Multi-data center optimization planning and scheduling method containing calculation-electricity-heat-carbon synergy

Abstract

The application discloses a kind of multi-data center optimization planning and scheduling method containing calculation-electricity-heat-carbon coordination, comprising: establishing each data center portrait model;Analysis on the preset variety of modification scheme and data center new scheme planning to existing multi-data center, the dynamic characteristic change of each data center;Establish data center planning design optimization model, quantitative analysis comparison to the preset variety of modification scheme and data center new scheme;Establish data center global space-time model, predict the calculation power demand, power demand, renewable energy power supply, cold demand, carbon emission and waste heat recyclable amount prediction value of each data center in different time periods;Set calculation power task migration and power migration mechanism;Preliminarily formulate calculation power task migration strategy set, power transmission strategy set and heat collaborative utilization strategy set;Build the cooperation game model between the servers in the first layer data center, the master-slave game model of the second layer across data center.

Description

Technical Field

[0001] This invention belongs to the field of data center technology, specifically relating to a multi-data center optimization planning and scheduling method that incorporates computing, electricity, heat, and carbon synergy. Background Technology

[0002] With the rapid growth in computing power demand in data centers, the energy system of data centers is undergoing restructuring. Computing power is a new type of productivity that integrates information computing power, network carrying capacity, and data storage capacity. It mainly provides services to society through computing infrastructure. In recent years, the demand for artificial intelligence training and inference has grown rapidly, driving the accelerated construction of computing infrastructure, which in turn leads to a continuous increase in the electricity demand of data centers. Computing power puts forward three core requirements for power supply: stability, greenness, and low cost. However, the traditional power system has certain contradictions in terms of renewable energy consumption and power regulation. The energy supply model of data centers is undergoing profound changes. With the continuous expansion of computing infrastructure, data centers are gradually transforming from a single electricity load into an important adjustable power demand-side resource. Computing power is a flexible regulator of the new power system, and green electricity is the green engine of the digital economy. As a high-energy-consuming entity, the flexible load characteristics of computing centers can empower the power grid to absorb fluctuating renewable energy, while the decarbonization of the power system provides a green direction for computing power deployment.

[0003] However, current efforts are primarily focused on the synergistic advancement of computing power, electricity, and decarbonization, integrating and upgrading power systems with digital infrastructure such as networks, computing, and storage. There is relatively little attention paid to the synergy between computing power, electricity, heat, and carbon emissions. During data center operation, changes in computing load affect server heat generation and cooling system heat dissipation, generating recoverable waste heat. Heat, in turn, constrains computing power scheduling and power allocation. For example, high-power, high-heat computing tasks can be prioritized for data centers with well-developed waste heat recovery facilities and located near industrial heat loads, while low-heat tasks can be placed in data centers without waste heat recovery or in remote areas. Therefore, deep synergy between computing power, electricity, heat, and carbon emissions is the future development trend. Furthermore, most data centers lack resource and energy exchange mechanisms, resulting in poor flexibility and becoming a primary obstacle to the efficient, low-carbon, and collaborative development of the data center industry. For example, some data centers face problems of excess computing resources and long-term low-load server operation, leading to a waste of computing resources. On the other hand, some data centers frequently experience computing power shortages due to undertaking core business and high-performance computing tasks, and have to alleviate the pressure by temporarily adding servers and expanding data centers. This not only increases operating costs but also leads to low resource allocation efficiency. At the same time, due to the lack of cross-data center computing power migration and scheduling mechanisms, computing tasks cannot be dynamically allocated according to the load status and energy supply of each data center, resulting in extremely poor overall operational flexibility of data centers. They are unable to adapt to the temporal fluctuations and type changes of computing tasks and cannot achieve global optimization of computing resources.

[0004] Moreover, with the development of computing power, electricity, heat and decarbonization, existing data centers also need to be upgraded and newly built. Most of the existing data centers were built a long time ago and have problems such as outdated server performance, high energy consumption of energy equipment, lack of waste heat recovery facilities and insufficient utilization of renewable energy. They urgently need to improve energy efficiency and enhance collaboration capabilities through technological transformation. At the same time, they also need to build new data centers to cope with the explosive growth in computing power demand. These are all problems that urgently need to be solved for the future operation and development of data centers.

[0005] Based on the above technical issues, a new method for optimizing planning and scheduling of multi-data centers that incorporates computing, electricity, heat, and carbon coordination is needed. Summary of the Invention

[0006] The technical problem to be solved by this invention is to overcome the shortcomings of the prior art and provide a multi-data center optimization planning and scheduling method with computing-electricity-heat-carbon synergy. It mainly achieves global optimization of computing power, electricity, and heat resources through data collection, profile establishment, planning and design, predictive scheduling, and optimization game theory, reduces resource gaps in each data center, improves the utilization rate of renewable energy, and reduces global carbon emissions through collaborative scheduling and low-carbon management, thereby improving economic efficiency and low carbon emissions. Moreover, it can dynamically adapt to scenarios such as computing power task growth, changes in energy endowment, and adjustments in industrial heat demand, achieving deep synergy between computing, electricity, heat, and carbon.

[0007] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: This invention provides a multi-data center optimization planning and scheduling method incorporating computation-electricity-heat-carbon synergy, comprising: S1. Data Collection and Profile Building: Collect multi-source data related to computing power, electricity, heat and carbon emissions of existing data centers in the previous year, analyze the computing power resource gap, electricity resource gap, heat recovery resource gap, carbon emissions, server load status and energy utilization efficiency of each data center in the previous year, and build a profile model of each data center. S2. Renovation and New Construction Analysis: Based on the profile model of each data center, we simultaneously acquire data on renewable energy endowment, industrial heat demand, planning budget, and available site selection around each data center. Combined with the phased evolution data of computing power tasks of each data center, we analyze the dynamic characteristics of each data center after planning various pre-set renovation schemes and new data center construction schemes for existing multi-data centers. S3. Establishment of planning and design model: Based on the dynamic characteristics of each data center, with the goal of minimizing the cost of building new data centers and transforming existing data centers, the resource gap of each data center and the carbon emissions, establish a data center planning and design optimization model, conduct quantitative analysis and comparison of various preset transformation schemes and new data center schemes, and decide on the planning and design scheme of existing multiple data centers. S4. Spatiotemporal Feature Extraction and Demand Forecasting: After the planning and design are completed, a global spatiotemporal model of the data center is established to characterize the spatial correlation structure and coordinated operation relationship of all data centers over time. Then, the spatial correlation features and temporal features of all data centers at the current time are extracted to predict the computing power demand, power demand, renewable energy power supply, cooling demand, carbon emissions and waste heat recovery of each data center in different time periods. S5. Resource Migration Mechanism Settings: When migrating computing power tasks in a data center, priority is given to migrating computing power tasks between servers within the data center. If, after the internal migration, at least one of the following conditions is met: computing power task computation still has a shortfall, carbon emissions exceed the standard, or server operating temperature exceeds the required range, then cross-data center computing power task migration is initiated to ensure that each data center meets the preset standards after migration. Additionally, a cross-data center power migration mechanism is set up to migrate surplus power to data centers with power shortages through preset transmission links. S6. Preliminary scheduling strategy formulation: Based on the current power consumption information, pending computing power task information, available computing power information, power supply capacity, carbon emission limits and server operating temperature of each data center, combined with the resource migration mechanism and various prediction values ​​of each data center, a preliminary set of computing power task migration strategy, power transmission strategy and thermal synergistic utilization strategy is formulated. S7. Two-layer game model construction: Construct a cooperative game model among servers within the first layer of the data center and a master-slave game model across the second layer of the data center. Through iterative interaction and solution of the two-layer game model, obtain the final computing power task migration scheme, power transmission scheme, heat co-utilization scheme and related pricing strategy.

[0008] Furthermore, S1 includes: Each data center is modeled as an independent intelligent agent, serving as an intelligent management and control unit for autonomously sensing, analyzing, and making decisions regarding the operational status of each data center. Each intelligent agent collects multi-source data related to computing power, electricity, heat and carbon emissions for each period of the previous year. For data centers with their own renewable energy generator sets, additional data are collected on renewable energy power generation, generator set operating efficiency, renewable energy power supply ratio and energy storage device charging and discharging. For data centers with waste heat recovery devices, additional data are collected on waste heat recovery amount, waste heat recovery efficiency, waste heat output and waste heat recovery device operating parameters. Each agent analyzes the computing power resource gap based on the collected multi-source data, outputting the computing power gap value, gap duration, gap peak value, and remaining amount of computing power for each time period of the previous year, forming a computing power resource gap time series curve; analyzes the power resource gap, outputting the power gap value, gap duration, remaining amount of power for each time period of the previous year, and remaining power availability; analyzes the heat recovery resource gap, outputting the heat recovery gap value, recovery efficiency, and total amount of data center waste heat that can be recovered without waste heat recovery devices, as well as its potential recovery value, for each time period of the previous year; analyzes carbon emissions, outputting the carbon emissions, exceedances, and carbon emission reductions corresponding to renewable energy generation for each time period of the previous year; analyzes server load status, outputting the average server load rate for each time period of the previous year; and analyzes energy utilization efficiency, outputting the PUE value, thermal utilization efficiency, and renewable energy utilization efficiency for each time period of the previous year. By combining the multidimensional analysis data of each intelligent agent, core operational feature labels and structural feature labels are formed for each data center profile model. The data centers are classified using feature quantization, weight allocation and cluster analysis methods, and a profile model for each data center is formed.

[0009] Furthermore, S2 includes: Based on the profile models of each data center, and simultaneously acquiring data on the renewable energy endowment, industrial heat demand, planning budget, and feasible site selection around each data center, and combining this with the phased evolution data of computing power tasks for each data center, multiple renovation schemes and multiple new data center construction schemes are pre-set. The renewable energy endowment data includes the exploitable amount and power generation potential time-series data of surrounding photovoltaic and wind power renewable energy sources; the phased evolution data of computing power tasks includes the growth trend and changes in computing power task types for each data center at different times within a pre-set future period; the renovation schemes include adding renewable energy generators and energy storage equipment to data centers without renewable energy generators, adding waste heat recovery devices to data centers without waste heat recovery devices, optimizing the waste heat output link and recovery efficiency for data centers with waste heat recovery devices, and upgrading and optimizing existing server clusters and energy equipment; the new construction schemes include site selection for new data centers, the scale of computing server construction, equipment deployment types, and energy equipment capacity planning. Based on each pre-set renovation and new construction plan, the operation status of each data center is simulated after the implementation of the plan, and the dynamic characteristics of each data center are analyzed, including dynamic changes in the supply and demand of computing resources, dynamic changes in the supply and demand and supply structure of power resources, dynamic changes in the utilization of heat recovery resources and heat synergy, dynamic changes in carbon emissions, dynamic changes in server load balance, dynamic changes in energy utilization efficiency, and changes in economic benefits.

[0010] Furthermore, S3 includes: With the objectives of minimizing the costs of building new data centers and upgrading existing data centers, as well as the resource gaps and carbon emissions of each data center, an optimization model for data center planning and design is established, which is expressed as follows: ; The total cost of new construction and renovation is n; n is the number of existing data centers that need to be renovated. The equipment procurement cost for the renovation of the i-th existing data center; The installation and construction cost for the renovation of the i-th existing data center; Let m be the maintenance cost after the i-th data center is upgraded; m is the number of newly built data centers. The site selection and land acquisition cost for the j-th newly built data center; The construction cost of the j-th newly built data center; The equipment deployment cost for the j-th newly built data center; The operating cost of the j-th newly built data center; ; To address the overall resource gap; , , These are the weighting coefficients for the resource gaps in computing power, electricity, and heat recovery, respectively; t is the time period within the planning cycle; k is the number of all data centers after renovation and new construction. This represents the computing power resource gap for the k-th data center during time period t. This represents the power resource gap for the k-th data center during time period t. This represents the heat recovery resource gap for the k-th data center during time period t. ; Carbon emissions corresponding to powering the grid for the k-th data center during time period t; This represents the carbon emissions corresponding to the operation of equipment in the k-th data center during time period t. The constraints of the planning and design optimization model include: planning budget constraints, site selection constraints, equipment deployment constraints, renewable energy utilization constraints, thermal synergistic utilization constraints, carbon emission compliance constraints, server load balancing constraints, and resource supply and demand balance constraints. The site selection constraints are adapted to the geographical conditions, energy supply conditions, and surrounding industrial heat demand distribution of the plannable site selection data. The equipment deployment constraints are adapted to the structural differences of different data centers and the equipment configuration requirements of renovation and new construction schemes. The resource supply and demand balance constraints include computing power supply and demand balance, power supply and demand balance, and thermal resource supply and demand balance. Among them, the power supply and demand balance needs to be adapted to the power supply capacity of renewable energy generator sets, and the thermal resource supply and demand balance needs to be adapted to the recovery efficiency of waste heat recovery devices and surrounding industrial heat demand. Substitute each preset renovation scheme and new construction scheme into the planning and design optimization model, solve the model, perform quantitative analysis and comparison of each scheme, and output the multi-objective optimization target achievement degree of each scheme; Based on the quantitative analysis and comparison results, and combined with the priority of multi-objective optimization, the scheme that simultaneously meets the constraints and the requirement of minimizing resource gaps, and achieves the comprehensive optimization of transformation cost, new construction cost and carbon emissions, is selected as the final planning and design scheme for the existing multi-data center.

[0011] Furthermore, S4 includes: After the planning and design are completed, obtain the operational and spatial data of all data centers in the current period. The spatial data includes the geographical location, physical distance, power transmission links, computing network topology, waste heat transmission links, and regional energy supply boundaries of each data center. Based on operational and spatial data, each data center is abstracted as a spatiotemporal node. By establishing the physical connections, energy transmission relationships, and computing power scheduling relationships between data centers, a global spatiotemporal model of the data center is formed, which represents the spatial relationship structure and coordinated operation relationship of all data centers that dynamically change over time. A spatiotemporal graph convolutional neural network is used to extract features from the global spatiotemporal model of the data center, outputting temporal features and spatial correlation features respectively. The features are then fused to obtain the spatiotemporal joint features of each data center. The temporal features include the time-series trends, periodic patterns, fluctuation characteristics, and peak-valley patterns of the operational data of each data center. The spatial correlation features include the spatial distance correlation between data centers, energy transmission loss characteristics, computing power network coupling characteristics, regional energy endowment constraints, and the coupling correlation of multiple indicators such as computing, electricity, heat, and carbon. The spatiotemporal joint features are input into the preset prediction model to predict the key operating parameters of each data center at different times within the preset period in the future, including: computing power demand, electricity demand, renewable energy power supply, cooling demand, carbon emissions, and waste heat recovery.

[0012] Furthermore, in S5, the computing power task migration mechanism is set up, including: The data center computing power migration mechanism adopts a layered migration logic of internal migration and cross-center migration: When migrating computing power tasks in a data center, priority is given to migrating computing power tasks between servers within the corresponding data center, migrating the computing power of overloaded servers to servers in a light-load, normal-load state; after the internal computing power task migration is completed, the operating status of the data center is detected, and if the detection result meets at least one preset condition, cross-data center computing power task migration is initiated. The conditions include: there is still a gap in computing power for the task, meaning that the current available computing power of the data center cannot meet the real-time computing power demand; carbon emissions exceed the standard, meaning that the current carbon emissions of the data center exceed the preset carbon emission compliance threshold; and server operating temperature exceeds the preset reasonable range, meaning that the server operating temperature is higher than the preset temperature threshold.

[0013] Furthermore, in S5, the power migration mechanism is configured as follows: By assessing the power supply and demand status of each data center, distinguishing between data centers with surplus power and those with power shortages, calculating the power surplus or shortage for each data center, and determining the supply and demand correspondence, migration amount, and transmission links for power migration based on power transmission links, the link with the least transmission loss is prioritized for power migration.

[0014] Furthermore, S6 includes: The system acquires current power consumption, pending computing tasks, available computing power, power supply capacity, carbon emission limits, and server operating temperatures for each data center. Combining this with resource migration mechanisms and various predicted values ​​from each data center, it performs feasibility assessments: determining whether inter-server computing task migration within a data center is feasible and whether cross-data center computing task migration needs to be initiated; determining whether each data center has surplus or shortage of power and whether it meets the conditions for cross-data center power transmission; and determining whether the recoverable waste heat, industrial heat demand, and heat transmission links of each data center meet the conditions for coordinated heat utilization. Based on the above judgment results, a set of computing power task migration strategies, a set of power transmission strategies, and a set of thermal co-utilization strategies were initially formulated. The set of computing power task migration strategies includes internal migration schemes, cross-center migration directions, migration task volume, migration priority, and constraints. The set of power transmission strategies includes supply and demand matching relationships, transmission volume, path, and loss constraints. The set of thermal co-utilization strategies includes waste heat supply and demand matching relationships, transmission volume, path, and utilization efficiency constraints. The generated strategy set is initially verified to ensure that each strategy meets the requirements of computing power supply and demand matching, power supply and demand matching, carbon emission compliance, server temperature compliance, and thermal supply and demand matching, thus forming a preliminary feasible scheduling strategy set.

[0015] Furthermore, in S7, a cooperative game model is constructed among the servers within the first-layer data center, including: Within a single data center, servers engage in cooperative game-playing. The game participants are all servers within the k-th data center. The local status of each server includes computing load, computing capacity, power consumption, operating temperature, and carbon emissions. The objective of cooperative game theory is to maximize global utility. The global utility function for the k-th data center is... Represented as: ; , , These are the weighting coefficients for each sub-utility; For load balancing effectiveness, it represents the load difference between servers; the smaller the difference, the higher the effectiveness. For carbon emission compliance utility, it is characterized that the lower the carbon emissions, the higher the utility. To ensure compliance with operating temperature requirements, the higher the performance of the temperature within a reasonable range, the higher the operational safety and stability. By solving the Pareto optimal solution of the cooperative game model using the Nash bargaining algorithm, the optimal computing power task migration allocation scheme for each server is obtained, expressed as: ; This refers to the amount of computing power allocated to the i-th server in the k-th data center. Let i be the individual utility of the i-th server in the k-th data center; This is the point at which negotiations break down for the i-th server in the k-th data center; This represents the total number of servers in the k-th data center.

[0016] Furthermore, in S7, a second-layer cross-data center master-slave game model is constructed, including: A multi-indicator comprehensive evaluation method is adopted to construct a data center evaluation index system. The evaluation indexes include data center scale, operational reliability, computing power capacity, power supply reliability, network connectivity, historical scheduling success rate, and carbon emission compliance rate. Master-slave game participant setup: The comprehensive evaluation value of each data center is obtained through weighted scoring. The data center with the highest evaluation value is selected as the game leader, and the other data centers are followers. Master-follower game sequence setting: Each follower reports its own state to the leader. The leader combines the initial scheduling strategy set and resource migration mechanism to maximize the global collaborative utility and decides on the global resource allocation plan. Each follower then adjusts its own operating strategy based on the leader's decision to maximize its own utility and iterates to the Stackelberg equilibrium, where the leader's decision is optimal and the followers have no incentive to change their own strategies. Among them, the leader's overall synergistic effect The target is represented as: ; , , , These are the sub-utility weighting coefficients; The total cost utility includes computing power migration costs, power migration costs, and heat transfer costs. For the overall resource gap utility; For the overall carbon emissions utility; To maximize the overall utility, including revenue from computing power services and revenue from electricity and heat trading; The self-utility of the s-th follower The target is represented as: ; , Weighting coefficients for local operational utility and transaction revenue utility; For local operating utility, the same as the utility of the first-level game model. ; For the benefit of transaction gains.

[0017] The beneficial effects of this invention are: (1) This invention collects multi-source data such as computing power, electricity, heat, and carbon emissions from the previous year, and systematically analyzes key indicators such as resource gaps, server load, and energy utilization efficiency in each period. It clearly identifies the core problems of each data center (such as excessive computing power gaps in some data centers, frequent carbon emissions exceeding standards in some data centers, and low energy utilization efficiency in some data centers), avoiding blind planning. In addition, the established profile model of each data center quantifies the core dimensions such as structural characteristics, computing power characteristics, and electricity characteristics of each data center. The design of subsequent renovation and new construction plans can fit the actual situation of each data center, avoiding a one-size-fits-all planning model and improving the adaptability of the planning scheme. (2) This invention combines data from multiple dimensions, such as data center profiles, surrounding renewable energy endowments, and industrial heat demand, to pre-set various renovation and new construction schemes, avoiding the limitations of a single scheme and ensuring that the scheme can adapt to the future computing power and energy demand changes of different data centers; by simulating the dynamic characteristics changes of data centers after the implementation of each scheme (computing power, electricity, carbon emissions, etc.), the implementation effect of each scheme can be known in advance, the advantages and disadvantages of the scheme can be clarified, and quantitative support can be provided for subsequent planning decisions, avoiding the problem of the scheme not achieving the expected results after implementation; and, by combining the phased evolution data of computing power tasks, the future computing power demand growth trend can be predicted in advance, ensuring that the renovation and new construction schemes can adapt to future development, avoiding repeated renovation and new construction in the short term, improving resource utilization efficiency, and reducing long-term operating costs; at the same time, by combining the surrounding renewable energy and industrial heat demand, the energy synergy path can be planned in advance, laying the foundation for subsequent computing-electricity-heat-carbon synergy; (3) The present invention takes the minimization of cost, resource gap and carbon emissions as its core objective. While controlling the total cost of renovation and new construction, it reduces the resource gap of computing power, electricity and other resources of each data center, reduces the overall carbon emissions, and achieves a triple balance of economy, efficiency and low carbon. By establishing a planning and design optimization model, it conducts quantitative analysis and comparison of multiple preset schemes, quantifies the cost, resource gap and carbon emissions of each scheme, avoids the bias of subjective decision-making, and ensures that the final selected planning scheme is the best overall and feasible. (4) This invention establishes a global spatiotemporal model, abstracts each data center into a spatiotemporal node, captures the spatial correlation and temporal dynamic changes between data centers, breaks the limitation of independent scheduling of each data center in the existing scheduling, and provides support for cross-data center collaborative scheduling; it uses a spatiotemporal graph convolutional neural network to extract spatiotemporal joint features, and combines time features and spatial correlation features to achieve high-precision prediction of key parameters such as computing power, electricity, and carbon emissions, avoids scheduling errors caused by prediction deviations, provides accurate basis for the formulation of preliminary scheduling strategies, and realizes advance scheduling and proactive response; (5) This invention prioritizes the migration of computing power between servers within a data center to avoid problems such as increased network latency and energy consumption caused by cross-data center migration. It also reduces the risk of task interruption during the computing power migration process and improves the efficiency and stability of computing power migration. Cross-data center migration is only initiated when internal migration cannot meet the demand to avoid blind migration and ensure that each data center meets the preset standards after migration. Furthermore, through the cross-data center power migration mechanism, surplus power (such as surplus from renewable energy generation) is migrated to data centers with power shortages to reduce power curtailment, improve the utilization rate of renewable energy, and reduce the grid purchase cost of power shortage data centers, thereby achieving global optimization of power resource allocation. (6) This invention combines the real-time operating status, predicted values ​​and resource migration mechanisms of each data center, and formulates three types of scheduling strategy sets: computing power, electricity and heat, to ensure that the scheduling strategy can cope with various operating scenarios; the preliminary scheduling strategy set provides an initial basis for solving the subsequent two-layer game model, reduces the number of iterations of game solving, and improves the optimization efficiency of the scheduling scheme. (7) The first-layer cooperative game model of this invention optimizes the resource allocation among servers within the data center and maximizes the local utility of a single data center; the second-layer master-slave game model optimizes the collaborative scheduling across data centers and maximizes the global collaborative utility, avoiding the problem of local optima but global inefficiency; through the iterative interactive solution of the two-layer game, the scheduling scheme can be dynamically adjusted according to the real-time operating status of each data center, ensuring that the scheduling scheme can adapt to changes in the operating status and improving the flexibility and adaptability of the scheduling; and, through the game model, computing power scheduling, power transmission, thermal coordination and carbon emission control are deeply bound together to achieve multi-dimensional resource collaborative optimization, break the barriers of independent operation of each system, improve the overall system operating efficiency and low carbon level, and achieve deep collaboration of computing-electricity-heat-carbon.

[0018] Other features and advantages will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a multi-data center optimization planning and scheduling method with computing-electricity-heat-carbon synergy according to the present invention; Figure 2 This is a schematic diagram illustrating the principle of the two-layer game model of the present invention; Figure 3 The flowchart for establishing the global spatiotemporal model and spatiotemporal prediction method for the data center in this invention is shown. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] like Figure 1 , Figure 2 As shown, this embodiment provides a multi-data center optimization planning and scheduling method with computing-electricity-heating-carbon synergy, which includes: S1. Data Collection and Profile Building: Collect multi-source data related to computing power, electricity, heat and carbon emissions of existing data centers in the previous year, analyze the computing power resource gap, electricity resource gap, heat recovery resource gap, carbon emissions, server load status and energy utilization efficiency of each data center in the previous year, and build a profile model of each data center. S2. Renovation and New Construction Analysis: Based on the profile model of each data center, we simultaneously acquire data on renewable energy endowment, industrial heat demand, planning budget, and available site selection around each data center. Combined with the phased evolution data of computing power tasks of each data center, we analyze the dynamic characteristics of each data center after planning various pre-set renovation schemes and new data center construction schemes for existing multi-data centers. S3. Establishment of planning and design model: Based on the dynamic characteristics of each data center, with the goal of minimizing the cost of building new data centers and transforming existing data centers, the resource gap of each data center and the carbon emissions, establish a data center planning and design optimization model, conduct quantitative analysis and comparison of various preset transformation schemes and new data center schemes, and decide on the planning and design scheme of existing multiple data centers. S4. Spatiotemporal Feature Extraction and Demand Forecasting: After the planning and design are completed, a global spatiotemporal model of the data center is established to characterize the spatial correlation structure and coordinated operation relationship of all data centers over time. Then, the spatial correlation features and temporal features of all data centers at the current time are extracted to predict the computing power demand, power demand, renewable energy power supply, cooling demand, carbon emissions and waste heat recovery of each data center in different time periods. S5. Resource Migration Mechanism Settings: When migrating computing power tasks in a data center, priority is given to migrating computing power tasks between servers within the data center. If, after the internal migration, at least one of the following conditions is met: computing power task computation still has a shortfall, carbon emissions exceed the standard, or server operating temperature exceeds the required range, then cross-data center computing power task migration is initiated to ensure that each data center meets the preset standards after migration. Additionally, a cross-data center power migration mechanism is set up to migrate surplus power to data centers with power shortages through preset transmission links. S6. Preliminary scheduling strategy formulation: Based on the current power consumption information, pending computing power task information, available computing power information, power supply capacity, carbon emission limits and server operating temperature of each data center, combined with the resource migration mechanism and various prediction values ​​of each data center, a preliminary set of computing power task migration strategy, power transmission strategy and thermal synergistic utilization strategy is formulated. S7. Two-layer game model construction: Construct a cooperative game model among servers within the first layer of the data center and a master-slave game model across the second layer of the data center. Through iterative interaction and solution of the two-layer game model, obtain the final computing power task migration scheme, power transmission scheme, heat co-utilization scheme and related pricing strategy.

[0024] In this embodiment, S1 includes: Each data center is modeled as an independent intelligent agent, serving as an intelligent management and control unit for autonomously sensing, analyzing, and making decisions regarding the operational status of each data center. Each intelligent agent collects multi-source data related to computing power, electricity, heat and carbon emissions for each period of the previous year. For data centers with their own renewable energy generator sets, additional data are collected on renewable energy power generation, generator set operating efficiency, renewable energy power supply ratio and energy storage device charging and discharging. For data centers with waste heat recovery devices, additional data are collected on waste heat recovery amount, waste heat recovery efficiency, waste heat output and waste heat recovery device operating parameters. Each agent analyzes the computing power resource gap based on the collected multi-source data, outputting the computing power gap value, gap duration, gap peak value, and remaining amount of computing power for each time period of the previous year, forming a computing power resource gap time series curve; analyzes the power resource gap, outputting the power gap value, gap duration, remaining amount of power for each time period of the previous year, and remaining power availability; analyzes the heat recovery resource gap, outputting the heat recovery gap value, recovery efficiency, and total amount of data center waste heat that can be recovered without waste heat recovery devices, as well as its potential recovery value, for each time period of the previous year; analyzes carbon emissions, outputting the carbon emissions, exceedances, and carbon emission reductions corresponding to renewable energy generation for each time period of the previous year; analyzes server load status, outputting the average server load rate for each time period of the previous year; and analyzes energy utilization efficiency, outputting the PUE value, thermal utilization efficiency, and renewable energy utilization efficiency for each time period of the previous year. By combining the multidimensional analysis data of each intelligent agent, core operational feature labels and structural feature labels are formed for each data center profile model. The data centers are classified using feature quantization, weight allocation and cluster analysis methods, and a profile model for each data center is formed.

[0025] In practical applications, each data center is modeled as an independent intelligent agent. Combining the structural differences and profile characteristics of each data center, each intelligent agent is configured with corresponding functional modules and serves as an intelligent management and control unit for each data center to autonomously perceive, analyze, and make decisions on the data center's operational status. Each intelligent agent can then perform subsequent modifications and new construction analyses, planning and design model establishment, spatiotemporal feature extraction and demand forecasting, and scheduling strategy decisions based on the corresponding functional modules.

[0026] When building a profile model, each intelligent agent needs to collect data on computing power load, computing power task completion, computing power idle time, computing power request, total electricity consumption, grid power supply, purchased electricity price, total server heat dissipation, heat dissipation energy consumption, indoor environmental heat, heat loss, total carbon emissions, carbon emissions per unit of computing power, carbon emission compliance threshold, number of servers, load rate of a single server, server runtime, total energy consumption, and the proportion of each type of energy consumption for each period of the previous year.

[0027] Structural feature tags: These indicate whether the data center has its own renewable energy generator sets, whether it includes waste heat recovery devices, and the core parameters of related equipment (such as generator set type and waste heat recovery device capacity). Core operational feature tags include computing power feature tags, power feature tags, thermal feature tags, carbon emission feature tags, server feature tags, and energy utilization feature tags. The data center classification includes at least high computing power gap type, low computing power gap type, high carbon emission type, green and efficient type, and thermal synergy type.

[0028] In this embodiment, S2 includes: Based on the profile models of each data center, and simultaneously acquiring data on the renewable energy endowment, industrial heat demand, planning budget, and feasible site selection around each data center, and combining this with the phased evolution data of computing power tasks for each data center, multiple renovation schemes and multiple new data center construction schemes are pre-set. The renewable energy endowment data includes the exploitable amount and power generation potential time-series data of surrounding photovoltaic and wind power renewable energy sources; the phased evolution data of computing power tasks includes the growth trend and changes in computing power task types for each data center at different times within a pre-set future period; the renovation schemes include adding renewable energy generators and energy storage equipment to data centers without renewable energy generators, adding waste heat recovery devices to data centers without waste heat recovery devices, optimizing the waste heat output link and recovery efficiency for data centers with waste heat recovery devices, and upgrading and optimizing existing server clusters and energy equipment; the new construction schemes include site selection for new data centers, the scale of computing server construction, equipment deployment types, and energy equipment capacity planning. Based on each pre-set renovation and new construction plan, the operation status of each data center is simulated after the implementation of the plan, and the dynamic characteristics of each data center are analyzed, including dynamic changes in the supply and demand of computing resources, dynamic changes in the supply and demand and supply structure of power resources, dynamic changes in the utilization of heat recovery resources and heat synergy, dynamic changes in carbon emissions, dynamic changes in server load balance, dynamic changes in energy utilization efficiency, and changes in economic benefits.

[0029] In practical applications, renewable energy endowment data refers to the basic information and dynamic power generation data of renewable energy sources such as photovoltaics and wind power available around each data center. This data is used to determine the feasibility, scale, and power generation benefits of adding renewable energy equipment to data centers. Data on the phased evolution of computing power tasks is used to predict changes in the computing power demand of each data center within a future preset period (such as a 5-year planning period). This ensures that renovation and new construction plans can adapt to the growth in computing power and adjustments in task types. The growth in computing power tasks includes the growth rate of computing power demand by year, quarter, and month, and changes in peak computing power demand. Adjustments in computing power task types include the distribution and evolution trend of computing power task types in different time periods. For example, some data centers currently primarily perform ordinary data processing and file storage tasks. In the next three years, they will add high-performance computing power tasks such as AI model training and big data analysis. These tasks have higher requirements for server performance and power stability. The data also includes changes in the latency requirements of computing power tasks, such as reducing the latency requirement for core business computing power tasks from less than 10ms to less than 5ms. Additionally, there are changes in task concurrency, such as an initial concurrent computing power task volume of 800 tasks per hour, which is expected to reach 1200 tasks per hour after two years. Industrial heat demand data refers to the heat load demand of industrial enterprises surrounding the data center, including the heat demand, temperature requirements, and heat transport distance at different times. Planning budget data specifies the total budget amount within the planning period, the allocation of sub-budgets (such as renovation budgets, new construction budgets, etc.), and budget adjustment thresholds. Plannable site selection data includes the possible locations for new data centers, land use costs for each location, surrounding infrastructure, environmental constraints, and the availability of renewable energy in the surrounding area.

[0030] The renovation plan primarily addresses the structural deficiencies and performance weaknesses of existing data centers. It incorporates their profile characteristics, renewable energy endowment data, phased evolution data of computing power tasks, industrial heat demand data, planning budget data, and planarable site selection data to pre-determine multiple differentiated renovation schemes, focusing on three core dimensions: renewable energy utilization, waste heat recovery, and equipment upgrades. The new construction plan, based on planarable site selection data, phased evolution data of computing power tasks, and renewable energy endowment data, pre-determines multiple differentiated new construction schemes, clearly defining the core configuration of the new data center to ensure it can adapt to future computing power demands and energy synergy requirements. Based on each pre-determined renovation and new construction scheme, professional simulation software is used to simulate the real-time operating status of each data center within the planning period after implementation. Combining the profile models and basic data of each data center, a comprehensive analysis of the dynamic characteristics changes of each data center is conducted to clarify the effectiveness of the scheme implementation.

[0031] In this embodiment, S3 includes: With the objectives of minimizing the costs of building new data centers and upgrading existing data centers, as well as the resource gaps and carbon emissions of each data center, an optimization model for data center planning and design is established, which is expressed as follows: ; The total cost of new construction and renovation is n; n is the number of existing data centers that need to be renovated. The equipment procurement cost for the renovation of the i-th existing data center; The installation and construction cost for the renovation of the i-th existing data center; Let m be the maintenance cost after the i-th data center is upgraded; m is the number of newly built data centers. The site selection and land acquisition cost for the j-th newly built data center; The construction cost of the j-th newly built data center; The equipment deployment cost for the j-th newly built data center; The operating cost of the j-th newly built data center; ; To address the overall resource gap; , , These are the weighting coefficients for the resource gaps in computing power, electricity, and heat recovery, respectively; t is the time period within the planning cycle; k is the number of all data centers after renovation and new construction. This represents the computing power resource gap for the k-th data center during time period t. This represents the power resource gap for the k-th data center during time period t. This represents the heat recovery resource gap for the k-th data center during time period t. ; Carbon emissions corresponding to powering the grid for the k-th data center during time period t; This represents the carbon emissions corresponding to the operation of equipment in the k-th data center during time period t. The constraints of the planning and design optimization model include: planning budget constraints, site selection constraints, equipment deployment constraints, renewable energy utilization constraints, thermal synergistic utilization constraints, carbon emission compliance constraints, server load balancing constraints, and resource supply and demand balance constraints. The site selection constraints are adapted to the geographical conditions, energy supply conditions, and surrounding industrial heat demand distribution of the plannable site selection data. The equipment deployment constraints are adapted to the structural differences of different data centers and the equipment configuration requirements of renovation and new construction schemes. The resource supply and demand balance constraints include computing power supply and demand balance, power supply and demand balance, and thermal resource supply and demand balance. Among them, the power supply and demand balance needs to be adapted to the power supply capacity of renewable energy generator sets, and the thermal resource supply and demand balance needs to be adapted to the recovery efficiency of waste heat recovery devices and surrounding industrial heat demand. Substitute each preset renovation scheme and new construction scheme into the planning and design optimization model, solve the model, perform quantitative analysis and comparison of each scheme, and output the multi-objective optimization target achievement degree of each scheme; Based on the quantitative analysis and comparison results, and combined with the priority of multi-objective optimization, the scheme that simultaneously meets the constraints and the requirement of minimizing resource gaps, and achieves the comprehensive optimization of transformation cost, new construction cost and carbon emissions, is selected as the final planning and design scheme for the existing multi-data center.

[0032] like Figure 3 As shown, in this embodiment, S4 includes: After the planning and design are completed, obtain the operational and spatial data of all data centers in the current period. The spatial data includes the geographical location, physical distance, power transmission links, computing network topology, waste heat transmission links, and regional energy supply boundaries of each data center. Based on operational and spatial data, each data center is abstracted as a spatiotemporal node. By establishing the physical connections, energy transmission relationships, and computing power scheduling relationships between data centers, a global spatiotemporal model of the data center is formed, which represents the spatial relationship structure and coordinated operation relationship of all data centers that dynamically change over time. A spatiotemporal graph convolutional neural network is used to extract features from the global spatiotemporal model of the data center, outputting temporal features and spatial correlation features respectively. The features are then fused to obtain the spatiotemporal joint features of each data center. The temporal features include the time-series trends, periodic patterns, fluctuation characteristics, and peak-valley patterns of the operational data of each data center. The spatial correlation features include the spatial distance correlation between data centers, energy transmission loss characteristics, computing power network coupling characteristics, regional energy endowment constraints, and the coupling correlation of multiple indicators such as computing, electricity, heat, and carbon. The spatiotemporal joint features are input into the preset prediction model to predict the key operating parameters of each data center at different times within the preset period in the future, including: computing power demand, electricity demand, renewable energy power supply, cooling demand, carbon emissions, and waste heat recovery.

[0033] It should be noted that the data center's operational data includes real-time computing load, CPU / GPU utilization, number of concurrent tasks, electricity purchased from the grid, actual renewable energy generation, energy storage charging and discharging status, server inlet and outlet water temperatures, chiller power consumption, air conditioning energy consumption, real-time PUE value, real-time carbon emissions, waste heat recovery temperature and flow rate, actual heat supply, peak and off-peak periods of power load, peak computing periods, and equipment failure and redundancy status.

[0034] Spatial data is used to characterize the location relationships, transmission characteristics, energy boundaries, and network structure of data centers within a physical area, specifically including: Geographic location: Latitude and longitude coordinates of each data center, distribution within the park, and administrative division; Physical distance: the straight-line distance between any two data centers, and the actual pipeline path length; Power transmission links: substation affiliation, line impedance, upper limit of transmission capacity, allowable transmission power, and line loss coefficient; Computing network topology: fiber optic routing, switch hierarchy, bandwidth limit, inter-node communication latency, packet loss rate, available computing power scheduling channels; Waste heat transport link: routing of heat pipe network, pipe diameter, insulation level, allowable transport temperature, transport efficiency, and maximum transport distance; Regional energy supply boundaries: maximum power supply capacity of the regional power grid, total capacity of the regional heating network, upper limit of renewable energy consumption, and regional carbon emission quota boundaries.

[0035] Each data center is abstracted as a spatiotemporal node, and directed and undirected edges are constructed based on the actual relationships between data centers, ultimately forming a global spatiotemporal topology model covering all data centers. Each node contains not only static attributes (location, capacity, equipment configuration, etc.) but also dynamic attributes (load, energy consumption, carbon emissions, waste heat, etc. over time periods), hence the name spatiotemporal node, which reflects the dual characteristics of changing over time and being spatially related. The edges between nodes are established based on real interaction relationships, including physical connections: power line connections, heat pipe connections, and fiber optic connections, constructing basic edges; energy transmission relationships: enabling power sharing, waste heat sharing, and cooling sharing, constructing energy interaction edges; and computing power scheduling relationships: enabling task migration, load balancing, and computing power offloading, constructing computing power scheduling edges. The global spatiotemporal model can characterize the spatial coupling relationships of multiple data centers, the cross-regional collaborative relationships of multiple energy flows, the constraints of the spatiotemporal synchronous changes of computing power and energy, and the dynamic evolution structure of the overall system over time.

[0036] The temporal characteristics reflect the time-series properties that govern the operation of a single data center, including: Time-series trends: Daily and weekly increases or decreases in computing power, electricity, and carbon emissions; Cyclical patterns: daily cycle (morning and evening peak hours), weekly cycle (weekdays and weekends), seasonal cycle; Fluctuation characteristics: load change amplitude, renewable energy generation volatility, and electricity consumption volatility; Peak and trough patterns: peak computing power periods, peak electricity consumption periods, peak waste heat generation periods, and peak carbon emission periods.

[0037] Spatial association characteristics reflect the mutual influence, constraints, and coupling between nodes, including: Spatial distance correlation: The closer the distance, the lower the cost of computing power scheduling and energy sharing; Energy transmission loss characteristics: The greater the distance, the greater the loss in the transmission of electricity and heat, and the stronger the line constraints. The coupling characteristics of computing power networks: network bandwidth and latency determine whether computing power can be migrated quickly; Regional energy endowment constraints: If a region has abundant solar power, surrounding data centers can share green electricity; The coupling and correlation of multiple indicators such as computing power, electricity, heat and carbon: The increase in computing power will lead to an increase in electricity, which in turn will lead to an increase in heat, carbon emissions and the amount of waste heat that can be recovered.

[0038] By inputting spatiotemporal joint features into a pre-trained prediction model (such as Transformer or graph prediction network), high-precision predictions can be made for future preset periods (such as the next 24 hours, the next 7 days, or the next 96 periods) on a time-by-time basis.

[0039] In this embodiment, the computing power task migration mechanism setting in step S5 includes: The data center computing power migration mechanism adopts a layered migration logic of internal migration and cross-center migration: When migrating computing power tasks in a data center, priority is given to migrating computing power tasks between servers within the corresponding data center, migrating the computing power of overloaded servers to servers in a light-load, normal-load state; after the internal computing power task migration is completed, the operating status of the data center is detected, and if the detection result meets at least one preset condition, cross-data center computing power task migration is initiated. The conditions include: there is still a gap in computing power for the task, meaning that the current available computing power of the data center cannot meet the real-time computing power demand; carbon emissions exceed the standard, meaning that the current carbon emissions of the data center exceed the preset carbon emission compliance threshold; and server operating temperature exceeds the preset reasonable range, meaning that the server operating temperature is higher than the preset temperature threshold.

[0040] In this embodiment, the power migration mechanism setting in step S5 includes: By assessing the power supply and demand status of each data center, distinguishing between data centers with surplus power and those with power shortages, calculating the power surplus or shortage for each data center, and determining the supply and demand correspondence, migration amount, and transmission links for power migration based on power transmission links, the link with the least transmission loss is prioritized for power migration.

[0041] In this embodiment, S6 includes: The system acquires current power consumption, pending computing tasks, available computing power, power supply capacity, carbon emission limits, and server operating temperatures for each data center. Combining this with resource migration mechanisms and various predicted values ​​from each data center, it performs feasibility assessments: determining whether inter-server computing task migration within a data center is feasible and whether cross-data center computing task migration needs to be initiated; determining whether each data center has surplus or shortage of power and whether it meets the conditions for cross-data center power transmission; and determining whether the recoverable waste heat, industrial heat demand, and heat transmission links of each data center meet the conditions for coordinated heat utilization. Based on the above judgment results, a set of computing power task migration strategies, a set of power transmission strategies, and a set of thermal co-utilization strategies were initially formulated. The set of computing power task migration strategies includes internal migration schemes, cross-center migration directions, migration task volume, migration priority, and constraints. The set of power transmission strategies includes supply and demand matching relationships, transmission volume, path, and loss constraints. The set of thermal co-utilization strategies includes waste heat supply and demand matching relationships, transmission volume, path, and utilization efficiency constraints. The generated strategy set is initially verified to ensure that each strategy meets the requirements of computing power supply and demand matching, power supply and demand matching, carbon emission compliance, server temperature compliance, and thermal supply and demand matching, thus forming a preliminary feasible scheduling strategy set.

[0042] In this embodiment, step S7 involves constructing a cooperative game model among the servers within the first-layer data center, including: Within a single data center, servers engage in cooperative game-playing. The game participants are all servers within the k-th data center. The local status of each server includes computing load, computing capacity, power consumption, operating temperature, and carbon emissions. The objective of cooperative game theory is to maximize global utility. The global utility function for the k-th data center is... Represented as: ; , , These are the weighting coefficients for each sub-utility; For load balancing effectiveness, it represents the load difference between servers; the smaller the difference, the higher the effectiveness. For carbon emission compliance utility, it is characterized that the lower the carbon emissions, the higher the utility. To ensure compliance with operating temperature requirements, the higher the performance of the temperature within a reasonable range, the higher the operational safety and stability. By solving the Pareto optimal solution of the cooperative game model using the Nash bargaining algorithm, the optimal computing power task migration allocation scheme for each server is obtained, expressed as: ; This refers to the amount of computing power allocated to the i-th server in the k-th data center. Let i be the individual utility of the i-th server in the k-th data center; This is the point at which negotiations break down for the i-th server in the k-th data center; This represents the total number of servers in the k-th data center.

[0043] In this embodiment, step S7 involves constructing a second-layer cross-data center master-slave game model, including: A multi-indicator comprehensive evaluation method is adopted to construct a data center evaluation index system. The evaluation indexes include data center scale, operational reliability, computing power capacity, power supply reliability, network connectivity, historical scheduling success rate, and carbon emission compliance rate. Master-slave game participant setup: The comprehensive evaluation value of each data center is obtained through weighted scoring. The data center with the highest evaluation value is selected as the game leader, and the other data centers are followers. Master-follower game sequence setting: Each follower reports its own state to the leader. The leader combines the initial scheduling strategy set and resource migration mechanism to maximize the global collaborative utility and decides on the global resource allocation plan. Each follower then adjusts its own operating strategy based on the leader's decision to maximize its own utility and iterates to the Stackelberg equilibrium, where the leader's decision is optimal and the followers have no incentive to change their own strategies. Among them, the leader's overall synergistic effect The target is represented as: ; , , , These are the sub-utility weighting coefficients; The total cost utility includes computing power migration costs, power migration costs, and heat transfer costs. For the overall resource gap utility; For the overall carbon emissions utility; To maximize the overall utility, including revenue from computing power services and revenue from electricity and heat trading; The self-utility of the s-th follower The target is represented as: ; , Weighting coefficients for local operational utility and transaction revenue utility; For local operating utility, the same as the utility of the first-level game model. ; For the benefit of transaction gains.

[0044] It should be noted that the solution to the second-layer cross-datacenter master-slave game model uses backward induction to solve for the Stackelberg equilibrium, including: Follower-optimal response: Given the leader's decisions, including the amount of computing power migration, the amount of electricity transmission, the thermal coordination scheme, and the prices when trading computing power, electricity, and heat, each follower solves its own utility maximization objective to obtain the follower-optimal response; Leader's optimal decision: The leader substitutes the followers' optimal response into his own utility function, solves the global utility maximization problem, and obtains the leader's optimal decision. Balanced solution output: Obtain computing power task migration scheme (the amount of tasks migrated from one data center to another), power transmission scheme (the amount of electricity transmitted from one data center to another), heat co-utilization scheme (the amount of heat transmitted from one data center to industrial heat demanders), and pricing strategy (the transaction price of computing power, electricity, and heat).

[0045] In the iterative interactive solution of the two-layer game model, the internal cooperative game ensures the local physical feasibility of each layer. Each data center reports its local feasible domain, resource reserves, and operating status to the leader of the master-slave game model. The master-slave game guarantees global optimality, with the two layers mutually correcting each other to achieve global collaboration. Specifically, the local optimal operating status is obtained through the cooperative game of the servers within the first-layer data center and fed back to the upper layer. This status serves as the input condition for the second-layer cross-data center master-slave game. The master-slave game outputs global scheduling and pricing instructions, which in turn constrain the first-layer game, forming an iterative interactive mechanism of inner-layer solution, status reporting, global decision-making, instruction issuance, and inner-layer re-optimization. After multiple iterations until both layers of the game reach stable equilibrium, the iterative interactive solution of the two-layer game model is completed, yielding the final computing power task migration scheme, power transmission scheme, thermal synergistic utilization scheme, and related pricing strategies.

[0046] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can also be implemented in other ways. The system embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0047] Furthermore, the functional modules in the various embodiments of this invention can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part. If the function is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0048] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A multi-data center optimization planning and scheduling method incorporating computation-electricity-heat-carbon synergy, characterized in that, It includes: S1. Data Collection and Profile Building: Collect multi-source data related to computing power, electricity, heat and carbon emissions of existing data centers in the previous year, analyze the computing power resource gap, electricity resource gap, heat recovery resource gap, carbon emissions, server load status and energy utilization efficiency of each data center in the previous year, and build a profile model of each data center. S2. Renovation and New Construction Analysis: Based on the profile model of each data center, we simultaneously acquire data on renewable energy endowment, industrial heat demand, planning budget, and available site selection around each data center. Combined with the phased evolution data of computing power tasks of each data center, we analyze the dynamic characteristics of each data center after planning various pre-set renovation schemes and new data center construction schemes for existing multi-data centers. S3. Establishment of planning and design model: Based on the dynamic characteristics of each data center, with the goal of minimizing the cost of building new data centers and transforming existing data centers, the resource gap of each data center and the carbon emissions, establish a data center planning and design optimization model, conduct quantitative analysis and comparison of various preset transformation schemes and new data center schemes, and decide on the planning and design scheme of existing multiple data centers. S4. Spatiotemporal Feature Extraction and Demand Forecasting: After the planning and design are completed, a global spatiotemporal model of the data center is established to characterize the spatial correlation structure and coordinated operation relationship of all data centers over time. Then, the spatial correlation features and temporal features of all data centers at the current time are extracted to predict the computing power demand, power demand, renewable energy power supply, cooling demand, carbon emissions and waste heat recovery of each data center in different time periods. S5. Resource Migration Mechanism Settings: When migrating computing power tasks in a data center, priority is given to migrating computing power tasks between servers within the data center. If, after the internal migration, at least one of the following conditions is met: computing power task computation still has a shortfall, carbon emissions exceed the standard, or server operating temperature exceeds the required range, then cross-data center computing power task migration is initiated to ensure that each data center meets the preset standards after migration. Additionally, a cross-data center power migration mechanism is set up to migrate surplus power to data centers with power shortages through preset transmission links. S6. Preliminary scheduling strategy formulation: Based on the current power consumption information, pending computing power task information, available computing power information, power supply capacity, carbon emission limits and server operating temperature of each data center, combined with the resource migration mechanism and various prediction values ​​of each data center, a preliminary set of computing power task migration strategy, power transmission strategy and thermal synergistic utilization strategy is formulated. S7. Two-layer game model construction: Construct a cooperative game model among servers within the first layer of the data center and a master-slave game model across the second layer of the data center. Through iterative interaction and solution of the two-layer game model, obtain the final computing power task migration scheme, power transmission scheme, heat co-utilization scheme and related pricing strategy.

2. The multi-data center optimization planning and scheduling method according to claim 1, characterized in that, S1 includes: Each data center is modeled as an independent intelligent agent, serving as an intelligent management and control unit for autonomously sensing, analyzing, and making decisions regarding the operational status of each data center. Each intelligent agent collects multi-source data related to computing power, electricity, heat and carbon emissions for each period of the previous year. For data centers with their own renewable energy generator sets, additional data are collected on renewable energy power generation, generator set operating efficiency, renewable energy power supply ratio and energy storage device charging and discharging. For data centers with waste heat recovery devices, additional data are collected on waste heat recovery amount, waste heat recovery efficiency, waste heat output and waste heat recovery device operating parameters. Each agent analyzes the computing power resource gap based on the collected multi-source data, outputting the computing power gap value, gap duration, gap peak value, and remaining amount of computing power for each time period of the previous year, forming a computing power resource gap time series curve; analyzes the power resource gap, outputting the power gap value, gap duration, remaining amount of power for each time period of the previous year, and remaining power availability; analyzes the heat recovery resource gap, outputting the heat recovery gap value, recovery efficiency, and total amount of data center waste heat that can be recovered without waste heat recovery devices, as well as its potential recovery value, for each time period of the previous year; analyzes carbon emissions, outputting the carbon emissions, exceedances, and carbon emission reductions corresponding to renewable energy generation for each time period of the previous year; analyzes server load status, outputting the average server load rate for each time period of the previous year; and analyzes energy utilization efficiency, outputting the PUE value, thermal utilization efficiency, and renewable energy utilization efficiency for each time period of the previous year. By combining the multidimensional analysis data of each intelligent agent, core operational feature labels and structural feature labels are formed for each data center profile model. The data centers are classified using feature quantization, weight allocation and cluster analysis methods, and a profile model for each data center is formed.

3. The multi-data center optimization planning and scheduling method according to claim 1, characterized in that, S2 includes: Based on the profile models of each data center, and simultaneously acquiring data on the renewable energy endowment, industrial heat demand, planning budget, and feasible site selection around each data center, and combining this with the phased evolution data of computing power tasks for each data center, multiple renovation schemes and multiple new data center construction schemes are pre-set. The renewable energy endowment data includes the exploitable amount and power generation potential time-series data of surrounding photovoltaic and wind power renewable energy sources; the phased evolution data of computing power tasks includes the growth trend and changes in computing power task types for each data center at different times within a pre-set future period; the renovation schemes include adding renewable energy generators and energy storage equipment to data centers without renewable energy generators, adding waste heat recovery devices to data centers without waste heat recovery devices, optimizing the waste heat output link and recovery efficiency for data centers with waste heat recovery devices, and upgrading and optimizing existing server clusters and energy equipment; the new construction schemes include site selection for new data centers, the scale of computing server construction, equipment deployment types, and energy equipment capacity planning. Based on each pre-set renovation and new construction plan, the operation status of each data center is simulated after the implementation of the plan, and the dynamic characteristics of each data center are analyzed, including dynamic changes in the supply and demand of computing resources, dynamic changes in the supply and demand and supply structure of power resources, dynamic changes in the utilization of heat recovery resources and heat synergy, dynamic changes in carbon emissions, dynamic changes in server load balance, dynamic changes in energy utilization efficiency, and changes in economic benefits.

4. The multi-data center optimization planning and scheduling method according to claim 1, characterized in that, S3 includes: With the objectives of minimizing the costs of building new data centers and upgrading existing data centers, as well as the resource gaps and carbon emissions of each data center, an optimization model for data center planning and design is established, which is expressed as follows: ； The total cost of new construction and renovation is n; n is the number of existing data centers that need to be renovated. The equipment procurement cost for the renovation of the i-th existing data center; The installation and construction cost for the renovation of the i-th existing data center; Let m be the maintenance cost after the i-th data center is upgraded; m is the number of newly built data centers. The site selection and land acquisition cost for the j-th newly built data center; The construction cost of the j-th newly built data center; The equipment deployment cost for the j-th newly built data center; The operating cost of the j-th newly built data center; ； To address the overall resource gap; , , These are the weighting coefficients for the resource gaps in computing power, electricity, and heat recovery, respectively; t is the time period within the planning cycle; k is the number of all data centers after renovation and new construction. This represents the computing power resource gap for the k-th data center during time period t. This represents the power resource gap for the k-th data center during time period t. This represents the heat recovery resource gap for the k-th data center during time period t. ； Carbon emissions corresponding to powering the grid for the k-th data center during time period t; This represents the carbon emissions corresponding to the operation of equipment in the k-th data center during time period t. The constraints of the planning and design optimization model include: planning budget constraints, site selection constraints, equipment deployment constraints, renewable energy utilization constraints, thermal synergistic utilization constraints, carbon emission compliance constraints, server load balancing constraints, and resource supply and demand balance constraints. The site selection constraints are adapted to the geographical conditions, energy supply conditions, and surrounding industrial heat demand distribution of the plannable site selection data. The equipment deployment constraints are adapted to the structural differences of different data centers and the equipment configuration requirements of renovation and new construction schemes. The resource supply and demand balance constraints include computing power supply and demand balance, power supply and demand balance, and thermal resource supply and demand balance. Among them, the power supply and demand balance needs to be adapted to the power supply capacity of renewable energy generator sets, and the thermal resource supply and demand balance needs to be adapted to the recovery efficiency of waste heat recovery devices and surrounding industrial heat demand. Substitute each preset renovation scheme and new construction scheme into the planning and design optimization model, solve the model, perform quantitative analysis and comparison of each scheme, and output the multi-objective optimization target achievement degree of each scheme; Based on the quantitative analysis and comparison results, and combined with the priority of multi-objective optimization, the scheme that simultaneously meets the constraints and the requirement of minimizing resource gaps, and achieves the comprehensive optimization of transformation cost, new construction cost and carbon emissions, is selected as the final planning and design scheme for the existing multi-data center.

5. The multi-data center optimization planning and scheduling method according to claim 1, characterized in that, S4 includes: After the planning and design are completed, obtain the operational and spatial data of all data centers in the current period. The spatial data includes the geographical location, physical distance, power transmission links, computing network topology, waste heat transmission links, and regional energy supply boundaries of each data center. Based on operational and spatial data, each data center is abstracted as a spatiotemporal node. By establishing the physical connections, energy transmission relationships, and computing power scheduling relationships between data centers, a global spatiotemporal model of the data center is formed, which represents the spatial relationship structure and coordinated operation relationship of all data centers that dynamically change over time. A spatiotemporal graph convolutional neural network is used to extract features from the global spatiotemporal model of the data center, outputting temporal features and spatial correlation features respectively. The features are then fused to obtain the spatiotemporal joint features of each data center. The temporal features include the time-series trends, periodic patterns, fluctuation characteristics, and peak-valley patterns of the operational data of each data center. The spatial correlation features include the spatial distance correlation between data centers, energy transmission loss characteristics, computing power network coupling characteristics, regional energy endowment constraints, and the coupling correlation of multiple indicators such as computing, electricity, heat, and carbon. The spatiotemporal joint features are input into the preset prediction model to predict the key operating parameters of each data center at different times within the preset period in the future, including: computing power demand, electricity demand, renewable energy power supply, cooling demand, carbon emissions, and waste heat recovery.

6. The multi-data center optimization planning and scheduling method according to claim 1, characterized in that, In S5, the computing power task migration mechanism is set up, including: The data center computing power migration mechanism adopts a layered migration logic of internal migration and cross-center migration: When migrating computing power tasks in a data center, priority is given to migrating computing power tasks between servers within the corresponding data center, migrating the computing power of overloaded servers to servers in a light-load, normal-load state; after the internal computing power task migration is completed, the operating status of the data center is detected, and if the detection result meets at least one preset condition, cross-data center computing power task migration is initiated. The conditions include: there is still a gap in computing power for the task, meaning that the current available computing power of the data center cannot meet the real-time computing power demand; carbon emissions exceed the standard, meaning that the current carbon emissions of the data center exceed the preset carbon emission compliance threshold; and server operating temperature exceeds the preset reasonable range, meaning that the server operating temperature is higher than the preset temperature threshold.

7. The multi-data center optimization planning and scheduling method according to claim 1, characterized in that, In S5, the power migration mechanism settings include: By assessing the power supply and demand status of each data center, distinguishing between data centers with surplus power and those with power shortages, calculating the power surplus or shortage for each data center, and determining the supply and demand correspondence, migration amount, and transmission links for power migration based on power transmission links, the link with the least transmission loss is prioritized for power migration.

8. The multi-data center optimization planning and scheduling method according to claim 1, characterized in that, S6 includes: The system acquires current power consumption, pending computing tasks, available computing power, power supply capacity, carbon emission limits, and server operating temperatures for each data center. Combining this with resource migration mechanisms and various predicted values ​​from each data center, it performs feasibility assessments: determining whether inter-server computing task migration within a data center is feasible and whether cross-data center computing task migration needs to be initiated; determining whether each data center has surplus or shortage of power and whether it meets the conditions for cross-data center power transmission; and determining whether the recoverable waste heat, industrial heat demand, and heat transmission links of each data center meet the conditions for coordinated heat utilization. Based on the above judgment results, a set of computing power task migration strategies, a set of power transmission strategies, and a set of thermal co-utilization strategies were initially formulated. The set of computing power task migration strategies includes internal migration schemes, cross-center migration directions, migration task volume, migration priority, and constraints. The set of power transmission strategies includes supply and demand matching relationships, transmission volume, path, and loss constraints. The set of thermal co-utilization strategies includes waste heat supply and demand matching relationships, transmission volume, path, and utilization efficiency constraints. The generated strategy set is initially verified to ensure that each strategy meets the requirements of computing power supply and demand matching, power supply and demand matching, carbon emission compliance, server temperature compliance, and thermal supply and demand matching, thus forming a preliminary feasible scheduling strategy set.

9. The multi-data center optimization planning and scheduling method according to claim 1, characterized in that, In S7, a cooperative game model is constructed among the servers within the first-layer data center, including: Within a single data center, servers engage in cooperative game-playing. The game participants are all servers within the k-th data center. The local status of each server includes computing load, computing capacity, power consumption, operating temperature, and carbon emissions. The objective of cooperative game theory is to maximize global utility. The global utility function for the k-th data center is... Represented as: ； , , These are the weighting coefficients for each sub-utility; For load balancing effectiveness, it represents the load difference between servers; the smaller the difference, the higher the effectiveness. For carbon emission compliance utility, it is characterized that the lower the carbon emissions, the higher the utility. To ensure compliance with operating temperature requirements, the higher the performance of the temperature within a reasonable range, the higher the operational safety and stability. By solving the Pareto optimal solution of the cooperative game model using the Nash bargaining algorithm, the optimal computing power task migration allocation scheme for each server is obtained, expressed as: ; This refers to the amount of computing power allocated to the i-th server in the k-th data center. Let i be the individual utility of the i-th server in the k-th data center; This is the point at which negotiations break down for the i-th server in the k-th data center; This represents the total number of servers in the k-th data center.

10. The multi-data center optimization planning and scheduling method according to claim 1, characterized in that, In S7, a second-layer cross-data center master-slave game model is constructed, including: A multi-index comprehensive evaluation method is adopted to construct a data center evaluation index system. The evaluation indexes include data center scale, operational reliability, computing power capacity, power supply reliability, network connectivity, historical scheduling success rate, and carbon emission compliance rate. Master-slave game participant setup: The comprehensive evaluation value of each data center is obtained through weighted scoring. The data center with the highest evaluation value is selected as the game leader, and the other data centers are the followers. Master-follower game sequence setting: Each follower reports its own state to the leader. The leader combines the initial scheduling strategy set and resource migration mechanism to maximize the global collaborative utility and decides on the global resource allocation plan. Each follower then adjusts its own operating strategy based on the leader's decision to maximize its own utility and iterates to the Stackelberg equilibrium, where the leader's decision is optimal and the followers have no incentive to change their own strategies. Among them, the leader's overall synergistic effect The target is represented as: ； , , , These are the sub-utility weighting coefficients; The total cost utility includes computing power migration costs, power migration costs, and heat transmission costs. For the overall resource gap utility; For the overall carbon emissions utility; To maximize the overall utility, including revenue from computing power services and revenue from electricity and heat trading; The self-utility of the s-th follower The target is represented as: ； , Weighting coefficients for local operational utility and transaction revenue utility; For local operating utility, the same as the utility of the first-level game model. ; For the benefit of transaction gains.

Images