A computing power center multi-agent energy consumption collaboration method and system

Abstract

This application relates to the field of intelligent agent energy consumption coordination technology, and discloses a multi-agent energy consumption coordination method and system for a computing center, including: simulating the computing power working state and intelligent agent energy consumption state of the computing center based on pre-acquired computing center data; analyzing the virtual energy bus reflecting the computing power allocation and the energy efficiency credit reflecting the energy efficiency of the intelligent agents to construct a virtual computing power allocation architecture; configuring a coordination strategy based on the real-time energy consumption data of each intelligent agent to analyze the ratio of energy consumption to performance of each intelligent agent to obtain the corresponding dynamic weight; simulating the dynamic weight and the bidirectional bidding coordination game process between the computing intelligent agent and the cooling intelligent agent in the virtual computing power allocation architecture to obtain a first coordination strategy; generating intelligent agent energy consumption control instructions according to the first coordination strategy to control each intelligent agent to execute corresponding energy consumption control and energy efficiency improvement actions; this application can reduce resource idleness and over-configuration, improve the overall energy efficiency of the computing center, and reduce total energy consumption.

Description

Technical Field

[0001] This application relates to the field of intelligent agent energy consumption coordination technology, and more specifically to a method and system for multi-agent energy consumption coordination in a computing center. Background Technology

[0002] With the rapid development of cloud computing, big data, and artificial intelligence, the scale and energy consumption of computing centers have increased dramatically. Computing centers typically contain multiple agents, such as computing agents and cooling agents. Computing agents are responsible for computational tasks, while cooling agents are responsible for heat dissipation. The energy consumption of these agents accounts for a large proportion of the total energy consumption of the computing center. The energy coordination process in computing centers is costly and wasteful. Computing agents and cooling agents often employ static strategies or independent control, lacking collaborative control and failing to dynamically adapt to real-time load changes and energy efficiency requirements, leading to problems such as low energy efficiency and uneven resource allocation.

[0003] Existing technologies suffer from the following problems: Optimization is performed separately for computing agents or cooling agents, ignoring the coupling relationship between the two, resulting in lag in the response of cooling agents and wasted computing power; when formulating energy consumption coordination strategies, they are based on historical data or fixed thresholds, which cannot adapt to dynamic workloads in real time, resulting in large fluctuations in the ratio of energy consumption to performance and low efficiency; the allocation of computing power and cooling resources is disconnected, which can easily lead to local overheating or overcooling, increasing additional energy consumption; to solve at least one of the above problems, this application proposes a multi-agent energy consumption coordination method and system for computing power centers. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this application is to provide a multi-agent energy consumption coordination method and system for computing power centers, which can effectively solve the problems in the background technology. The specific technical solution of this application is as follows:

[0005] A method for coordinated energy consumption among multiple intelligent agents in a computing center includes:

[0006] Based on the pre-acquired computing center data, the computing power working status and intelligent agent energy consumption status of the computing center are simulated. The virtual energy bus reflecting the computing power allocation and the energy efficiency credit reflecting the energy efficiency of the intelligent agent are analyzed to construct a virtual computing power allocation architecture.

[0007] Based on the real-time energy consumption data of each agent, a collaborative strategy is configured to analyze the ratio of energy consumption to performance of each agent to obtain the corresponding dynamic weight. The agents include computing agents and cooling agents.

[0008] In the virtual computing power allocation architecture, dynamic weights and a collaborative game process of bidirectional bidding between computing agents and cooling agents are simulated to obtain the first collaborative strategy;

[0009] According to the first collaborative strategy, intelligent agent energy consumption control instructions are generated to control each intelligent agent to perform corresponding energy consumption control actions and energy efficiency improvement actions, so as to coordinate the energy consumption of multiple intelligent agents in the computing center.

[0010] Specifically, based on pre-acquired computing center data, the process simulates the computing power operation status of the computing center and the energy consumption status of intelligent agents, analyzes the virtual energy bus reflecting the computing power allocation and the energy efficiency credit reflecting the energy efficiency of intelligent agents, and constructs a virtual computing power allocation architecture, including:

[0011] Based on the pre-acquired computing center data, the computing power working status and intelligent agent energy consumption status of the computing center are simulated, the coupling between load and cooling demand is analyzed, and a virtual energy bus reflecting the computing power allocation is constructed.

[0012] Analyze the historical energy efficiency benchmark value and real-time operating energy efficiency deviation of each intelligent agent, and compare the difference between the real-time operating energy efficiency deviation and the benchmark using the average value of the historical energy efficiency benchmark value as the benchmark to obtain the corresponding energy efficiency credit;

[0013] By combining the virtual energy bus and energy efficiency credits, a virtual computing power allocation architecture is constructed.

[0014] Specifically, the step of simulating the computing power operation status and intelligent agent energy consumption status of the computing power center based on pre-acquired computing power center data, analyzing the coupling between load and cooling demand, and constructing a virtual energy bus reflecting the computing power allocation includes:

[0015] Based on the pre-acquired computing center data, simulate the computing power working status of the computing center and the energy consumption status of the intelligent agent, analyze and calculate the real-time computing power consumption change curve when the intelligent agent performs computing tasks, and the corresponding cooling demand distribution map during the operation of the cooling intelligent agent.

[0016] Based on the real-time computing power consumption change curve, the computing power demand is analyzed according to the preset priority of computing tasks, and a computing power allocation layer is constructed.

[0017] The load aggregation characteristics are analyzed in the cooling demand distribution map, cooling zones are dynamically divided, corresponding cooling resources are configured for each cooling zone, and a cooling layer is constructed.

[0018] Based on the positional relationship between the computation agent and the cooling agent, analyze the distance, path accessibility and cooling efficiency between the computation agent and the nearest cooling agent, and calculate the corresponding positional coupling factor.

[0019] By using the location coupling factor, a mapping relationship is established between the computing power allocation layer and the cooling layer, thus constructing a virtual energy bus consisting of a dual-layer structure of computing power allocation layer and cooling layer.

[0020] Specifically, combining the virtual energy bus and energy efficiency credits, a virtual computing power allocation architecture is constructed, including:

[0021] In the virtual energy bus, analyze the computing power demand of the computing power allocation layer and the cooling resources of the cooling layer to calculate the computing power cooling situation and calculate the cooling efficiency.

[0022] The computing power requirement of the computing power allocation layer is calculated based on the energy efficiency credit analysis, and the efficiency value of the cooling resources provided by the cooling agent is calculated. The circulation efficiency is then obtained by combining the cooling efficiency.

[0023] Based on the circulation efficiency, an initial energy efficiency credit and dynamic credit limit are configured for each intelligent agent corresponding to the virtual energy bus through a preset credit allocation model, thereby constructing a virtual computing power allocation architecture.

[0024] Specifically, the step of configuring a collaborative strategy based on the real-time energy consumption data of each agent to analyze the ratio of energy consumption to performance of each agent and obtain the corresponding dynamic weight includes:

[0025] Based on the real-time energy consumption data of each agent, the real-time computing task requirements of the agent are analyzed and mapped to the corresponding computing power requirements to obtain the first weight of each computing agent.

[0026] The correlation between cooling efficiency and energy consumption of the cooling agent is analyzed, and the second weight of each cooling agent is obtained by combining the correlation with the real-time energy consumption data of the cooling agent.

[0027] Based on the coupling relationship between the computing agent and the cooling agent in the virtual computing power allocation architecture, the energy consumption to performance ratio of each agent is analyzed, and the first weight and the second weight are dynamically adjusted to obtain the corresponding dynamic weight.

[0028] Specifically, based on the coupling relationship between the computing agent and the cooling agent in the virtual computing power allocation architecture, the energy consumption to performance ratio of each agent is analyzed, and the first and second weights are dynamically adjusted to obtain the corresponding dynamic weights, including:

[0029] The real-time resource matching degree between the computing power allocation layer and the cooling layer in the virtual computing power allocation architecture is analyzed, and the performance coupling coefficient of the computing agent and the cooling agent is obtained through a preset coupling analysis model.

[0030] For agents whose efficiency coupling coefficient is greater than a preset coupling threshold, the weights are amplified by combining the corresponding efficiency coupling coefficient weights and the ratio of energy consumption to performance of each agent to obtain the first dynamic weights.

[0031] For agents whose performance coupling coefficient is less than or equal to a preset coupling threshold, the weights are reduced by analyzing the independent operation characteristics of each agent to obtain a second dynamic weight.

[0032] By combining the first dynamic weight and the second dynamic weight, the weights are calibrated and optimized using a preset weight calibration model to obtain the dynamic weights corresponding to each agent.

[0033] Specifically, in the virtual computing power allocation architecture, a collaborative game process involving dynamic weights and bidirectional bidding between computing agents and cooling agents is simulated to obtain a first collaborative strategy, including:

[0034] Based on the dynamic weights corresponding to each agent, the bidding parameters of the computing agent and the cooling agent are determined through a preset bidding analysis model.

[0035] Based on the bidding parameters, a collaborative game process of two-way bidding between computing agents and cooling agents is simulated in the virtual computing power allocation architecture to obtain the first collaborative strategy.

[0036] Specifically, according to the bidding parameters, a collaborative game process of two-way bidding between computing agents and cooling agents is simulated in the virtual computing power allocation architecture to obtain a first collaborative strategy, including:

[0037] Based on the bidding parameters, analyze the computing task requirements and performance requirements of the computing agent in the virtual computing power allocation architecture to determine the initial bidding scheme;

[0038] Analyze the real-time resource distribution of the virtual computing power allocation architecture and determine the resource benchmark through a preset benchmark analysis model;

[0039] Based on the initial bidding scheme and resource benchmark, resource allocation simulation is performed, and a two-way bidding game is conducted on the energy efficiency of the resources allocated by the computing agent and the cooling agent to obtain the first collaborative strategy.

[0040] Specifically, according to the first collaborative strategy, intelligent agent energy consumption control instructions are generated to control each intelligent agent to execute corresponding energy consumption control actions and energy efficiency improvement actions, including:

[0041] The first collaborative strategy is decomposed into energy consumption budget instructions for computational agents and cooling control instructions for cooling agents.

[0042] According to the energy consumption budget instructions, each computing agent is controlled to perform corresponding energy consumption control actions;

[0043] The cooling control command is used to control each cooling agent to perform corresponding energy efficiency improvement actions.

[0044] A multi-agent energy consumption coordination system for computing power centers, used to implement the aforementioned multi-agent energy consumption coordination method for computing power centers, includes:

[0045] The virtual computing power allocation architecture construction module simulates the computing power working status of the computing power center and the energy consumption status of intelligent agents based on the pre-acquired computing power center data, analyzes the virtual energy bus reflecting the computing power allocation and the energy efficiency credit reflecting the energy efficiency of intelligent agents, and constructs the virtual computing power allocation architecture.

[0046] The dynamic weight allocation module, based on the real-time energy consumption data of each agent, configures a collaborative strategy to analyze the ratio of energy consumption to performance of each agent to obtain the corresponding dynamic weight. The agents include computing agents and cooling agents.

[0047] The multi-agent collaborative strategy formulation module simulates dynamic weights and the collaborative game process of bidirectional bidding between computing agents and cooling agents in the virtual computing power allocation architecture to obtain the first collaborative strategy.

[0048] The multi-agent collaborative control module generates energy consumption control instructions for each agent according to the first collaborative strategy, and controls each agent to perform corresponding energy consumption control actions and energy efficiency improvement actions to coordinate the energy consumption of the multi-agents in the computing center.

[0049] The beneficial effects of this application are as follows: Based on dynamic mapping of computing power allocation and cooling requirements, a two-layer structure of computing and cooling layers is established through a location coupling factor, enabling precise resource matching; combining historical energy efficiency benchmarks and real-time deviation to evaluate the energy efficiency of intelligent agents, and analyzing the energy consumption to performance ratio of each intelligent agent based on real-time energy consumption data to generate dynamic weights, which can adaptively adjust the weights according to the working state; adopting a two-way bidding cooperative game model to simulate the interaction between computing and cooling intelligent agents, generating cooperative strategies through bidding parameters and resource benchmarks, and coordinating the control of multiple intelligent agents. Virtual energy bus and energy efficiency credits enable efficient matching of computing power and cooling resources, reducing resource idleness and over-configuration; the cooperative strategy, combined with the coupling relationship between multiple intelligent agents, can improve the operational stability of the computing center under fluctuating loads; through dynamic weights and cooperative game, optimizing the energy consumption to performance ratio can improve the overall energy efficiency of the computing center and reduce total energy consumption. Attached Figure Description

[0050] Figure 1 This is a flowchart illustrating a multi-agent energy consumption coordination method for computing centers, as described in an embodiment of this application.

[0051] Figure 2 This is a schematic diagram of the virtual energy bus in the embodiments of this application;

[0052] Figure 3This is a flowchart illustrating the dynamic weight calculation process in the embodiments of this application;

[0053] Figure 4 This is a schematic diagram of the structure of a multi-agent energy consumption collaborative system for computing centers, as described in an embodiment of this application. Detailed Implementation

[0054] The present application will be further described in detail below with reference to the accompanying drawings and embodiments.

[0055] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0056] Hereinafter, the terms "first," "second," and other generic terms are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0057] refer to Figure 1 The image shows a specific implementation of a multi-agent energy consumption coordination method for computing centers according to this application, including:

[0058] S101. Based on the pre-acquired computing center data, simulate the computing power working status and intelligent agent energy consumption status of the computing center, analyze the virtual energy bus reflecting the computing power allocation and the energy efficiency credit reflecting the energy efficiency of the intelligent agent, and construct a virtual computing power allocation architecture.

[0059] S102. Based on the real-time energy consumption data of each agent, configure a collaborative strategy to analyze the ratio of energy consumption to performance of each agent to obtain the corresponding dynamic weight. The agents include computing agents and cooling agents.

[0060] S103. Simulate the dynamic weights and the collaborative game process of bidirectional bidding between the computing agent and the cooling agent in the virtual computing power allocation architecture to obtain the first collaborative strategy.

[0061] S104. Generate intelligent agent energy consumption control instructions according to the first collaborative strategy, and control each intelligent agent to execute corresponding energy consumption control actions and energy efficiency improvement actions, so as to coordinate the energy consumption of multiple intelligent agents in the computing center.

[0062] Currently, energy consumption in computing centers is becoming increasingly prominent, and traditional methods cannot meet dynamic load and energy efficiency requirements. This embodiment, based on pre-acquired computing center data (including but not limited to server utilization, memory usage, rack inlet temperature, and cooling system power consumption), simulates the dynamic computing power operation status and agent energy consumption status of the computing center. It analyzes and constructs a virtual energy bus reflecting computing power allocation, defines energy efficiency credits for each agent reflecting energy efficiency, and builds a virtual computing power allocation architecture. By constructing this virtual computing power allocation architecture, the association and coupling of the computing and cooling processes within the computing center are achieved, providing a precise data foundation and operational architecture for subsequent collaborative optimization. Unlike traditional static and fixed resource management methods, this embodiment can achieve global dynamic optimization, improve the efficiency and accuracy of multi-agent collaborative control, enhance energy efficiency, and reduce energy consumption.

[0063] Specifically, based on the real-time energy consumption data of each agent, the energy consumption data of the agent is calculated, including but not limited to throughput and task processing speed. The energy consumption data for cooling agents includes but is not limited to the heat removed per unit of energy consumption. The collaborative strategy is configured to analyze the ratio of energy consumption to performance of each agent, which reflects the current energy efficiency level of the agent. Combined with the type of agent and its coupling relationship in the virtual energy bus, the corresponding dynamic weight is calculated. Through the dynamic weight mechanism, the changes in the computing power status within the computing center can be accurately responded to. If the energy efficiency of a computing agent is temporarily reduced due to processing critical tasks, its weight can be dynamically increased to ensure task completion, while guiding cooling resources to be tilted towards it to prevent overheating. For an agent with extremely low energy efficiency, its weight will be reduced to avoid resource waste. By dynamically adjusting the weight, it can be ensured that when the system optimizes resources, it always approaches the goal of global optimal energy efficiency and guarantees system performance, achieving precise resource regulation and improving the adaptability and effectiveness of multi-agent energy consumption collaborative control.

[0064] Furthermore, the virtual computing power allocation architecture simulates dynamic weights and a collaborative game process involving two-way bidding between computing agents and cooling agents. The computing agent, as the demand side, bids for computing resources and necessary cooling guarantees from the virtual energy bus based on its computing task requirements and performance goals. Its bid price is related to its dynamic weight and task urgency. The cooling agent, as the supplier, bids to provide cooling services based on its current cooling capacity, efficiency, and energy consumption costs. The optimal solution is determined through this two-way bidding collaborative game process, resulting in the first collaborative strategy. This two-way bidding collaborative game process effectively coordinates the conflicting objectives between computing agents and cooling agents, identifies the optimal point for energy consumption collaboration during the game, improves the fairness and efficiency of the game process, and enhances overall energy efficiency.

[0065] Specifically, energy consumption control instructions for intelligent agents are generated according to the first collaborative strategy. For computing agents, energy consumption budget instructions are generated, and for cooling agents, cooling regulation instructions are generated. Each agent is controlled to execute corresponding energy consumption control actions and energy efficiency improvement actions to achieve collaborative control of the energy consumption of multiple agents in the computing center. By executing precise control instructions generated by the optimal strategy produced by collaborative game theory, various agents within the computing center can perform highly coordinated operations, guiding the computing load to areas with higher energy efficiency. At the same time, cooling resources are precisely distributed to heat sources on demand, avoiding over-cooling or under-cooling of the computing power, saving energy and improving the service reliability and stability of the computing center.

[0066] This application, based on dynamic mapping of computing power allocation and cooling requirements, establishes a two-layer structure of computing and cooling layers through location coupling factors, enabling precise resource matching. It assesses agent energy efficiency by combining historical energy efficiency benchmarks and real-time deviation, and analyzes the energy consumption-to-performance ratio of each agent based on real-time energy consumption data to generate dynamic weights that can adaptively adjust according to the working state. A two-way bidding cooperative game model is employed to simulate the interaction between computing and cooling agents, generating cooperative strategies through bidding parameters and resource benchmarks, and coordinating the control of multiple agents. Virtual energy buses and energy efficiency credits enable efficient matching of computing power and cooling resources, reducing resource idleness and over-configuration. The cooperative strategy, combined with the coupling relationship between multiple agents, improves the operational stability of the computing center under fluctuating loads. Through dynamic weights and cooperative game theory, the energy consumption-to-performance ratio is optimized, improving the overall energy efficiency of the computing center and reducing total energy consumption.

[0067] Furthermore, based on the pre-acquired computing center data, the computing power operation status of the computing center and the energy consumption status of the intelligent agent are simulated. The virtual energy bus reflecting the computing power allocation and the energy efficiency credit reflecting the energy efficiency of the intelligent agent are analyzed to construct a virtual computing power allocation architecture, including:

[0068] S201. Based on the pre-acquired computing center data, simulate the computing power working status and intelligent agent energy consumption status of the computing center, analyze the coupling between load and cooling demand, and construct a virtual energy bus reflecting the computing power allocation.

[0069] S202. Analyze the historical energy efficiency benchmark value and real-time operating energy efficiency deviation of each intelligent agent. Using the average value of the historical energy efficiency benchmark value as the benchmark, compare the difference between the real-time operating energy efficiency deviation and the benchmark to obtain the corresponding energy efficiency credit.

[0070] S203. Combining the virtual energy bus and energy efficiency credit, construct a virtual computing power allocation architecture.

[0071] In this embodiment, based on pre-acquired computing center data, the computing power operation status and intelligent agent energy consumption status of the computing center are simulated to obtain the real-time computing power consumption change curve of the computing agent and the cooling demand distribution map of the cooling agent. The coupling between load and cooling demand is analyzed, and a virtual energy bus reflecting the computing power allocation is constructed. By constructing the virtual energy bus, the correlation coupling between computing agents and cooling agents within the computing center is realized, thereby determining the correspondence between computing power growth and cooling demand. This provides an accurate mapping architecture for subsequent linkage and coordinated control of computing agents and cooling agents, improving the predictability and accuracy of computing power resource allocation.

[0072] Specifically, for each agent, based on pre-acquired historical energy efficiency data, including but not limited to calculating the agent's performance-to-power ratio and the cooling capacity-to-power ratio for cooling the agent, statistical analysis is performed on each agent's historical energy efficiency benchmark value and real-time operational energy efficiency deviation. The real-time operational energy efficiency deviation is the relative difference between the real-time energy efficiency value and its own historical energy efficiency benchmark value. Using the average of the historical energy efficiency benchmark values ​​as a benchmark, the difference between the real-time operational energy efficiency deviation and the benchmark is compared to obtain the corresponding energy efficiency credit. By calculating the agent's energy efficiency credit, the progress or regression of each agent's performance relative to historical norms can be analyzed, providing accurate data support for subsequent allocation of computing resources and game theory, enabling computing resources to be prioritized for agents with better energy efficiency performance, thereby improving overall energy efficiency.

[0073] Specifically, by combining a virtual energy bus and energy efficiency credits, and using the virtual energy bus as the physical architecture, the energy efficiency credits of each agent are defined into the corresponding structure to construct a virtual computing power allocation architecture. By combining the virtual energy bus and energy efficiency credits to construct this virtual computing power allocation architecture, when scheduling computing resources, the virtual energy bus ensures matching computing power with cooling capacity, and the energy efficiency credits guide resource flow to high-efficiency units. This ensures that the resulting collaborative strategy conforms to both physical laws and energy efficiency optimization goals, providing a corresponding efficient operating environment for subsequent collaborative game processes.

[0074] Furthermore, based on the pre-acquired computing center data, the computing power operation status and intelligent agent energy consumption status of the computing center are simulated, the coupling between load and cooling demand is analyzed, and a virtual energy bus reflecting the computing power allocation is constructed, including:

[0075] S301. Based on the pre-acquired computing center data, simulate the computing power working state of the computing center and the energy consumption state of the intelligent agent, analyze and calculate the real-time computing power consumption change curve when the intelligent agent performs computing tasks and the corresponding cooling demand distribution map during the operation of the cooling intelligent agent.

[0076] S302. Based on the real-time computing power consumption change curve, analyze the computing power demand according to the preset priority of computing tasks, and construct a computing power allocation layer;

[0077] S303. Analyze the load aggregation characteristics in the cooling demand distribution map, dynamically divide the cooling areas, configure corresponding cooling resources for each cooling area, and construct a cooling layer.

[0078] S304. Based on the positional relationship between the computational agent and the cooling agent, analyze the distance, path smoothness, and cooling efficiency between the computational agent and the nearest cooling agent, and calculate the corresponding positional coupling factor.

[0079] S305. By using the location coupling factor, a mapping relationship is established between the computing power allocation layer and the cooling layer to construct a virtual energy bus consisting of a dual-layer structure of computing power allocation layer and cooling layer.

[0080] In this embodiment, based on pre-acquired computing center data, the computing power operation status of the computing center and the energy consumption status of the intelligent agent are simulated. The system analyzes the real-time computing power consumption change curve when the intelligent agent performs computing tasks and the corresponding cooling demand distribution map during the operation of the cooling intelligent agent. The system acquires computing center data such as server utilization and power consumption of the computing intelligent agent and water flow and chiller power consumption of the cooling intelligent agent through integrated infrastructure management software and deployed sensor network. Through time series data analysis methods, discrete power consumption sampling points are fitted into a continuous real-time computing power consumption change curve, reflecting the power consumption value and the trend and fluctuation frequency of power consumption over time. Through spatial interpolation algorithm, temperature data is combined with cooling equipment exhaust parameters to reflect the cooling demand intensity and distribution hotspots in different areas, generating a corresponding cooling demand distribution map.

[0081] It should be noted that by analyzing the real-time computing power and power consumption change curves of computing agents when performing computing tasks, the trend of computing load changes can be predicted. By analyzing the cooling demand distribution map corresponding to the operation of cooling agents, the heat load accumulation area can be quickly and accurately located, providing a data foundation for subsequent matching of computing power and cooling, and improving the accuracy and effectiveness of the allocation process of computing power and cooling resources.

[0082] Specifically, based on real-time computing power and power consumption curves, computing power requirements are analyzed according to the preset priorities of computing tasks. The computing resources consumed by high-priority tasks are marked as core requirements that need to be prioritized, and the corresponding power consumption fluctuations are given higher attention weight. The computing power requirement information marked with priority is integrated into the corresponding logical plane to construct a computing power allocation layer. By constructing a computing power allocation layer, the computing power supply and cooling support for critical tasks can be prioritized in resource coordination and game theory, ensuring the continuity of computing tasks even with limited resources, reducing task lag, improving the running quality of computing tasks, and enhancing the overall efficiency of the computing center.

[0083] Furthermore, the load aggregation characteristics are analyzed in the cooling demand distribution map, and the load aggregation characteristics are clustered. Cooling zones are dynamically divided according to the degree of heat load aggregation, and corresponding cooling resources are configured for each cooling zone to construct a cooling layer. By constructing a dynamically adaptive cooling layer, it is possible to adapt to changes in heat load, prevent cooling blind spots or resource waste under fixed zone division, configure dedicated cooling resources for each zone, accurately execute corresponding cooling control commands, improve resource cooling efficiency, and avoid overcooling.

[0084] Specifically, based on the positional relationship between the computational agent and the cooling agent, the distance between each computational agent and its physically nearest cooling agent is calculated; by analyzing whether there are obstructions such as cabinet blockages or cable obstructions in the airflow path between them, the corresponding path unobstructedness is calculated; the temperature reduction value per unit energy consumption is analyzed, and the cooling efficiency of the cooling agent on the computational agent is calculated; the calculated distance, path unobstructedness, and cooling efficiency are weighted and multiplied to obtain the corresponding position coupling factor. The higher the value of the position coupling factor, the tighter the coupling relationship between the corresponding computational agent and the cooling agent, and the more effective the cooling response.

[0085] It is important to emphasize that by calculating the location coupling factor, the performance of the computing power resource cooling process can be analyzed, providing accurate structural support for subsequent resource mapping and strategy formulation in the virtual energy bus, ensuring that the generated collaborative strategy conforms to the physical structure and is executable, and improving the effectiveness of the strategy.

[0086] like Figure 2As shown in the diagram, the numbers reflect the coupling relationships between corresponding intelligent agents. Using the location coupling factor as the key, a weighted mapping relationship is established between each computing demand unit in the computing power allocation layer and the cooling resource unit responsible for that area in the cooling layer. For a computing node containing a high-priority computing task, if the location coupling factor is high, the weight of the corresponding mapping link connected to the cooling resource is also set high, indicating that resource flow on this link should be prioritized. The weighted mapping relationships are integrated to construct a virtual energy bus consisting of a two-layer structure: the computing power allocation layer and the cooling layer. By constructing the virtual energy bus, the coupling relationship between computing agents and cooling resources can be reflected and processed in real time. During computing power resource scheduling, the energy consumption of the cooling process can be analyzed, and the dynamic demand for computing power can be accurately responded to during the allocation of cooling resources. Through collaborative mapping, the mapping between the computing process and the cooling process can be correlated, optimizing the operation of computing resources and the energy efficiency of cooling resources during the cooling process, thereby improving resource utilization.

[0087] Furthermore, combining the aforementioned virtual energy bus and energy efficiency credits, a virtual computing power allocation architecture is constructed, including:

[0088] S401. Analyze the computing power demand of the computing power allocation layer and the cooling resources of the cooling layer in the virtual energy bus, and calculate the cooling efficiency.

[0089] S402. Calculate the computing power requirement of the computing power allocation layer and the efficiency value of the cooling resources provided by the cooling agent based on the energy efficiency credit analysis, and calculate the circulation efficiency in combination with the cooling efficiency.

[0090] S403. Based on the circulation efficiency, configure initial energy efficiency credit and dynamic credit limit for each intelligent agent corresponding to the virtual energy bus through a preset credit allocation model, and construct a virtual computing power allocation architecture.

[0091] In this embodiment, within the virtual energy bus, the total power consumption of the computing agents in the region is obtained from the computing power allocation layer, and the total energy consumption of the mapped cooling agents to eliminate this heat load is obtained from the cooling layer. The computing power cooling situation between the computing power demand of the computing power allocation layer and the cooling resources of the cooling layer is analyzed. The degree of matching between the actual average temperature of the region and the target set temperature is calculated to obtain the cooling efficiency, which is expressed as the ratio of the calculated equipment heat load to the cooling energy consumed to eliminate this load. By calculating the cooling efficiency, efficient mapping relationships of the virtual energy bus can be identified, providing accurate data support for the computing power resource allocation and cooling resource adjustment process, and improving the efficiency and effectiveness of computing power resource scheduling and cooling resource matching.

[0092] Specifically, based on energy efficiency credit analysis, the computing power requirements of the computing power allocation layer are calculated by the computing agents, and the efficiency value of the cooling resources provided by the cooling agents is also considered. Based on the energy efficiency credit calculated for each agent, the energy efficiency credit of the computing agent and the energy efficiency credit of the cooling agents providing cooling services are extracted. These two credit values ​​are summed and averaged to obtain a comprehensive credit base. The calculated cooling efficiency is multiplied by the comprehensive credit base to calculate the circulation efficiency. By calculating the circulation efficiency, computing power resources can be allocated in conjunction with the accumulated energy efficiency credit values ​​of the agents. In the resource allocation process, not only the current energy efficiency credit value of the agents is considered, but also the historically accumulated energy efficiency credit values ​​are combined to select the better agents, thereby improving the effectiveness of resource allocation and enhancing overall energy efficiency.

[0093] Specifically, based on circulation efficiency, a virtual computing power allocation architecture is constructed by configuring initial energy efficiency credits and dynamic credit limits for each intelligent agent corresponding to the virtual energy bus through a pre-defined credit allocation model. The credit allocation model includes, but is not limited to, a neural network model pre-trained using a large amount of historical intelligent agent resource allocation data. The calculated circulation efficiency is input into the pre-trained neural network model, and the model analyzes historical data on the circulation efficiency of resource circulation paths to allocate corresponding initial energy efficiency credits and dynamic credit limits to each intelligent agent, thus constructing the virtual computing power allocation architecture. This architecture reflects the location of computing power resources and the flow mechanism of resources based on energy efficiency. By combining circulation efficiency and energy efficiency credits, resource allocation and energy efficiency value are integrated, enabling accurate allocation of computing power resources according to corresponding energy efficiency values, thereby improving the overall energy efficiency of allocating computing power resources to corresponding intelligent agents for executing computing tasks.

[0094] Furthermore, based on the real-time energy consumption data of each agent, a collaborative strategy is configured to analyze the energy consumption-performance ratio of each agent to obtain the corresponding dynamic weights, including:

[0095] S501. Based on the real-time energy consumption data of each agent, analyze and calculate the real-time computing task requirements of the agent, map them to the corresponding computing power requirements, and obtain the first weight of each computing agent.

[0096] S502. Analyze the correlation between the cooling efficiency and energy consumption of the cooling agent, and combine the correlation with the real-time energy consumption data of the cooling agent to obtain the second weight of each cooling agent.

[0097] S503. Based on the coupling relationship between the computing agent and the cooling agent in the virtual computing power allocation architecture, analyze the ratio of energy consumption to performance of each agent, and dynamically adjust the first weight and the second weight to obtain the corresponding dynamic weight.

[0098] In this embodiment, based on the real-time energy consumption data of each agent, including but not limited to server utilization, cache hit rate, and memory bandwidth usage, the real-time computing task requirements of the computing agents are analyzed. The characteristics of the computing tasks currently being executed by each computing agent are identified, including but not limited to high-performance computing, AI training, and web services. Tasks with different characteristics have different types and intensities of computing resource requirements. By analyzing historical task requirements, real-time computing task requirements are mapped to corresponding computing power requirements. Using a pre-trained linear regression model based on a large amount of computing power requirement data, a first weight is calculated for each computing agent. This first weight reflects the intensity of resource requirements generated by the computing agent's computing tasks at the current moment and its importance in global energy efficiency optimization. By calculating the first weight, the differences in computing power requirements among different computing tasks can be differentiated and quantified. When formulating collaborative strategies, priority can be given to core tasks that significantly impact the overall performance of the computing center, avoiding performance fluctuations caused by fixed energy consumption control and improving the operational efficiency of computing tasks.

[0099] Furthermore, a pre-defined second-weight calculation model is used to analyze the correlation between cooling efficiency and energy consumption of the cooling agents. Combining this correlation with real-time energy consumption data of the cooling agents, a second weight is obtained for each cooling agent. This second-weight calculation model includes, but is not limited to, pre-trained nonlinear models using a large amount of historical cooling efficiency and energy consumption data. The model combines the correlation function and real-time energy consumption data to weight and correct cooling efficiency. Cooling agents operating within a high-efficiency range with controllable energy consumption are assigned higher second weights; those with large absolute cooling capacity but extremely high energy consumption and low overall efficiency are assigned lower second weights. The model outputs the calculated second weight for each cooling agent. By calculating these second weights, the energy efficiency of the cooling agents can be accurately analyzed. This allows for the identification and priority utilization of high-efficiency cooling resources that can meet cooling needs with lower energy consumption. This effectively avoids energy waste caused by blindly relying on high-power cooling equipment in collaborative control, improving resource cooling efficiency and overall energy efficiency.

[0100] Specifically, based on the coupling relationship between the computing agent and the cooling agent in the virtual computing power allocation architecture, the energy consumption to performance ratio of each agent is analyzed, and the first and second weights are dynamically adjusted to obtain the corresponding dynamic weights. Through the dynamic adjustment of the weights, combined with the mutual coupling relationship between the computing agent and the cooling agent, the dynamically adjusted weights provide accurate data support for the collaborative game process, and combine them to obtain an agent combination that can efficiently link up, thereby improving the intelligence of the collaborative control strategy and the effectiveness of the control effect.

[0101] like Figure 3As shown, based on the coupling relationship between the computing agent and the cooling agent in the virtual computing power allocation architecture, the energy consumption to performance ratio of each agent is analyzed, and the first and second weights are dynamically adjusted to obtain the corresponding dynamic weights, including:

[0102] S601. Analyze the real-time resource matching degree between the computing power allocation layer and the cooling layer in the virtual computing power allocation architecture, and obtain the performance coupling coefficient of the computing agent and the cooling agent through the preset coupling analysis model.

[0103] S602. For agents whose efficiency coupling coefficient is greater than a preset coupling threshold, the weights are amplified by combining the corresponding efficiency coupling coefficient weights and the ratio of energy consumption to performance of each agent to obtain the first dynamic weight.

[0104] S603. For agents whose performance coupling coefficient is less than or equal to a preset coupling threshold, analyze the independent operation characteristics of each agent to reduce the weights and obtain a second dynamic weight.

[0105] S604. Combining the first dynamic weight and the second dynamic weight, the weights are calibrated and optimized using a preset weight calibration model to obtain the dynamic weights corresponding to each agent.

[0106] In this embodiment, within the virtual computing power allocation architecture, the resource requirements of each computing unit in the computing power allocation layer and the resource supply status of the corresponding cooling unit in the cooling layer are monitored in real time. The real-time resource matching degree between the computing power allocation layer and the cooling layer is analyzed. A preset coupling analysis model is used to obtain the performance coupling coefficient between the computing agent and the cooling agent. The coupling analysis model includes, but is not limited to, a calculation model based on multi-dimensional vector space similarity, including but not limited to cosine similarity analysis. The model analyzes the supply and demand matching degree between the computing agent and the cooling agent and outputs the calculated performance coupling coefficient. By calculating the performance coupling coefficient, the coupling relationship is quantitatively analyzed, providing accurate data support for weight adjustment. This enables the identification of efficient combinations of computing agents and cooling agents, improving the effectiveness of the collaborative strategy.

[0107] Specifically, for agents with an efficiency coupling coefficient greater than a preset coupling threshold, the weights are amplified by combining the corresponding efficiency coupling coefficient weight with the energy consumption to performance ratio of each agent, resulting in a first dynamic weight. Based on historical operating data or energy efficiency targets, a preset coupling threshold is set. For agents with an efficiency coupling coefficient greater than this threshold, the weights are weighted amplified by combining the extent to which the efficiency coupling coefficient exceeds the threshold with the agent's current actual energy consumption to performance ratio, resulting in the first dynamic weight. By amplifying the weights, agents within the efficient collaborative region and with good energy efficiency can obtain higher weights and priorities in resource competition. This ensures that computing and cooling resources can prioritize and guarantee the needs of efficient combinations, thereby maintaining the effectiveness of efficient collaborative links, improving overall energy efficiency, and significantly enhancing the overall energy utilization efficiency of the computing center.

[0108] Meanwhile, for agents with an efficiency coupling coefficient less than or equal to a preset coupling threshold, the weights are reduced by analyzing the independent operating characteristics of each agent to obtain a second dynamic weight. Independent operating characteristics include, but are not limited to, absolute energy consumption level, energy efficiency during independent operation, and task latency. The corresponding feature vector energy efficiency ratio is calculated and combined with task priority to obtain weights. These weights are then weighted and reduced to obtain the second dynamic weight. By reducing the weights, the excessive consumption of global resources by units with poor coordination or low energy efficiency can be effectively suppressed, reducing energy waste in inefficient operations. Optimizing the resource structure improves the operating efficiency and stability of the computing center, thereby enhancing overall energy efficiency.

[0109] Specifically, by combining the first and second dynamic weights, the weights are calibrated and optimized using a pre-defined weight calibration model to obtain the dynamic weights corresponding to each agent. The weight calibration model includes, but is not limited to, a constraint optimization model. The model's objective function is to maximize the system's estimated overall energy efficiency. Constraints include, but are not limited to, the upper limit of the system's total resources, a reasonable range for the sum of the weights of various agents, and a threshold to prevent any single agent from having an excessively high weight. Under the premise of satisfying various constraints, the model seeks the weight allocation scheme that optimizes the objective function, analyzes the weight balance between different types of agents and the supply and demand of global resources, calibrates and optimizes the weights, and outputs the dynamic weights corresponding to each agent. By calibrating and optimizing the weights, the coordination and stability of multi-agent combinations in the collaborative process can be improved. The calibrated dynamic weights can meet the overall resource constraints and operational balance requirements of the system, improve the globality and feasibility of the collaborative strategy, thereby enhancing the effectiveness of energy consumption coordination and improving the overall energy efficiency of the computing center.

[0110] Furthermore, in the virtual computing power allocation architecture, dynamic weights and a collaborative game process of bidirectional bidding between computing agents and cooling agents are simulated to obtain a first collaborative strategy, including:

[0111] S701. Based on the dynamic weights corresponding to each agent, determine the bidding parameters of the computing agent and the cooling agent through a preset bidding analysis model.

[0112] S702. Simulate the collaborative game process of two-way bidding between computing agents and cooling agents in the virtual computing power allocation architecture according to the bidding parameters, and obtain the first collaborative strategy.

[0113] In this embodiment, based on the dynamic weights corresponding to each agent, the bidding parameters of the computation agent and the cooling agent are determined through a pre-set bidding analysis model. The bidding analysis model includes, but is not limited to, a deep neural network model pre-trained using a large amount of historical agent operation status data and dynamic weights. The bidding parameters of the computation agent include, but are not limited to, maximum acceptable energy consumption cost and expected computing power performance indicators. Based on the dynamic weights of the computation agent, the model increases the upper limit of its maximum acceptable energy consumption cost and ensures that its expected computing power performance indicators reflect the needs of its high-priority tasks. The bidding parameters of the cooling agent include, but are not limited to, cooling service price and maximum cooling capacity. The model, combined with the dynamic weights of the cooling agent, sets the cooling agents with higher weights at a more competitive level. The model outputs the bidding parameters of the computation agent and the cooling agent. By calculating the bidding parameters of the agents, the dynamic weights are transformed into competitive capabilities in the game process. This allows for the analysis of capability differences between agents, ensuring that the collaborative game process is based on the true capabilities of the agents, and improving the effectiveness and rationality of collaborative strategies.

[0114] Specifically, the collaborative game process of two-way bidding between computing agents and cooling agents is simulated in the virtual computing power allocation architecture according to the bidding parameters to obtain the first collaborative strategy. By simulating the collaborative game process of two-way bidding, the conflict of objectives between computing agents and cooling agents can be effectively coordinated. Through multiple rounds of game, the balance point of the overall interests of the system can be identified. The obtained first collaborative strategy is not a top-down mandatory command, but a consensus solution formed by the participants after expressing their intentions and weighing their interests during the game process. This can improve the rationality and feasibility of the strategy, enhance its adaptability, and better cope with the dynamic changes in the internal state of the computing center, thereby achieving better global energy efficiency control.

[0115] Furthermore, based on the bidding parameters, a collaborative game process of two-way bidding between computing agents and cooling agents is simulated in the virtual computing power allocation architecture to obtain a first collaborative strategy, including:

[0116] S801. Analyze the computational task requirements and performance requirements of the computational agent in the virtual computing power allocation architecture according to the bidding parameters, and determine the initial bidding scheme.

[0117] S802. Analyze the real-time resource distribution of the virtual computing power allocation architecture and determine the resource benchmark through a preset benchmark analysis model;

[0118] S803. Combining the initial bidding scheme and resource benchmark, simulate the allocation of resources and conduct a two-way bidding game on the energy efficiency of the resources allocated by the computing agent and the cooling agent to obtain the first collaborative strategy.

[0119] In this embodiment, the computational task requirements and performance requirements of computing agents are analyzed in the virtual computing power allocation architecture according to the bidding parameters to determine the initial bidding scheme. Within the virtual computing power allocation architecture, based on the bidding parameters of each computing agent, the computational task requirements and performance requirements of that agent are parsed. A preset scheme generation algorithm associates and maps these requirements with the bidding parameters. For example, for high-priority tasks, if the energy consumption to performance ratio is temporarily high, the algorithm, based on its dynamic weight, determines a bid close to the maximum acceptable energy consumption cost in the initial scheme, and determines the precise type and quantity of required computing power resources. Simultaneously, it generates a requirement for collaborative cooling services, generating an initial bidding scheme for each computing agent. This scheme determines the maximum energy consumption cost paid for a specific quantity and quality of computing power resources, as well as the expected cooling guarantee. By formulating the initial bidding scheme, the computational task requirements are transformed into a game process, clearly identifying the resource requirements of each computing agent and providing data support for resource matching.

[0120] Specifically, the real-time resource distribution of the virtual computing power allocation architecture is analyzed, including but not limited to the number of available CPU cores and idle memory capacity. Simultaneously, the total computing power requirement and corresponding total heat load for all initial bidding schemes are estimated. A pre-defined benchmark analysis model is used to determine the resource benchmark. This benchmark analysis model includes, but is not limited to, algorithms based on market equilibrium theory. The model comprehensively analyzes the current total supply and demand, and refers to historical transaction data and energy efficiency levels from the same period to calculate the corresponding resource benchmark. By calculating the resource benchmark, accurate value metrics and resource allocation benchmark data are provided for the two-way bidding game process, effectively preventing irrational situations in the game process and guiding the agent's bidding behavior towards the system's current overall optimal energy efficiency and resource utilization, thus optimizing the overall flow and allocation efficiency of resources.

[0121] Specifically, combining the initial bidding scheme and resource benchmarks, resource allocation simulation is performed, and a two-way bidding game is conducted on the energy efficiency of the resources allocated to the computing agent and the cooling agent to obtain the first collaborative strategy. A supply tender is generated based on the initial bidding scheme, the cooling agent's bidding parameters, and the resource benchmarks. Resource allocation simulation is then performed, and computing power resource blocks are initially matched with cooling resource blocks based on the bids, asking prices, and resource benchmarks of both parties. The computing agent can not only bid for its required computing power resources but also make demands on the cooling efficiency of the matched cooling agent. The cooling agent commits to its service energy efficiency level when providing cooling services. The game process can be multi-round; agents can adjust their bidding strategies in the next round based on the matching results of the previous round and the dynamic changes in the resource benchmarks, until a Nash equilibrium is reached, where no agent can benefit by unilaterally changing its bid. At this point, the computing power allocation result of the computing agent and the service provided by the cooling agent are obtained, and the first collaborative strategy is output. By combining resource benchmarks with two-way bidding, the allocation of computing resources can be clearly defined. Through two-way bidding for energy efficiency, it can be ensured that the final output resource combination is the best balance in terms of performance, energy consumption, and cost, thereby improving the feasibility and rationality of the strategy. This can effectively transform the optimization of the virtual space into efficient operation in the physical world, and improve the overall energy efficiency and efficiency of the computing center under variable loads.

[0122] Furthermore, according to the first collaborative strategy, intelligent agent energy consumption control instructions are generated to control each intelligent agent to execute corresponding energy consumption control and energy efficiency improvement actions, including:

[0123] S901, Decompose the first cooperative strategy into energy consumption budget instructions for computing agents and cooling control instructions for cooling agents;

[0124] S902. Control each computing agent to perform corresponding energy consumption control actions according to the energy consumption budget instruction;

[0125] S903. Control each cooling agent to perform the corresponding energy efficiency improvement action according to the cooling control command.

[0126] In this embodiment, the first collaborative strategy is decomposed into energy consumption budget instructions for computing agents and cooling control instructions for cooling agents. Instructions are decomposed based on the resource allocation results in the strategy. These resource allocation results include, but are not limited to, the amount of computing resources obtained by each computing agent and its corresponding energy consumption limit, as well as the service objects and cooling output of each cooling agent. For computing agents, energy consumption budget instructions are generated, including but not limited to the power consumption limit. For cooling agents, cooling control instructions are generated, including but not limited to the target cooling area, the temperature or humidity range to be maintained, and the setpoint for cooling output. By obtaining specific operation instructions through strategy decomposition, the control boundaries and responsibilities of computing agents and cooling agents are clearly defined, avoiding execution confusion that may be caused by overlapping or omitted instructions, thereby improving the overall energy efficiency of the computing center.

[0127] Specifically, each computing agent is controlled to perform corresponding energy consumption control actions according to energy consumption budget instructions. The computing agents dynamically adjust the processor's operating frequency and voltage, and intelligently migrate and merge computing tasks among processor cores based on task priority and real-time load, thereby placing idle cores in a low-power sleep state. While ensuring that task completion deadlines are met, the execution speed of non-critical computing tasks is moderately adjusted. By executing precise energy consumption control actions, redundant power consumption and idle losses in the computing process can be effectively eliminated, reducing the power consumption of the computing center and improving energy efficiency.

[0128] Specifically, each cooling agent is controlled to perform corresponding energy efficiency improvement actions according to cooling control commands. Based on the service area and target temperature set in the commands, the cooling agents dynamically adjust the compressor speed, the opening of the electronic expansion valve, or the frequency of the variable frequency water pump to ensure that the cooling output precisely matches the real-time heat load, avoiding overcooling or insufficient cooling. Based on the load distribution information provided by the virtual energy bus, the fan speed or the opening and closing angle of the intelligent air valves is adjusted to optimize airflow organization, accurately delivering cooling capacity to the most needed hot spots and reducing heat and cold mixing losses. For systems consisting of multiple cooling devices, the number of devices activated and their respective load rates are dynamically determined through a group control strategy based on the overall cooling demand, ensuring that the system always operates at the point of highest overall energy efficiency. By executing energy efficiency improvement actions, energy waste caused by overcooling and ineffective cooling capacity delivery can be reduced, improving the overall energy efficiency ratio of the cooling system.

[0129] like Figure 4 As shown, a multi-agent energy consumption coordination system for a computing center is used to implement a multi-agent energy consumption coordination method for a computing center, including:

[0130] The virtual computing power allocation architecture construction module simulates the computing power working status of the computing power center and the energy consumption status of intelligent agents based on the pre-acquired computing power center data, analyzes the virtual energy bus reflecting the computing power allocation and the energy efficiency credit reflecting the energy efficiency of intelligent agents, and constructs the virtual computing power allocation architecture.

[0131] The dynamic weight allocation module, based on the real-time energy consumption data of each agent, configures a collaborative strategy to analyze the ratio of energy consumption to performance of each agent to obtain the corresponding dynamic weight. The agents include computing agents and cooling agents.

[0132] The multi-agent collaborative strategy formulation module simulates dynamic weights and the collaborative game process of bidirectional bidding between computing agents and cooling agents in the virtual computing power allocation architecture to obtain the first collaborative strategy.

[0133] The multi-agent collaborative control module generates energy consumption control instructions for each agent according to the first collaborative strategy, and controls each agent to perform corresponding energy consumption control actions and energy efficiency improvement actions to coordinate the energy consumption of the multi-agents in the computing center.

[0134] In this embodiment, the virtual computing power allocation architecture construction module integrates real-time and historical operating data of the computing center through digital twin technology to construct a dynamic virtual architecture that combines a virtual energy bus reflecting resource coupling relationships and an energy efficiency credit mechanism reflecting individual energy efficiency performance. This enables the mapping of computing power and cooling processes and energy consumption assessment, providing data support for collaborative energy consumption optimization. The dynamic weight allocation module, based on real-time collected energy consumption and performance data, analyzes the task requirements of computing agents and the service efficiency of cooling agents. Combining this with the coupling relationships in the virtual architecture, it assigns an adaptively adjusted dynamic weight to each agent. This constructs a weight allocation mechanism that can accurately respond to system state changes, distinguish task priorities, and incentivize energy efficiency improvements. This allows collaborative decision-making to analyze and calculate task priorities, ensuring resources are tilted towards high-efficiency and critical links, achieving precise guidance towards optimization goals.

[0135] In the virtual computing power allocation architecture, the multi-agent collaborative strategy formulation module uses dynamic weights as influence parameters to drive the computing agent and the cooling agent to engage in a two-way bidding collaborative game around resource allocation and energy efficiency commitments. The first collaborative strategy is generated by simulating the market clearing process. The command-based scheduling is transformed into a distributed market adjustment mechanism. Through game theory, individual and global goals are coordinated, so that the final strategy is not only technically feasible, but also approaches the overall energy efficiency optimal solution under the current constraints, thus improving the scientific nature and adaptability of the decision-making.

[0136] The multi-agent collaborative control module parses the first collaborative strategy derived from the game into specific control instructions, including energy consumption budget instructions for the computation agent and cooling regulation instructions for the cooling agent, and drives physical devices to perform corresponding energy consumption control actions and energy efficiency improvement actions. Through linkage control, deep collaboration between the computation agent and the cooling agent is ensured, thereby reducing energy consumption and improving energy efficiency, and ensuring the smooth progress of the multi-agent collaborative process.

[0137] The above description is merely a preferred embodiment of this application. The scope of protection of this application is not limited to the above embodiments. All technical solutions falling within the scope of this application's concept are within the scope of protection of this application. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of this application should also be considered within the scope of protection of this application.

Claims

1. A method for coordinated energy consumption among multiple intelligent agents in a computing center, characterized in that, include: Based on the pre-acquired computing center data, simulate the computing power working status of the computing center and the energy consumption status of the intelligent agent, analyze and calculate the real-time computing power consumption change curve when the intelligent agent performs computing tasks, and the corresponding cooling demand distribution map during the operation of the cooling intelligent agent. Based on the real-time computing power consumption change curve, the computing power demand is analyzed according to the preset priority of computing tasks, and a computing power allocation layer is constructed. The load aggregation characteristics are analyzed in the cooling demand distribution map, cooling zones are dynamically divided, corresponding cooling resources are configured for each cooling zone, and a cooling layer is constructed. Based on the positional relationship between the computation agent and the cooling agent, analyze the distance, path accessibility and cooling efficiency between the computation agent and the nearest cooling agent, and calculate the corresponding positional coupling factor. By using the location coupling factor, a mapping relationship is established between the computing power allocation layer and the cooling layer, and a virtual energy bus consisting of a dual-layer structure of computing power allocation layer and cooling layer is constructed. Analyze the historical energy efficiency benchmark value and real-time operating energy efficiency deviation of each intelligent agent, and compare the difference between the real-time operating energy efficiency deviation and the benchmark using the average value of the historical energy efficiency benchmark value as the benchmark to obtain the corresponding energy efficiency credit; By combining the virtual energy bus and energy efficiency credits, a virtual computing power allocation architecture is constructed. Based on the real-time energy consumption data of each agent, the real-time computing task requirements of the agent are analyzed and mapped to the corresponding computing power requirements to obtain the first weight of each computing agent. The correlation between cooling efficiency and energy consumption of the cooling agent is analyzed, and the second weight of each cooling agent is obtained by combining the correlation with the real-time energy consumption data of the cooling agent. Based on the coupling relationship between computing agents and cooling agents in the virtual computing power allocation architecture, the energy consumption to performance ratio of each agent is analyzed, and the first weight and the second weight are dynamically adjusted to obtain the corresponding dynamic weight. Based on the dynamic weights corresponding to each agent, the bidding parameters of the computing agent and the cooling agent are determined through a pre-set bidding analysis model. The bidding analysis model includes a deep neural network model pre-trained using historical agent operation status data and dynamic weights. The first collaborative strategy is obtained by simulating a two-way bidding collaborative game process between computing agents and cooling agents in a virtual computing power allocation architecture according to the bidding parameters. According to the first collaborative strategy, intelligent agent energy consumption control instructions are generated to control each intelligent agent to perform corresponding energy consumption control actions and energy efficiency improvement actions, so as to coordinate the energy consumption of multiple intelligent agents in the computing center.

2. The multi-agent energy consumption coordination method for computing centers according to claim 1, characterized in that, Combining the aforementioned virtual energy bus and energy efficiency credits, a virtual computing power allocation architecture is constructed, including: In the virtual energy bus, analyze the computing power demand of the computing power allocation layer and the cooling resources of the cooling layer to calculate the computing power cooling situation and calculate the cooling efficiency. The computing power requirement of the computing power allocation layer is calculated based on the energy efficiency credit analysis, and the efficiency value of the cooling resources provided by the cooling agent is calculated. The circulation efficiency is then obtained by combining the cooling efficiency. Based on the circulation efficiency, an initial energy efficiency credit and dynamic credit limit are configured for each intelligent agent corresponding to the virtual energy bus through a preset credit allocation model, thereby constructing a virtual computing power allocation architecture. The credit allocation model includes a neural network model pre-trained using historical intelligent agent resource allocation data.

3. The multi-agent energy consumption coordination method for computing centers according to claim 1, characterized in that, The step involves analyzing the energy consumption to performance ratio of each agent based on the coupling relationship between the computing agent and the cooling agent in the virtual computing power allocation architecture, and dynamically adjusting the first and second weights to obtain the corresponding dynamic weights, including: The real-time resource matching degree of the computing power allocation layer and the cooling layer in the virtual computing power allocation architecture is analyzed. The efficiency coupling coefficient of the computing agent and the cooling agent is obtained through a preset coupling analysis model. The coupling analysis model includes a calculation model based on cosine similarity. For agents whose efficiency coupling coefficient is greater than a preset coupling threshold, the first weight or second weight corresponding to the agent is amplified by combining the corresponding efficiency coupling coefficient weight and the ratio of energy consumption to performance of each agent to obtain the first dynamic weight. The efficiency coupling coefficient weight is the extent to which the efficiency coupling coefficient exceeds the coupling threshold. For agents whose performance coupling coefficient is less than or equal to a preset coupling threshold, the weights are reduced by analyzing the independent operation characteristics of each agent to obtain a second dynamic weight. By combining the first dynamic weight and the second dynamic weight, the weights are calibrated and optimized using a preset weight calibration model to obtain the dynamic weights corresponding to each agent.

4. The multi-agent energy consumption coordination method for computing centers according to claim 1, characterized in that, Based on the bidding parameters, a collaborative game process of two-way bidding between computing agents and cooling agents is simulated in the virtual computing power allocation architecture to obtain a first collaborative strategy, including: Based on the bidding parameters, analyze the computing task requirements and performance requirements of the computing agent in the virtual computing power allocation architecture to determine the initial bidding scheme; Analyze the real-time resource distribution of the virtual computing power allocation architecture and determine the resource benchmark through a preset benchmark analysis model; Based on the initial bidding scheme and resource benchmark, resource allocation simulation is performed, and a two-way bidding game is conducted on the energy efficiency of the resources allocated by the computing agent and the cooling agent to obtain the first collaborative strategy.

5. The multi-agent energy consumption coordination method for a computing center according to claim 1, characterized in that, According to the first collaborative strategy, intelligent agent energy consumption control instructions are generated to control each intelligent agent to execute corresponding energy consumption control actions and energy efficiency improvement actions, including: The first collaborative strategy is decomposed into energy consumption budget instructions for computational agents and cooling control instructions for cooling agents. According to the energy consumption budget instructions, each computing agent is controlled to perform corresponding energy consumption control actions; The cooling control command is used to control each cooling agent to perform corresponding energy efficiency improvement actions.

6. A multi-agent energy consumption collaborative system for a computing center, characterized in that, A method for coordinating the energy consumption of multiple agents in a computing center as described in any one of claims 1 to 5 includes: The virtual computing power allocation architecture construction module simulates the computing power working status of the computing power center and the energy consumption status of intelligent agents based on the pre-acquired computing power center data, analyzes the virtual energy bus reflecting the computing power allocation and the energy efficiency credit reflecting the energy efficiency of intelligent agents, and constructs the virtual computing power allocation architecture. The dynamic weight allocation module, based on the real-time energy consumption data of each agent, configures a collaborative strategy to analyze the ratio of energy consumption to performance of each agent to obtain the corresponding dynamic weight. The agents include computing agents and cooling agents. The multi-agent collaborative strategy formulation module simulates dynamic weights and the collaborative game process of bidirectional bidding between computing agents and cooling agents in the virtual computing power allocation architecture to obtain the first collaborative strategy. The multi-agent collaborative control module generates energy consumption control instructions for each agent according to the first collaborative strategy, and controls each agent to perform corresponding energy consumption control actions and energy efficiency improvement actions to coordinate the energy consumption of the multi-agents in the computing center.

Images