, etc.) are integrated into a linear programming model, and the objective function is defined as minimizing economic operating costs, thus obtaining a complete scheduling model.

[0098] In some embodiments, at least with economic operating cost as the optimization objective, solving the scheduling model includes one of the following steps: solving the scheduling model linearly with minimizing economic operating cost as the optimization objective; solving the scheduling model linearly with minimizing the linear fit result of both economic operating cost and carbon management accounting amount as the optimization objective; and solving the scheduling model linearly with minimizing economic operating cost as the optimization objective, based on the constraint of carbon management accounting amount.

[0099] When the optimization objective is to minimize economic operating costs, the core lies in finding the scheduling scheme that minimizes the total operating cost of the data center while satisfying all operational constraints. This typically encompasses various economic factors such as electricity purchase costs, carbon purchase costs, and energy storage charging and discharging costs. Solving the scheduling model linearly means that when the scheduling model is constructed as a linear programming problem, mature linear programming algorithms can be used to solve it, such as the simplex method or interior-point method. These methods can efficiently find the global optimum.

[0100] In some embodiments, when the goal is to minimize economic operating costs, equation (7) can be used to define the cost summary for each scheduling period and each data center: , in, , and These represent the regular electricity purchase price, the green electricity price, and the carbon purchase price per unit, respectively. and These represent the unit operation and maintenance costs for energy storage charging and discharging, respectively. Costs of economic operation.

[0101] When the optimization objective is to minimize the linear fit between economic operating costs and carbon management accounting quantities, this is a multi-objective optimization strategy. By weighting and summing or linearly combining the two objective functions—economic operating costs and carbon management accounting quantities—it is transformed into a single optimization objective. This method allows schedulers to adjust the weights of different objectives based on actual needs and priorities to achieve a balance between economic benefits and carbon management benefits. For example, a comprehensive objective function can be defined, which includes economic operating costs and carbon management accounting quantities, and the weight coefficients of each can be adjusted to reflect preferences for economic efficiency and carbon emissions. Solving this model linearly can also employ the linear programming algorithm described above.

[0102] In some embodiments, to eliminate the impact of different dimensions of cost and emissions, the values ​​can be normalized using a reference value first, and then a single linear weighted optimization objective can be formed using equation (8): , in, and Positive reference costs and reference net emissions. Cost preference weight, This refers to net emissions after deducting carbon credits for green electricity. The dimensionless comprehensive objective function value obtained by weighted summation after normalizing economic operating costs and net emissions is used to evaluate the economic efficiency and carbon emission performance of the scheduling scheme. The smaller the value, the better the scheme under the set weights.

[0103] In one specific embodiment, the optimization objective is to minimize the linear fit between economic operating costs and carbon management accounting quantities. Alternatively, the optimization objective can be to minimize the linear fit between economic operating costs and actual emissions in carbon management accounting quantities.

[0104] Based on constrained carbon management accounting, minimizing economic operating costs as the optimization objective means setting a hard constraint on carbon management accounting, such as setting a carbon emission ceiling or a green electricity usage floor. Under this premise, the solution is then performed with the goal of minimizing economic operating costs. This method ensures that economic operating costs are reduced as much as possible while meeting the preset carbon emission requirements. For example, the actual emissions can be set to not exceed a certain threshold, or the green electricity carbon credit can be set to not be lower than a certain threshold.

[0105] In some embodiments, when carbon management requirements are a hard boundary, an upper limit for net emissions can be set using equation (9), or a lower limit for green electricity carbon credits can be set: , in, This is the preset upper limit for net emissions. This is the preset lower limit for green electricity carbon credit.

[0106] In one specific embodiment, the optimization objective of minimizing economic operating costs, based on the constrained carbon management accounting amount, can be the actual emissions in the constrained carbon management accounting amount.

[0107] This application's solution effectively addresses the potential for insufficient carbon management due to a single economic objective by providing multiple flexible optimization objectives and solution strategies. Based on a constructed scheduling model, the solution allows schedulers to select different optimization paths according to actual operational needs and carbon management policy guidance. When the primary objective is purely economic benefit, minimizing economic operating costs can be directly used as the optimization objective, and the cost-optimal scheduling decision can be quickly obtained through linear solution. This approach is suitable for scenarios with extremely high economic requirements and relatively lenient carbon emission constraints. To achieve a balance between economic benefit and carbon management, the solution provides a strategy that linearly fits economic operating costs and carbon management accounting quantities, using their minimization as the optimization objective. This method, by introducing weighting coefficients, can actively optimize carbon management performance while reducing costs based on preset priorities. Furthermore, for scenarios with strict carbon emission limits or quota requirements, the solution allows for solving the problem with minimizing economic operating costs as the optimization objective, while constraining carbon management accounting quantities. This means first ensuring that carbon emissions meet the preset upper limit or green electricity usage reaches the lower limit, and then seeking the lowest economic operating cost under this premise. Through the different optimization objectives and solution methods described above, the solution proposed in this application enables data center cluster scheduling to move beyond single-dimensional optimization and flexibly adapt to changing market environments and policy requirements, finding the optimal balance between economic efficiency and environmental friendliness, thereby improving the overall effectiveness and adaptability of scheduling decisions.

[0108] The following is a concrete example to illustrate this. Suppose a data center cluster needs to schedule resources within a scheduling cycle based on its operational data (such as node carbon intensity, carbon quotas, green electricity carbon offset coefficients, online task load, offline batch task load, energy storage capacity, etc.) and operational scheduling variables (such as regular electricity purchases, green electricity usage, carbon purchases, online task processing volume, offline batch task processing volume, energy storage charging and discharging volume, etc.). One specific implementation example is as follows: When the data center cluster's operational strategy focuses on pure economic benefits, the optimization objective can be set to minimize economic operating costs. In this case, the scheduling model will be constructed as a linear programming problem, where the objective function only includes various economic costs (such as electricity purchase costs, carbon purchase costs, etc.). This linear programming problem can be quickly solved using commercial optimization solvers such as Gurobi or CPLEX, utilizing their built-in simplex method or interior-point method algorithms, to obtain the scheduling decision with the lowest economic operating cost while satisfying all operational constraints. For example, when data center clusters face carbon emission quota restrictions but also want to balance economic efficiency, a strategy can be adopted that minimizes the linear fit between economic operating costs and carbon management accounting costs as the optimization objective. For instance, a comprehensive objective function can be set, with cost weighting at 0.7 and net emissions weighting at 0.3. This comprehensive objective function is then used as the optimization objective, and a linear solver is used to solve it, resulting in a scheduling decision that balances economic costs and carbon emissions. As another example, when data center clusters must strictly adhere to carbon emission caps, a strategy can be adopted that minimizes economic operating costs while constraining carbon management accounting costs. Specifically, a hard constraint can be added, such as requiring "actual emissions not to exceed 1000 tons of CO2 equivalent" or "green electricity usage not less than 50% of total electricity consumption." Under this carbon emission constraint, a linear solution is then performed with the goal of minimizing economic operating costs. In this way, regardless of the final economic cost, it can be ensured that the data center's carbon emissions will not exceed the preset cap, while minimizing operating costs as much as possible.

[0109] Correspondingly, under the above weight settings, the comprehensive objective function can be expressed as equation (10).

[0110] , In equation (10) With equation (8) For the same comprehensive objective function; Equation (10) is Equation (8) with the same cost preference weight. When the net emission weight is 0.7 and the net emission weight is 0.3, the meanings of all variables and reference values ​​remain consistent.

[0111] like Figure 4As shown, in one embodiment, the method for configuring the operation scheduling variables of the data center includes S401 to S403.

[0112] S401 converts the operation scheduling variables in the scheduling decision into power trading instructions.

[0113] The S402 sends power trading instructions to the data center via the automatic generation control interface.

[0114] S403 generates emission analysis reports under different scheduling modes and / or sensitivity analysis reports under different parameter configurations based on the actual emissions, green electricity carbon credits, and carbon purchases calculated by scheduling decisions.

[0115] This application's solution transforms abstract scheduling decisions output by the scheduling model into executable power trading instructions and utilizes an automatic generation control interface to automate the issuance of these instructions, thereby ensuring that scheduling strategies can be implemented efficiently and accurately at the data center level. This transformation and issuance mechanism enables the data center to dynamically adjust its power procurement, green electricity usage, and energy storage charging and discharging variables based on real-time scheduling optimization results, thus optimizing economic operating costs and finely controlling carbon management accounting in actual operation. Furthermore, by conducting in-depth analysis of the actual emissions, green electricity carbon credits, and carbon purchases calculated from the scheduling decisions, and generating emission analysis and sensitivity analysis reports, this solution not only provides a quantitative assessment of scheduling effectiveness but also reveals the impact of different scheduling modes and parameter configurations on carbon management and economics. This allows operators to continuously optimize scheduling strategies based on detailed data insights. Overall, this solution closely integrates theoretical scheduling decisions with actual operational control and effect evaluation, forming a closed-loop intelligent scheduling and management system that effectively solves the problems of difficult-to-implement scheduling decisions and difficult-to-evaluate their effects.

[0116] The following is a concrete example. Upon receiving a dispatch decision, the system first converts the values ​​regarding regular electricity purchases, green electricity usage, and energy storage charging / discharging into electricity trading instructions conforming to electricity market trading standards, according to a preset protocol format. For example, if the dispatch decision instructs the purchase of 100MWh of regular electricity and 50MWh of green electricity in a future period, this can be converted into two independent electricity trading instructions, each containing information such as the type of electricity purchased, quantity, time period, and price ceiling. Simultaneously, if the dispatch decision involves energy storage systems, such as instructing them to charge or discharge during a specific period, corresponding energy storage control instructions are generated, including charging power, discharging power, and duration. These electricity trading instructions and energy storage control instructions are then encapsulated using the TCP / IP protocol through the automatic generation control interface within the data center and transmitted to the data center's energy management system via a fiber optic network. Upon receiving the instructions, the energy management system further parses and drives the corresponding power equipment to execute commands, such as submitting electricity purchase requests to the grid company or controlling the energy storage inverter to perform charging and discharging operations. After the scheduling cycle ends, based on actual operational data and scheduling decisions, the actual emissions, green electricity carbon credits, and carbon purchases are calculated. Based on this data, an emissions analysis report can be automatically generated, which may include comparative charts of key indicators such as total carbon emissions, green electricity utilization, and carbon trading costs under different scheduling modes (e.g., compared to a scheduling mode that only considers economics). Simultaneously, sensitivity analysis can be performed. For example, by simulating scenarios of a 20% increase in carbon prices or a 10% fluctuation in electricity prices, the scheduling model can be rerun and a sensitivity analysis report generated, demonstrating the potential impact of these parameter changes on the economic operating costs of data centers and carbon management accounting.

[0117] In some embodiments, see Figure 5 To adapt to the time-varying nature of task load, electricity price, node carbon intensity, and green energy availability, as well as prediction errors, a prediction-driven, safe rolling optimization method can be used to generate scheduling decisions. This safe rolling optimization method executes only the safe-verified actions for the current scheduling period in each scheduling period, and then re-predicts and solves using the latest observations in the next scheduling period.

[0118] S501 generates a rolling prediction sequence for the next H scheduling periods based on the current status and historical operation data.

[0119] The current status includes at least the backlog of online and offline batch tasks, available server capacity, migration link capacity, energy storage charge status, electricity purchase status, and carbon quota status.

[0120] The rolling forecast sequence should include at least the task load, electricity price, node carbon intensity, and green electricity availability. The rolling forecast sequence can be generated using physical information digital twins, time-series graph neural networks, Transformers, or other sequence prediction networks, but is not limited to these methods.

[0121] S502, construct an uncertainty set based on the prediction residuals between predicted values ​​and corresponding actual observed values ​​in historical periods.

[0122] The uncertainty set is used to characterize the range of deviations that a rolling forecast sequence may produce.

[0123] S503, within the prediction window, uses the rolling prediction sequence and uncertainty set as input to perform robust rolling optimization or shrinking time-domain model predictive control solution on the scheduling model, generating capacity quotas and task scheduling references for each data center and each time period.

[0124] S504 encodes the current state, rolling prediction sequence, uncertainty set, and capacity quota and task scheduling reference for the current time period into a time-series state representation, and inputs it into the candidate policy network to obtain candidate scheduling actions.

[0125] The candidate policy network can adopt a constrained Markov decision process strategy or a multi-agent reinforcement learning strategy to enable each data center to generate candidate quantities for task processing, task migration, electricity purchase, green electricity use, and energy storage charging and discharging based on local resource status. The coordination unit then combines these with the capacity quota to form a cluster-level candidate decision. The candidate scheduling actions are then input into the security projection layer, and feasibility verification and repair are performed based on the hard constraints corresponding to equations (1) to (9). The security projection relationship can be expressed as equation (11): , in, The executable scheduling action vector is obtained after time period t is safely projected. To map candidate scheduling actions to a set of feasible decisions The safety projection operator is used to correct candidate actions that do not meet hard constraints; Let t be the set of feasible decisions that satisfy the hard constraints of the scheduling model within the range of uncertainties considered; This is a constraint function used to verify the feasibility of scheduling actions. Each component of the function corresponds to a constraint, and the fact that the function value of each component is not greater than zero indicates that the corresponding constraint is satisfied. The candidate scheduling action vector for time period t is output by the candidate policy network based on the current state, rolling prediction sequence, uncertainty set, capacity quota, and scheduling reference. The time period t is the scheduling action vector to be tested for constructing a feasible decision set. Its components include operation scheduling variables such as task processing volume, task migration volume, electricity purchase volume, green electricity usage volume, and energy storage charging and discharging volume. This is the set of prediction bias uncertainties constructed based on historical prediction residuals over time period t; Let be a prediction deviation vector within this set, used to characterize the potential deviations in prediction parameters such as task load, electricity price, nodal carbon intensity, and green energy availability. The constraint function combines the rolling prediction values ​​and this deviation vector to perform a feasibility check, where for all The constraint requirement indicates that the scheduling action should meet the corresponding hard constraints under each of the considered prediction deviation scenarios.

[0126] The safety projection layer can be implemented through linear programming, mixed-integer linear programming, quadratic programming, or control barrier function constraint repair, so as to reject or modify candidate actions when hard constraints are not met.

[0127] S505 executes feasible scheduling actions for the current scheduling period and sends task processing, task migration, power purchase, green electricity use, and energy storage control instructions to the corresponding data centers.

[0128] S506: Collect actual load, actual electricity purchase, actual carbon emissions, green electricity carbon credit, and constraint violations to form feedback data; use the feedback data to update the predictor and / or candidate strategies, and advance the rolling window forward by one scheduling period. Return to execute S501.

[0129] Feedback can be used simultaneously to calculate an evaluation value consisting of economic operating costs, net emissions, and penalties for violating constraints.

[0130] Therefore, the above embodiments can improve decision-making efficiency in large-scale cluster scenarios. Safe projection and rolling optimization are used to ensure that each actual execution action still conforms to the hard constraints of the original scheduling model. When the prediction module or candidate policy network is abnormal, the confidence is insufficient, or the calculation times out, it can directly fall back to the solution results of linear programming or mixed integer linear programming based on equations (1) to (9).

[0131] This application also provides an electronic device. Figure 6 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application. For example... Figure 6 As shown, the electronic device in this embodiment mainly includes a processor 601 and a memory 602. The memory 602 can be configured to store a program for executing the data center cluster scheduling method of the above-described method embodiments, and the processor 601 can be configured to execute the program in the memory 602. This program includes, but is not limited to, a program for executing the data center cluster scheduling method of the above-described method embodiments. For ease of explanation, only the parts related to the embodiments of this application are shown. For specific technical details not disclosed, please refer to the method section of the embodiments of this application.

[0132] In some embodiments, the electronic device may include multiple processors 601 and multiple memories 602. The program executing the data center cluster scheduling method of the above method embodiments can be divided into multiple subroutines, each of which can be loaded and run by a processor 601 to perform different steps of the data center cluster scheduling method of the above method embodiments. Specifically, each subroutine can be stored in a different memory 602, and each processor 601 can be configured to execute programs in one or more memories 602 to jointly implement the data center cluster scheduling method of the above method embodiments; that is, each processor 601 executes different steps of the data center cluster scheduling method of the above method embodiments to jointly implement the data center cluster scheduling method of the above method embodiments.

[0133] The aforementioned multiple processors 601 can be processors deployed on the same device. For example, the aforementioned electronic device can be a high-performance device composed of multiple processors, and the aforementioned multiple processors 601 can be processors configured on that high-performance device. Alternatively, the aforementioned multiple processors 601 can also be processors deployed on different devices. For example, the aforementioned electronic device can be a server cluster, and the aforementioned multiple processors 601 can be processors on different servers within the server cluster.

[0134] This application also provides a computer-readable storage medium. In one embodiment of the computer-readable storage medium according to this application, the computer-readable storage medium can be configured to store a program that performs the data center cluster scheduling method of the above-described method embodiments. This program can be loaded and run by a processor to implement the above-described data center cluster scheduling method. For ease of explanation, only the parts related to the embodiments of this application are shown; for specific technical details not disclosed, please refer to the method section of the embodiments of this application. The computer-readable storage medium can be a memory formed by various electronic devices. Optionally, in the embodiments of this application, the computer-readable storage medium is a non-transitory computer-readable storage medium.

[0135] The data center cluster scheduling method, electronic device, and readable storage medium provided in this application embodiment construct an accurate scheduling model, incorporate the accounting relationship between carbon management accounting, operational data, and operational scheduling variables into the scheduling model, and solve the problem with at least economic operating cost as the optimization objective. This enables accurate prediction of economic operating cost and carbon management accounting under different scheduling strategies, avoids conflicts between economic cost and carbon compliance cost in scheduling decisions, achieves comprehensive optimization of data center cluster operation, and improves the overall operational efficiency of data center clusters.

[0136] Exemplary embodiments of this disclosure have been specifically shown and described above. It should be understood that this disclosure is not limited to the detailed structures, arrangements, or implementations described herein; rather, this disclosure is intended to cover various modifications and equivalent arrangements contained within the spirit and scope of the appended claims.

Claims

1. A data center cluster scheduling method, characterized in that, include: Obtain operational data from the data center cluster; Based on the aforementioned operational data, a scheduling model is constructed; The decision variables of the scheduling model are the operation scheduling variables of the data center. The scheduling model is used to describe the accounting relationship between carbon management accounting, the operation data and the operation scheduling variables. The scheduling model is solved with at least economic operating cost as the optimization objective to obtain the scheduling decision; the economic operating cost is determined by the operation scheduling variables. Based on the scheduling decision, configure the operation scheduling variables of the data center.

2. The data center cluster scheduling method according to claim 1, characterized in that, The operational data includes the node carbon intensity, carbon quota, and green electricity carbon offset coefficient of the data center. The operational scheduling variables include the data center's regular electricity purchase, green electricity usage, and carbon purchase. The carbon management accounting quantities include actual emissions and green electricity carbon offset. The accounting relationships include the accounting relationship between actual emissions, regular electricity purchase, and node carbon intensity; the accounting relationship between green electricity carbon offset, green electricity usage, and green electricity carbon offset coefficient; and the accounting relationship between carbon purchase, actual emissions, green electricity carbon offset, and carbon quota.

3. The data center cluster scheduling method according to claim 1, characterized in that, The operational data includes the online task load and offline batch task load of the data center. The operational scheduling variables include the online task processing volume and offline batch task processing volume of the data center. Within a preset time period, the sum of the online task processing volume equals the online task load. Within a preset delay window, the cumulative amount of the offline batch task processing volume equals the offline batch task load. The accounting relationship includes the accounting relationship between the carbon management accounting volume, the online task load, and the offline batch task load.

4. The data center cluster scheduling method according to claim 3, characterized in that, The operational data also includes the task migration load of the data center, and the operational scheduling variables also include the task migration amount. Within a preset time period, the task migration amount does not exceed the task migration load. The accounting relationship includes the accounting relationship between the carbon management accounting amount, the online task load, the offline batch task load, and the task migration amount.

5. The data center cluster scheduling method according to claim 1, characterized in that, The operational data includes the data center's energy storage capacity, initial power, charging power limit, discharging power limit, charging efficiency, and discharging efficiency. The operational scheduling variables include the data center's energy storage charging amount and energy storage discharging amount. The energy storage charging amount is determined by the energy storage capacity, the initial power, the charging power limit, and the charging efficiency. The energy storage discharging amount is determined by the energy storage capacity, the initial power, the discharging power limit, and the discharging efficiency. The accounting relationship includes the accounting relationship between the carbon management accounting amount, the energy storage charging amount, and the energy storage discharging amount.

6. The data center cluster scheduling method according to any one of claims 1 to 5, characterized in that, The process of constructing a scheduling model based on the operational data includes: Boundary constraints are applied to the running data to obtain constrained running data; A linear relationship is established between the constrained running data and the running schedule variables to obtain the accounting relationship; The scheduling model is obtained by linearly fitting each of the aforementioned accounting relationships.

7. The data center cluster scheduling method according to any one of claims 1 to 5, characterized in that, Solving the scheduling model, with at least economic operating cost as the optimization objective, includes one of the following steps: The scheduling model is solved linearly with the goal of minimizing the economic operating cost. The scheduling model is solved linearly with the optimization objective of minimizing the linear fit between the economic operating cost and the carbon management accounting amount. Based on the constraints of the carbon management accounting amount, and with the minimization of the economic operating cost as the optimization objective, the scheduling model is solved linearly.

8. The data center cluster scheduling method according to any one of claims 1 to 5, characterized in that, The step of configuring the operation scheduling variables of the data center based on the scheduling decision includes: Convert the operational scheduling variables in the scheduling decision into power trading instructions; The power trading instructions are sent to the data center via the automatic power generation control interface; Based on the actual emissions, green electricity carbon credits, and carbon purchases calculated by the scheduling decision, emission analysis reports under different scheduling modes and / or sensitivity analysis reports under different parameter configurations are generated.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the data center cluster scheduling method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the data center cluster scheduling method according to any one of claims 1 to 8.