Green intelligent computing center computing power dynamic regulation and control method based on carbon emission demand analysis

Abstract

The invention relates to the field of green computing and intelligent computing center management, and discloses a green intelligent computing center computing power dynamic regulation and control method based on carbon emission demand analysis, which comprises the following steps: data acquisition and analysis: acquiring multi-dimensional data of a green intelligent computing center in real time, including carbon emission intensity, server load, energy consumption, environmental conditions and cooling system energy efficiency data, preprocessing and analyzing the data; energy efficiency management and control standards are dynamically generated, the energy efficiency management and control standards of the intelligent calculation center are generated based on the data analysis result, and the energy efficiency management and control standards comprise a server load threshold value, carbon emission intensity limitation and a cooling system energy efficiency standard; dynamic regulation and control of computing power are implemented based on an energy efficiency management and control standard; and real-time load conditions, energy consumption characteristics and carbon emission intensity of the server are combined. According to the invention, by dynamically generating the energy efficiency management and control standard, feeding back the regulation and control strategy in real time and optimizing the computing power distribution, the efficient distribution of the computing power of the intelligent computing center is realized, the resource utilization rate is improved, and the carbon emission and the energy consumption are obviously reduced.

Description

Technical Field

[0001] The present invention relates to the field of green computing and intelligent computing center management technology, and specifically to a method for dynamically controlling the computing power of a green intelligent computing center based on carbon emission demand analysis. Background Art

[0002] With the rapid development of information technology, intelligent computing centers have become an important infrastructure to support large-scale computing tasks and artificial intelligence model training. In life, online education, cloud computing, autonomous driving and other application scenarios all require intelligent computing centers to provide efficient computing support. However, the operation of intelligent computing centers is accompanied by huge energy consumption problems, which directly affects their economic and sustainable development capabilities. Especially in the context of the global promotion of a green and low-carbon economy, how to ensure the efficient operation of the computing power of intelligent computing centers while achieving the goal of energy conservation and emission reduction has become a topic of great concern to all sectors of society.

[0003] In the existing technology, some solutions have been proposed to improve the operating efficiency and optimize energy consumption of intelligent computing centers. For example, through load balancing technology, computing tasks can be assigned to different servers, thereby reducing the overload of individual servers and improving overall resource utilization. At the same time, some technical solutions use hardware energy efficiency optimization methods, such as using low-power chips or high-efficiency refrigeration equipment to reduce unit energy consumption. In addition, optimization technologies for cooling systems have gradually matured, such as dynamically adjusting fan speed and coolant temperature to reduce unnecessary cooling energy consumption. These technologies have achieved remarkable results in improving the operating efficiency of intelligent computing centers and reducing some energy consumption.

[0004] However, the existing technology still has many shortcomings. First, computing power regulation mostly adopts a static allocation strategy, which fails to dynamically adjust according to the real-time load of the task, resulting in low resource utilization and serious overload or idleness of some servers. Second, the current energy consumption optimization scheme mainly focuses on a single factor, such as load balancing or cooling system efficiency, while ignoring the comprehensive impact of multi-dimensional factors such as carbon emission intensity, and cannot adapt to the current needs of green and low-carbon development. Third, the existing technology lacks a closed-loop feedback mechanism and cannot make real-time adjustments to the computing power allocation according to the actual deviations in operation, resulting in rigid control strategies and difficulty in adapting to changes in task requirements. Summary of the invention

[0005] In view of the shortcomings of the existing technology, the present invention provides a method for dynamically controlling the computing power of a green intelligent computing center based on carbon emission demand analysis, which solves the comprehensive problems of rigid computing power control, uneven resource allocation, single energy consumption optimization and lack of carbon emission control in the existing technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for dynamically controlling the computing power of a green intelligent computing center based on carbon emission demand analysis, comprising the following steps: Data collection and analysis: Real-time acquisition of multi-dimensional data from the Green Intelligent Computing Center, including carbon emission intensity, server load, energy consumption, environmental conditions, and cooling system energy efficiency data, and pre-processing and analysis of the data; Dynamically generate energy efficiency control standards: Based on the data analysis results, generate energy efficiency control standards for the intelligent computing center, including server load thresholds, carbon emission intensity limits, and cooling system energy efficiency benchmarks; Implement dynamic control of computing power based on energy efficiency control standards: dynamically adjust the allocation path of computing tasks based on the real-time load of the server, energy consumption characteristics and carbon emission intensity; Computing power allocation optimization and energy efficiency feedback control: Through a closed-loop feedback mechanism, real-time operation data is collected, the actual operation results are compared with energy efficiency management standards, the computing power allocation path is optimized, and the control strategy is dynamically adjusted.

[0007] Preferably, the data collection and analysis includes: Obtain carbon emission intensity data through sensors and energy interfaces; Obtain load rate and energy consumption data from the server; Obtain regional temperature and humidity data from environmental monitoring equipment; Get real-time energy efficiency data from cooling equipment; The collected data is cleaned and standardized, and a carbon emission prediction model is built using machine learning algorithms.

[0008] Preferably, the dynamically generated energy efficiency control standard includes: Generate carbon emission intensity limits based on carbon emission forecast models; Generate server load safety operation threshold based on load analysis; Generate cooling system energy efficiency benchmark based on cooling system operation data; Dynamically predict future carbon emission trends and adjust the energy efficiency control standards in real time.

[0009] Preferably, the dynamic control of computing power includes: Establish a computing power allocation optimization model based on multi-dimensional factors, including server load, energy consumption, carbon emission intensity and environmental conditions; A task priority evaluation model is used to classify tasks according to their priority and computational complexity; Dynamically adjust the task allocation path to assign high-priority tasks to servers with low energy consumption or low carbon emission intensity.

[0010] Preferably, the task priority evaluation model includes: Obtain key parameters of the task, including the timeliness of the task, computational complexity, and the task's demand for computing resources; A hierarchical model for task priority evaluation is constructed through the analytic hierarchy process, with task timeliness as the highest priority factor and computational complexity and computational resource requirements as secondary factors; Calculate the task priority scores and sort the tasks according to the scores; According to the task priority sorting, high-priority tasks are preferentially assigned to servers with low load and high energy efficiency, and the execution time of low-priority tasks is adjusted or assigned to low-energy consumption mode.

[0011] Preferably, the computing power allocation optimization includes: By real-time monitoring of the server's load and energy consumption, when a server's load exceeds the set threshold, task migration is triggered; Prioritize migrating high-energy-consuming tasks to servers with lower carbon emission intensity or under low load; Dynamically adjust the distribution of computing resources to achieve load balancing and energy efficiency optimization.

[0012] Preferably, the closed-loop feedback mechanism includes: Real-time collection of the intelligent computing center's operating data, including task completion time, unit energy consumption, and actual carbon emissions; Compare actual operation data with energy efficiency control standards to evaluate the deviation of computing power regulation; Dynamically adjust energy efficiency control standards and computing power allocation strategies based on the evaluation results.

[0013] Preferably, the dynamic adjustment and control strategy includes: When carbon emission intensity decreases, high energy consumption tasks are allocated to low carbon emission areas first; Delay the execution of low-priority tasks during peak load periods and adjust them to run during low-load or off-peak periods; Dynamically adjust server workloads based on changes in cooling system operating efficiency.

[0014] Preferably, the dynamic control of computing power based on energy efficiency control standards is implemented: combining the real-time load situation of the server, energy consumption characteristics and carbon emission intensity, dynamically adjusting the allocation path of computing tasks, specifically including: The server load rate, real-time carbon emission intensity, ambient temperature and humidity data, and server power consumption are used as input variables; Use supervised learning algorithms to train the model and build a mapping relationship between carbon emissions and the above input variables based on historical operating data; Use the long short-term memory network to predict the carbon emission intensity in a specified time period in the future and generate a carbon emission intensity trend curve; Dynamically adjust the task allocation strategy based on the prediction results, including prioritizing high-energy-consuming tasks to low-carbon areas or off-peak hours.

[0015] Preferably, the computing power allocation optimization and energy efficiency feedback control: through a closed-loop feedback mechanism, real-time collection of operating data, comparison of actual operating results with energy efficiency control standards, optimization of computing power allocation paths and dynamic adjustment of control strategies, specifically include: Minimize the server's energy consumption per task; Minimize task execution time; Maximize overall server load balancing and cooling system energy efficiency.

[0016] The present invention provides a method for dynamically regulating the computing power of a green intelligent computing center based on carbon emission demand analysis. It has the following beneficial effects: 1. The present invention adopts a technical solution based on multi-dimensional data collection, dynamic generation of energy efficiency control standards, and optimization of computing power allocation, which achieves the effect of comprehensively improving the energy efficiency of the green intelligent computing center. Compared with the solution of static task allocation based on a single factor in the prior art, it solves the problem of resource waste and high carbon emissions caused by the lack of real-time feedback and dynamic adjustment.

[0017] 2. The present invention collects and compares the operation data in real time through a closed-loop feedback mechanism, dynamically optimizes the computing power allocation path, and realizes the combination of flexibility in task allocation and energy consumption control. Compared with the prior art solution that relies on preset rules for task allocation, it overcomes the shortcomings of slow response to task allocation and uneven resource allocation, and effectively improves the efficiency of task completion and overall resource utilization.

[0018] 3. The present invention combines regional carbon emission intensity and cooling system energy efficiency to dynamically generate energy efficiency control standards, significantly reducing the carbon emission intensity of the intelligent computing center. Compared with the traditional task scheduling scheme that lacks consideration of carbon emissions in the prior art, the present invention solves the problem of disconnection between carbon emission control and computing power regulation, and provides a feasible solution for the low-carbon operation of green data centers.

[0019] 4. The present invention combines reinforcement learning algorithms with task priority models to dynamically adjust task allocation strategies, fully realizing the timely completion of high-priority tasks and delayed execution of low-priority tasks. Compared with the rigid task scheduling strategies in the prior art that cannot adapt to real-time demand changes, the present invention solves the problems of low task execution efficiency and task backlogs, and enhances the flexibility and adaptability of the intelligent computing center. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Please see attached Figure 1 The embodiment of the present invention provides a method for dynamically controlling the computing power of a green intelligent computing center based on carbon emission demand analysis, comprising the following steps: S1. Data collection and analysis: Real-time acquisition of multi-dimensional data from the Green Intelligent Computing Center, including carbon emission intensity, server load, energy consumption, environmental conditions, and cooling system energy efficiency data, and pre-processing and analysis of the data; Data collection and analysis are the basic steps to achieve dynamic control of computing power in a green intelligent computing center. Its main function is to provide sufficient data support for the generation of subsequent energy efficiency management standards and computing power control. By comprehensively collecting multi-dimensional data from the intelligent computing center in real time and extracting key features in combination with the analysis process, it is possible to accurately characterize the energy consumption and carbon emission status of the intelligent computing center; the technical content of this part includes the design of the data acquisition system, the preprocessing process of the data, and the establishment of a data analysis model. By fully revealing the specific implementation method of the present invention in this technical link, those skilled in the art can accurately reproduce it.

[0023] Generally, the operation of the intelligent computing center involves a variety of energy consumption and environmental data. In order to achieve accurate collection, the present invention obtains relevant information through the following multi-dimensional data sources.

[0024] As an option, real-time server load data is the core part of the collection, which is mainly used to characterize the operating status and load level of each server during task processing. Specifically, key performance indicators such as CPU usage and GPU occupancy can be obtained through the resource monitoring module deployed inside the server. At the same time, power consumption data can also be directly extracted from the server hardware in watts (W). These data directly reflect the energy consumption characteristics of task execution.

[0025] In some embodiments, the carbon emission intensity is obtained by connecting to an external energy interface, which provides real-time carbon emission intensity data of regional energy supply. Specifically, when the carbon emission intensity of regional energy changes, such as switching from fossil energy to renewable energy, the system can dynamically update the value of carbon emission intensity.

[0026] In addition, the energy efficiency data of the cooling system also needs to be collected in real time. Generally, the energy efficiency ratio of the cooling system can be obtained in real time through the power meter and temperature sensor installed on the cooling equipment. For example, in the refrigeration system, the COP calculation formula is:

[0027] in, Indicates the amount of heat absorbed by the cooling system in kilowatts (kW). Indicates the input power consumed by the cooling equipment in kilowatts (kW).

[0028] In one possible implementation, environmental condition data is collected by environmental sensors, mainly including temperature, humidity, and regional meteorological data. These data have a direct impact on the operating efficiency of the cooling system. For example, when the external ambient temperature is low, the power consumption of the cooling system will be significantly reduced, so it is necessary to dynamically adjust the cooling parameters.

[0029] Generally, after data collection is completed, the system will pre-process the raw data to ensure its accuracy and availability. Specifically, the collected multi-dimensional data is first cleaned to remove sensor errors, communication noise, and outliers. For example, when the server load rate fluctuates abnormally, it can be corrected through the median filtering method.

[0030] As an option, standardization is one of the key steps in data preprocessing. Data such as server load rate and energy efficiency ratio usually have different dimensions. Through normalization, all data can be mapped to the [0,1] interval. The formula is as follows:

[0031] in, is the normalized value, is the original value, and are the minimum and maximum values ​​of the data respectively.

[0032] In some embodiments, in order to improve the temporal consistency of data, time series aggregation processing can be performed on multidimensional data. For example, real-time data in seconds can be aggregated into a one-minute average value, which can not only reduce the amount of data, but also avoid interference with subsequent model calculations caused by excessive fine-grained fluctuations.

[0033] The main purpose of data analysis is to extract key features from the collected multi-dimensional data to support the generation of subsequent energy efficiency standards and the establishment of computing power regulation models.

[0034] In a possible implementation, the present invention uses a machine learning method to establish a carbon emission prediction model. Specifically, the server load rate, power consumption data, carbon emission intensity and environmental conditions are used as input variables, and a carbon emission prediction model is constructed through a supervised learning algorithm. Random forest or support vector machine (SVM) can be selected for modeling, and the formula is:

[0035] C(t) is the predicted value of carbon emission intensity at time t; L(t) is the server load rate at time t; E(t) is the regional energy carbon emission intensity at time t; P(t) is the power consumption of the server at time t; W(t) is the environmental condition data at time t, including temperature and humidity.

[0036] Specifically, when the amount of data is large, the model can be partitioned. For example, the carbon emission prediction model can be trained for different load intervals to improve the prediction accuracy.

[0037] As an extension, data analysis can also introduce long short-term memory networks (LSTM) to predict the time series of carbon emission intensity. The LSTM model can capture nonlinear change trends over a long time span, thus better adapting to complex task scenarios.

[0038] S2. Dynamically generate energy efficiency control standards: Based on the data analysis results, generate energy efficiency control standards for the intelligent computing center, which include server load thresholds, carbon emission intensity limits, and cooling system energy efficiency benchmarks; Dynamically generating energy efficiency management and control standards is an important part of achieving energy conservation and emission reduction and efficient allocation of computing power in the intelligent computing center. Based on the results of the aforementioned data analysis, the present invention forms a dynamic energy efficiency management and control standard by calculating and adjusting the server load threshold, carbon emission intensity limit, and cooling system energy efficiency benchmark. This standard not only takes into account the current operating status of the intelligent computing center, but also can adapt to possible changes in the operating environment and task requirements in the future, thereby providing a reliable basis for computing power regulation.

[0039] In general, the generation of energy efficiency control standards relies on the comprehensive analysis of multi-dimensional data and the calculation results of the prediction model. By dynamically adjusting data such as server load, regional energy carbon emission intensity, and cooling system efficiency, the resource utilization efficiency of the intelligent computing center can be optimized.

[0040] As an option, the formulation of energy efficiency management and control standards needs to be combined with the priority and computational complexity of specific tasks to ensure that both the real-time requirements of the tasks are met and optimization effects are achieved in terms of energy consumption and carbon emissions.

[0041] In the intelligent computing center, the server load rate directly affects its energy consumption performance and stability. Generally speaking, the load rate of each server needs to be maintained in a certain safe operating range. Too low will lead to resource waste, and too high may cause overload and system instability.

[0042] Specifically, the present invention dynamically generates a load threshold for each server by analyzing historical load data and real-time load status. For example, the load interval of the server can be set to 50% to 80% to ensure a balance between its operating efficiency and cooling cost.

[0043] In some embodiments, the load threshold is calculated as follows:

[0044] in: is the load threshold of the server; is the average load in historical operation data; is the peak load in historical data; and They are weight coefficients, which are used to adjust the dynamic change requirements of the current load.

[0045] In one possible implementation, the load threshold can be further adjusted by real-time monitoring of server temperature. When the server temperature approaches the tolerance limit of the device, the system will automatically lower the load threshold and prioritize migrating tasks to other low-load devices to avoid the formation of local hot spots.

[0046] Generally, carbon emission intensity limits are set dynamically based on the carbon emission characteristics of regional energy supply. This limit can help the intelligent computing center prioritize the use of low-carbon energy and reduce overall carbon emissions.

[0047] Specifically, the present invention obtains the regional energy carbon emission intensity in real time and calculates the carbon emission upper limit of the current period in combination with the energy consumption characteristics of the task. As an option, the carbon emission intensity limit can be determined by the following formula:

[0048] in: is the limit value of carbon emission intensity; is the total energy consumption of the intelligent computing center, in kWh; is the total number of current tasks; The allowed carbon emission time window, in hours.

[0049] In some embodiments, when the proportion of renewable energy in the regional energy supply increases, the system will automatically increase the carbon emission intensity limit. For example, during the low-peak period at night, when the proportion of renewable energy power supply reaches 90%, the allowed carbon emission intensity will be appropriately relaxed to perform high-energy-consuming tasks.

[0050] In one possible implementation, carbon emission intensity limits can also be adjusted dynamically based on the forecast results of future energy supply. For example, by using the long short-term memory network (LSTM) to predict the energy carbon emission change trend in the next hour, the system can adjust the carbon emission limit in advance to adapt to the dynamic changes in the energy structure.

[0051] The cooling system is an important component of the energy consumption of the intelligent computing center, and its operating efficiency directly affects the total energy consumption. Generally speaking, the cooling system capacity efficiency ratio (COP) is used to measure the cooling efficiency, and a higher COP value indicates a higher cooling efficiency.

[0052] Specifically, the present invention dynamically generates an energy efficiency benchmark by collecting real-time operating data of the cooling system. For example, the optimal energy efficiency ratio of the cooling system can be calculated by the following formula:

[0053] in: The optimal energy efficiency ratio for the cooling system; is the ambient temperature in degrees Celsius; is the temperature inside the server after cooling, in degrees Celsius; is the input power of the cooling system in kilowatts (kW).

[0054] As an option, when the external ambient temperature is low, the system will dynamically increase the temperature baseline of the cooling system to reduce cooling power consumption. Conversely, under high summer temperatures, the system will maintain cooling efficiency by adjusting fan speed and water cooling circulation rate.

[0055] In some embodiments, the energy efficiency benchmark of the cooling system can also be adjusted in real time in combination with changes in server load and regional temperature and humidity. For example, when the server load is low and the ambient temperature is low, the system can reduce the cooling intensity, thereby further saving energy consumption.

[0056] In a possible extended application, the dynamic energy efficiency control standard of the present invention can be combined with a multi-task scheduling optimization strategy to further improve system performance. For example, by combining the priority and load distribution of tasks, the resource allocation standards of different tasks can be dynamically adjusted to achieve a higher task completion rate and lower energy consumption.

[0057] In other embodiments, the dynamically generated energy efficiency control standards can also be linked to the dynamic pricing mechanism of the external power grid. For example, when the power grid price decreases, the system can relax the energy efficiency restrictions to perform more high-energy-consuming tasks, thereby reducing operating costs.

[0058] S3. Implement dynamic control of computing power based on energy efficiency control standards: dynamically adjust the allocation path of computing tasks based on the real-time load of the server, energy consumption characteristics and carbon emission intensity; By combining energy efficiency control standards, the distribution path of computing tasks in the intelligent computing center is dynamically adjusted to achieve the dual goals of energy conservation and emission reduction and efficient operation. Based on the real-time load of the server, energy consumption characteristics and regional energy carbon emission intensity, dynamic allocation of task paths is a key link in achieving overall system optimization. Dynamic regulation can not only solve the problem of uneven server load, but also prioritize high-energy-consuming tasks to servers with lower carbon emission intensity or higher energy efficiency, thereby minimizing the carbon footprint of computing tasks.

[0059] In general, the dynamic control of computing tasks requires comprehensive consideration of the priority of the task, the current operating status of the server, and the external environmental conditions. These factors influence each other and have a significant impact on the control results. To this end, the present invention adopts a method of comprehensive analysis of multidimensional factors and dynamically adjusts the allocation path of computing tasks through optimization models and algorithms.

[0060] Specifically, by combining historical data with real-time monitoring data, the present invention can quickly identify the server that is currently most suitable for task execution, and at the same time reasonably allocate tasks to achieve overall load balancing and maximize resource utilization.

[0061] In one possible implementation, dynamic control of computing power is based on energy efficiency control standards as the core constraint and is implemented through the following logic: first, the current load rate and remaining available computing power of each server are obtained in real time to identify efficient servers and potential bottleneck servers; second, tasks are hierarchically sorted according to their priority and energy consumption characteristics; finally, tasks are dynamically adjusted to prioritize the servers with the lowest energy consumption or the most balanced load.

[0062] Generally speaking, load balancing is the primary goal of computing power regulation. When the server load is high, some tasks can be assigned to servers with lower loads through task migration strategies to achieve optimal utilization of global resources.

[0063] As an option, when the task has a high requirement for computing timeliness, the present invention will preferentially assign high-priority tasks to servers with faster response speeds. Correspondingly, for low-priority tasks, they will be preferentially assigned to servers with lower loads and lower energy consumption to reduce overall energy consumption.

[0064] In a possible implementation, the present invention constructs a computing power control optimization model, takes multi-dimensional factors as input, and dynamically generates a task allocation plan. Specifically, the multi-dimensional factors include: Server Load Rate :Indicates the The real-time load of a server is usually based on CPU usage or the number of tasks executed.

[0065] Server energy consumption characteristics : It indicates the unit computing power consumption of the server (in watts / number of computing tasks), reflecting the energy efficiency of executing tasks.

[0066] Regional carbon emission intensity : reflects the carbon emission characteristics of regional energy, in units of .

[0067] Ambient temperature : Directly related to the efficiency of the cooling system, high temperature conditions will increase the energy consumption of the server.

[0068] The present invention realizes task allocation optimization through the following objective function:

[0069] in: To optimize the target value, reflect the energy and carbon emission costs of the overall task allocation; is the number of servers in the intelligent computing center; To regulate the weights and balance the importance of energy consumption and carbon emissions factors.

[0070] By solving the above optimization model, the system can dynamically generate the optimal task allocation plan.

[0071] Generally speaking, task priority evaluation is an important pre-step for computing power regulation. Priority evaluation can ensure that high-priority tasks are executed first when resources are tight.

[0072] Specifically, the present invention evaluates the priority of tasks through the analytic hierarchy process (AHP). The calculation of the priority involves the following three main parameters: Timeliness : Indicates the urgency with which a task needs to be completed, usually expressed as the remaining execution time.

[0073] Computational complexity : Indicates the amount of computing resources required for the task, in terms of task runtime.

[0074] Resource requirements : Indicates the amount of specific hardware resources occupied by a task.

[0075] In some embodiments, the score calculation formula of the task priority is as follows:

[0076] in: Score the task priority; are the weights of the three parameters, satisfying .

[0077] Tasks with higher priority scores will be assigned to servers with lower loads to ensure timely completion of tasks.

[0078] During the operation of the intelligent computing center, some servers are prone to performance degradation or even failure due to task overload. To this end, the present invention provides a dynamic task migration mechanism for achieving dynamic balancing of server loads.

[0079] In one possible implementation, when the load rate of a server exceeds a preset threshold, the system triggers a task migration strategy. The goal of task migration is to assign some tasks to servers with lower loads while ensuring that the execution efficiency of tasks does not drop significantly during the migration process.

[0080] Specifically, the present invention determines the task migration path through the following rules: Prioritize low-load servers: Task migration is preferentially assigned to servers with a load rate below 50%.

[0081] Avoid hot spots: When multiple servers meet the migration conditions, servers with higher cooling efficiency are given priority to avoid increased cooling energy consumption due to hot spots.

[0082] Considering network latency: For tasks that require real-time processing, servers with lower network latency are preferred.

[0083] In one possible implementation, the present invention uses a reinforcement learning algorithm to dynamically optimize the task allocation path. Specifically, the system models the computing power regulation problem as a multi-armed bandit problem and implements real-time optimization of task allocation through a deep Q learning (DQL) algorithm.

[0084] The algorithm is implemented by the following steps: State space definition: includes server load rate, energy consumption characteristics, and task priority.

[0085] Action space definition: including task allocation paths and migration strategies.

[0086] Reward function design: The timeliness of task completion and the degree of energy consumption optimization are the main reward indicators.

[0087] Through continuous training, the system can adjust the task allocation strategy in real time to adapt to changing load and task requirements.

[0088] In some extended applications, the dynamic control strategy of the present invention can be combined with intelligent computing centers deployed in multiple regions to optimize the allocation of computing power across regions. For example, when the carbon emission intensity of a certain region is high, tasks can be preferentially migrated to other low-carbon regions, thereby further reducing overall carbon emissions.

[0089] In addition, in another possible implementation, the control algorithm of the present invention can also be combined with the energy cost optimization of the task. For example, when the price of the power grid fluctuates greatly, the system can prioritize the execution of high-energy-consuming tasks during the low-price period to further reduce the operating cost.

[0090] S4. Computing power allocation optimization and energy efficiency feedback control: Through a closed-loop feedback mechanism, real-time operation data is collected, the actual operation results are compared with energy efficiency control standards, the computing power allocation path is optimized, and the control strategy is dynamically adjusted; In the present invention, computing power allocation optimization and energy efficiency feedback control are important steps to realize the closed-loop dynamic control of the intelligent computing center. By collecting the operating data of the intelligent computing center in real time, the system can compare the actual operating results with the energy efficiency management and control standards, and then adjust the task allocation path and control strategy, so as to achieve efficient utilization of computing power resources and dynamic optimization of system energy efficiency. In general, the closed-loop feedback mechanism continuously adjusts the distribution of computing power resources by continuously collecting and analyzing data to achieve the accuracy of task execution and maximize resource utilization.

[0091] As an option, the real-time operation data collected in the present invention includes task completion time, unit task energy consumption, carbon emission intensity, server load distribution, etc. These data can fully reflect the operation status of the intelligent computing center and provide support for the adjustment of subsequent optimization strategies.

[0092] Specifically, by establishing a feedback model, the system can analyze operational deviations and, combined with current energy efficiency management standards, dynamically optimize task allocation paths and control parameters to achieve energy conservation and emission reduction effects while ensuring the completion of high-priority tasks.

[0093] In one possible implementation, real-time collection of operating data relies on a multi-dimensional monitoring system deployed in the intelligent computing center. Generally, the monitoring system includes server load sensors, power consumption sensors, cooling system efficiency monitoring equipment, and regional carbon emission intensity interfaces.

[0094] As an option, server load data is collected through internal hardware monitoring modules. These modules record CPU and GPU usage, memory usage, and task queue length in real time. Taking CPU usage as an example, the load data of each server per unit time is recorded as a percentage and stored in a distributed database.

[0095] Carbon emission intensity data comes from an external interface provided by regional energy suppliers. Specifically, when the energy supply structure changes, such as an increase in the proportion of renewable energy, the system can automatically update the real-time carbon emission intensity to support the adjustment of the regulation strategy.

[0096] In some embodiments, the cooling system efficiency data is obtained by calculating the Coefficient of Performance (COP) in real time, using the following formula:

[0097] in: is the heat absorbed by the cooling system, in kilowatts (kW); is the input power of the cooling equipment in kilowatts (kW).

[0098] By continuously collecting the above-mentioned multi-dimensional data, the system can form a panoramic monitoring of the operating status of the intelligent computing center.

[0099] In general, the system will compare the actual operating data collected with the energy efficiency control standards item by item to identify operating deviations and determine the optimization direction. As an option, the energy efficiency control standards include server load thresholds, carbon emission intensity limits, and cooling system energy efficiency benchmarks.

[0100] Specifically, the present invention calculates the deviation between the actual operating effect and the energy efficiency standard by establishing a comparison model, and the formula is as follows:

[0101] in: It is the deviation ratio, which is used to quantify the degree of deviation from the actual operating effect; is the specific value of the actual operating data; It is the reference value of energy efficiency control standard.

[0102] In some embodiments, when the deviation ratio exceeds a preset threshold, the system triggers an adjustment strategy. For example, when the actual energy consumption of a server exceeds the standard energy consumption, the system reduces the energy consumption of the server through task migration or load adjustment.

[0103] As an extension, when the carbon emission intensity limit is exceeded, the system can give priority to backup servers in low-carbon emission areas to perform high-energy consumption tasks, thereby reducing overall carbon emissions.

[0104] Generally, the optimization of task allocation path needs to be dynamically adjusted in combination with actual operation deviation and energy efficiency standard. The present invention generates the optimal allocation scheme by comprehensively considering load balancing, energy consumption minimization and task priority through a multi-objective optimization model.

[0105] In one possible implementation, the objective function of the optimization model is as follows:

[0106] in: To optimize the target value; is the total number of servers in the intelligent computing center; For the Energy consumption of each server; For the The carbon emission intensity of the region where the server is located; For the The execution time of the current task on the server; To optimize the weight, satisfy .

[0107] By solving the above optimization objectives, the system can dynamically generate the optimal task allocation path, thereby balancing task execution efficiency and resource utilization.

[0108] In some embodiments, when the actual operation deviation exceeds a preset threshold, the system will automatically trigger the adjustment of the control strategy. Generally, the adjustment of the control strategy includes task migration, load balancing, and updating of energy efficiency benchmarks.

[0109] As an option, the task migration strategy can prioritize the migration of high-energy-consuming tasks from servers with higher loads to servers with lower loads. For example, when the load rate of a server exceeds 80%, the system will immediately trigger a migration operation to transfer some tasks to servers with a load rate of less than 50%.

[0110] In one possible implementation, the load balancing strategy can be optimized in combination with real-time energy efficiency benchmarks. For example, when the energy efficiency ratio of the cooling system decreases, the system will reduce the task allocation in the high-temperature area and prioritize the tasks to servers with higher cooling efficiency.

[0111] In addition, the control strategy can be adjusted dynamically based on the task priority. For example, for high-priority tasks, the system will prioritize their execution time, while for low-priority tasks, execution can be delayed to reduce current energy consumption pressure.

[0112] In some extended applications, the closed-loop feedback mechanism of the present invention can also be combined with the dynamic pricing system of the external power grid. When the electricity price fluctuates greatly, the system can dynamically adjust the task execution strategy according to the electricity price level. For example, during the period of low electricity price, the system can give priority to high energy consumption tasks, thereby reducing the overall operating cost.

[0113] In addition, in another possible implementation, the control strategy of the present invention can also be combined with the dynamic changes of regional energy supply. For example, when the proportion of renewable energy in a certain area is high, the system can preferentially migrate tasks to servers in that area to further reduce carbon emissions.

[0114] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for dynamically controlling the computing power of a green intelligent computing center based on carbon emission demand analysis, characterized in that: The following steps are involved: Data collection and analysis: Real-time acquisition of multi-dimensional data from the Green Intelligent Computing Center, including carbon emission intensity, server load, energy consumption, environmental conditions, and cooling system energy efficiency data, and pre-processing and analysis of the data; Dynamically generate energy efficiency control standards: Based on the data analysis results, generate energy efficiency control standards for the intelligent computing center, including server load thresholds, carbon emission intensity limits, and cooling system energy efficiency benchmarks; Implement dynamic control of computing power based on energy efficiency control standards: dynamically adjust the allocation path of computing tasks based on the real-time load of the server, energy consumption characteristics and carbon emission intensity; Computing power allocation optimization and energy efficiency feedback control: Through a closed-loop feedback mechanism, real-time operation data is collected, the actual operation results are compared with energy efficiency management standards, the computing power allocation path is optimized, and the control strategy is dynamically adjusted.

2. The method for dynamically controlling computing power of a green intelligent computing center based on carbon emission demand analysis according to claim 1 is characterized in that: The data collection and analysis include: Obtain carbon emission intensity data through sensors and energy interfaces; Obtain load rate and energy consumption data from the server; Obtain regional temperature and humidity data from environmental monitoring equipment; Get real-time energy efficiency data from cooling equipment; The collected data is cleaned and standardized, and a carbon emission prediction model is built using machine learning algorithms.

3. The method for dynamic control of computing power of a green intelligent computing center based on carbon emission demand analysis according to claim 1 is characterized in that: The dynamically generated energy efficiency control standards include: Generate carbon emission intensity limits based on carbon emission forecast models; Generate server load safety operation threshold based on load analysis; Generate cooling system energy efficiency benchmark based on cooling system operation data; Dynamically predict future carbon emission trends and adjust the energy efficiency control standards in real time.

4. The method for dynamic control of computing power of a green intelligent computing center based on carbon emission demand analysis according to claim 1 is characterized in that: The dynamic control of computing power includes: Establish a computing power allocation optimization model based on multi-dimensional factors, including server load, energy consumption, carbon emission intensity and environmental conditions; A task priority evaluation model is used to classify tasks according to their priority and computational complexity; Dynamically adjust the task allocation path to assign high-priority tasks to servers with low energy consumption or low carbon emission intensity.

5. The method for dynamic control of computing power of a green intelligent computing center based on carbon emission demand analysis according to claim 4 is characterized in that: The task priority evaluation model includes: Obtain key parameters of the task, including the timeliness of the task, computational complexity, and the task's demand for computing resources; A hierarchical model for task priority evaluation is constructed through the analytic hierarchy process, with task timeliness as the highest priority factor and computational complexity and computational resource requirements as secondary factors; Calculate the task priority scores and sort the tasks according to the scores; According to the task priority sorting, high-priority tasks are preferentially assigned to servers with low load and high energy efficiency, and the execution time of low-priority tasks is adjusted or assigned to low-energy consumption mode.

6. The method for dynamic control of computing power of a green intelligent computing center based on carbon emission demand analysis according to claim 1 is characterized in that: The computing power allocation optimization includes: By monitoring the server's load and energy consumption status in real time, when the load of a server exceeds the set threshold, task migration is triggered; Prioritize migrating high-energy-consuming tasks to servers with lower carbon emission intensity or under low load; Dynamically adjust the distribution of computing resources to achieve load balancing and energy efficiency optimization.

7. The method for dynamic control of computing power of a green intelligent computing center based on carbon emission demand analysis according to claim 1 is characterized in that: The closed-loop feedback mechanism includes: Real-time collection of the intelligent computing center's operating data, including task completion time, unit energy consumption, and actual carbon emissions; Compare actual operation data with energy efficiency control standards to evaluate the deviation of computing power regulation; Dynamically adjust energy efficiency control standards and computing power allocation strategies based on the evaluation results.

8. The method for dynamic control of computing power of a green intelligent computing center based on carbon emission demand analysis according to claim 1 is characterized in that: The dynamic adjustment and control strategy includes: When carbon emission intensity decreases, high energy consumption tasks are allocated to low carbon emission areas first; Delay the execution of low-priority tasks during peak load periods and adjust them to run during low-load or off-peak periods; Dynamically adjust server workloads based on changes in cooling system operating efficiency.

9. The method for dynamic control of computing power of a green intelligent computing center based on carbon emission demand analysis according to claim 1 is characterized in that: The implementation of dynamic control of computing power based on energy efficiency control standards: combining the real-time load of the server, energy consumption characteristics and carbon emission intensity, dynamically adjusting the allocation path of computing tasks, specifically including: The server load rate, real-time carbon emission intensity, ambient temperature and humidity data, and server power consumption are used as input variables; Use supervised learning algorithms to train the model and build a mapping relationship between carbon emissions and the above input variables based on historical operating data; Use the long short-term memory network to predict the carbon emission intensity in a specified time period in the future and generate a carbon emission intensity trend curve; Dynamically adjust the task allocation strategy based on the prediction results, including prioritizing high-energy-consuming tasks to low-carbon areas or off-peak hours.

10. The method for dynamic control of computing power of a green intelligent computing center based on carbon emission demand analysis according to claim 1 is characterized in that: The computing power allocation optimization and energy efficiency feedback control: through the closed-loop feedback mechanism, real-time collection of operating data, comparison of actual operating results with energy efficiency control standards, optimization of computing power allocation path and dynamic adjustment of control strategies, specifically including: Minimize the server's energy consumption per task; Minimize task execution time; Maximize overall server load balancing and cooling system energy efficiency.

Images