Intelligent dynamic energy regulation method and system for large-scale computing center

Abstract

This application discloses an intelligent dynamic energy regulation method and system for large-scale computing centers, relating to the field of computing center control. The method includes: acquiring server load information and server internal environment information of a target computing center; determining energy consumption prediction information of the target computing center based on the server load information, the server internal environment information, and an energy consumption prediction model; acquiring operation information and external environment information of a target power grid; determining energy consumption cost prediction information of the target power grid based on the operation information, the external environment information, and an energy consumption cost model; and adjusting the computing task waiting queue of the target computing center based on the energy consumption prediction information, the energy consumption cost prediction information, computing task urgency information, and computing task continuity information.

Description

Technical Field

[0001] This specification relates to the field of computing center control, and more specifically, this application relates to an intelligent dynamic energy regulation method and system for large-scale computing centers. Background Technology

[0002] Energy consumption is a critical issue in large-scale computing centers, especially given the fluctuating global energy prices and increasingly stringent environmental protection requirements. Reducing energy costs in computing centers has become an important research direction. Traditional computing centers typically complete computing tasks through static scheduling strategies and fixed energy supply methods. This approach fails to adequately consider real-time grid operation, external environmental changes, and the urgency and continuity of computing tasks, resulting in low energy efficiency and difficulty in coping with grid fluctuations or insufficient energy supply.

[0003] Furthermore, traditional methods lack real-time monitoring and feedback of server internal environment information, which leads to inaccurate energy consumption prediction when the server is working under high load or harsh environment, further increasing the complexity of energy management. Summary of the Invention

[0004] The summary section introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. This summary section is not intended to limit the key and essential technical features of the claimed technical solution, nor is it intended to determine the scope of protection of the claimed technical solution.

[0005] Firstly, this application proposes an intelligent dynamic energy regulation method for large-scale computing centers, the method comprising: Obtain server load information and internal server environment information of the target computing center; Based on the above server load information, the above server internal environment information, and the energy consumption prediction model, the energy consumption prediction information of the above target computing center is determined; Acquire operational information and external environmental information of the target power grid; Based on the above-mentioned target power grid operation information, the above-mentioned external environment information, and the energy consumption cost model, the energy consumption cost prediction information of the target power grid is determined; Based on the aforementioned energy consumption prediction information, energy cost prediction information, computational task urgency information, and computational task continuity information, the computational task waiting queue of the aforementioned target computing center is adjusted.

[0006] In one feasible implementation, the aforementioned server load information includes CPU utilization, memory utilization, disk I / O, network loan utilization, number and status of processes, and average load. The aforementioned internal server environment information includes server humidity information, server temperature information, power status information, and cooling fan speed information; The energy consumption prediction model includes a first input layer, a first feature extraction layer, a first algorithm core layer, and a first output layer. The first feature extraction layer includes a time-series feature extraction module and a frequency domain analysis module. The first algorithm core layer includes an ARIMA module and an LSTM module. The first output layer is used to output energy consumption prediction information, which includes short-term energy consumption prediction information, long-term energy consumption prediction information, and uncertainty assessment information.

[0007] In one feasible implementation, the aforementioned operational information based on the target power grid includes real-time power supply information, forecast information on other users' power demand, electricity price information, power supply stability information, and backup power information. The aforementioned external environmental information includes weather conditions, natural disaster early warning information, seasonal factors, and solar radiation intensity information; The aforementioned energy consumption cost model includes a second input layer, a second feature extraction layer, a second algorithm core layer, and a second output layer. The second feature extraction layer includes a time feature extraction module, an environmental feature extraction module, and a power supply and demand matching feature extraction module. The second algorithm core layer includes a cost prediction module and a risk assessment module. The cost prediction module is used to evaluate cost prediction information under normal circumstances, and the risk assessment module is used to simulate the distribution of energy consumption costs under external environment and power grid operation conditions to assess cost risks under different conditions and to assess the impact of uncertainties caused by power price fluctuations or supply fluctuations on overall costs.

[0008] In one feasible implementation, the aforementioned cost prediction module is based on linear regression model, LASSO regression model and seasonal ARIMA model, and the aforementioned risk assessment module is based on Monte Carlo method and Copula model.

[0009] In one feasible implementation, the adjustment of the computing task waiting queue of the target computing center based on the aforementioned energy consumption prediction information, energy cost prediction information, computing task urgency information, and computing task continuity information includes: Based on the above energy consumption prediction information, energy cost prediction information, computational task urgency information, and computational task continuity information, determine the total energy cost function, optimization objective function, time of minimizing objective time function, continuity constraint formula, priority weighted total cost function, task scheduling time function, task resource cost function, and optimization problem, and obtain the waiting queue of computational tasks to be adjusted. The waiting queue for computing tasks in the target computing center is adjusted based on the aforementioned waiting queue for computing tasks to be adjusted.

[0010] In one feasible implementation, the adjustment of the computing task waiting queue of the target computing center based on the aforementioned energy consumption prediction information, energy cost prediction information, computing task urgency information, and computing task continuity information includes: The total energy cost is calculated using the following formula. TC i (t) : ; Define the optimization objective function J(t) : ; ; ; Solving for the time point of minimizing the objective function : ; Define the continuity constraint formula: ; Define the priority-weighted total cost function J p (t) : ; ; Task scheduling time is obtained by solving the following formula: ; Task T i Resources allocated R j The cost function is: ; The optimal allocation strategy obtains the waiting queue of computational tasks to be adjusted by solving the following optimization problem: ; P i It is a task T i priority, E i Indicates task T i The predicted energy consumption value, C(t), represents the electricity cost when performing the task at time t. Ui Indicates task T i The urgency weight, L i Indicates task T i The continuity weight N represents the total number of tasks. x ij It is a task T i Resources allocated R j Binary decision variables.

[0011] In one feasible implementation, it further includes: If the energy consumption prediction information exceeds the energy prediction information provided by the target power grid, the task urgency information exceeds the preset threshold information, and the task continuity information exceeds the continuity threshold information, then determine the energy prediction difference information between the energy consumption prediction information and the energy prediction information provided by the target power grid. The operating power information of the hydrogen-powered UPS is determined based on the above energy prediction difference information.

[0012] In one feasible implementation, determining the operating power information of the hydrogen-powered UPS based on the aforementioned energy prediction difference information includes: Based on the operating power information of the aforementioned hydrogen energy UPS, the internal environmental information of the aforementioned server, and the external environmental information, the output parameters for controlling the solid hydrogen storage system using physical adsorption materials and the solid hydrogen storage system using chemical hydrogen storage materials are determined respectively.

[0013] In one feasible implementation, it further includes: When the internal temperature of the server is higher than the preset internal temperature, solid-state hydrogen storage systems using chemical hydrogen storage materials should be used first. When the internal temperature of the server is less than or equal to the above-mentioned preset internal temperature, the solid hydrogen storage system with physical adsorption materials should be used first. When the ambient temperature is higher than the preset ambient temperature, solid-state hydrogen storage systems using chemical hydrogen storage materials should be used preferentially. When the ambient temperature is less than or equal to the preset ambient temperature, solid hydrogen storage systems using physical adsorption materials should be used preferentially.

[0014] Secondly, this application proposes an intelligent dynamic energy regulation system for large-scale computing centers, comprising: The first acquisition unit is used to acquire server load information and server internal environment information of the target computing center. The first determining unit is used to determine the energy consumption prediction information of the target computing center based on the server load information, the server internal environment information and the energy consumption prediction model. The second acquisition unit is used to acquire the target power grid's operating information and external environment information; The second determining unit is used to determine the energy cost prediction information of the target power grid based on the operation information of the target power grid, the external environment information and the energy cost model. The adjustment unit is used to adjust the computing task waiting queue of the target computing center based on the aforementioned energy consumption prediction information, energy cost prediction information, computing task urgency information, and computing task continuity information.

[0015] In summary, the intelligent dynamic energy regulation method for large-scale computing centers according to embodiments of this application includes: acquiring server load information and internal environment information of the target computing center; determining energy consumption prediction information of the target computing center based on the server load information, the internal environment information, and an energy consumption prediction model; acquiring the operation information and external environment information of the target power grid; determining the energy consumption cost prediction information of the target power grid based on the operation information, the external environment information, and an energy consumption cost model; and adjusting the computing task waiting queue of the target computing center based on the energy consumption prediction information, the energy consumption cost prediction information, computing task urgency information, and computing task continuity information. The method proposed in this application, by acquiring server load information, internal environment information, power grid operation information, and external environment information in real time, enables the system to dynamically adjust the scheduling of computing tasks under different environmental conditions and power grid conditions, thereby optimizing energy use. This significantly improves the flexibility and accuracy of energy management. This solution utilizes energy consumption prediction models and energy consumption cost models to predict and identify energy consumption and cost information in different time periods, enabling the computing center to execute non-urgent tasks during low-cost periods and avoid high-cost periods, maximizing energy cost savings. The system comprehensively considers the urgency and continuity of tasks, prioritizing urgent tasks through intelligent scheduling algorithms to ensure efficient completion within appropriate timeframes and avoid conflicts and resource waste. By monitoring and adjusting the server's internal environment, the system prevents excessive energy consumption and instability under high load and harsh environments, thereby extending server lifespan and improving overall energy efficiency. In summary, this solution integrates multi-dimensional real-time information with intelligent predictive models to achieve comprehensive optimization of energy regulation in computing centers. It not only reduces energy costs but also improves task scheduling efficiency and computing resource utilization, providing crucial technical support for the sustainable development of large-scale computing centers.

[0016] The intelligent dynamic energy regulation method for large-scale computing centers proposed in this application, along with other advantages, objectives, and features of this application, will be partly apparent from the following description and partly understood by those skilled in the art through study and practice of this application. Attached Figure Description

[0017] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 A flowchart illustrating an intelligent dynamic energy regulation method for a large-scale computing center, provided as an embodiment of this application; Figure 2 This is a schematic diagram of an intelligent dynamic energy regulation system for a large-scale computing center, provided as an embodiment of this application. Detailed Implementation

[0018] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The technical solutions of the embodiments of this application will now be clearly and completely described in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.

[0019] Please see Figure 1 This is a flowchart illustrating a method for intelligent dynamic energy regulation in a large-scale computing center, as provided in an embodiment of this application. Specifically, it may include: Firstly, this application proposes an intelligent dynamic energy regulation method for large-scale computing centers, the method comprising: S110. Obtain server load information and server internal environment information of the target computing center; S120. Based on the above server load information, the above server internal environment information, and the energy consumption prediction model, determine the energy consumption prediction information of the above target computing center. S130. Obtain the target power grid's operational information and external environment information; S140. Based on the above-mentioned target power grid operation information, the above-mentioned external environment information, and the energy consumption cost model, determine the energy consumption cost prediction information of the target power grid; S150. Based on the above-mentioned energy consumption prediction information, energy cost prediction information, computing task urgency information, and computing task continuity information, adjust the computing task waiting queue of the above-mentioned target computing center.

[0020] For example, firstly, the system needs to acquire real-time server load information from the target computing center, including server CPU utilization, memory utilization, disk I / O, and network bandwidth utilization. This information reflects the server's current workload. Simultaneously, the system also acquires internal server environmental information, such as internal temperature, humidity, and voltage status. This environmental information affects the server's energy efficiency and operational stability.

[0021] After acquiring server load and internal environment information, the system uses a pre-defined energy consumption prediction model to predict the future energy consumption of the target computing center. This prediction model considers the impact of current load and environment on energy consumption, thus providing data support for subsequent energy management.

[0022] The system then acquires operational information about the target power grid, including data on power supply, electricity prices, and demand forecasts. Simultaneously, the system collects external environmental information, such as local weather conditions, temperature, humidity, and wind speed, which can affect power demand and supply.

[0023] The system uses an energy cost model, combined with grid operation information and external environmental information, to predict energy costs over different time periods. This model can calculate the cost of using electricity in the current or future period, thereby identifying high-cost and low-cost electricity consumption periods.

[0024] Finally, the system optimizes and adjusts the computing task waiting queue based on energy consumption prediction information, energy cost prediction information, computing task urgency information (the degree to which tasks need to be completed as soon as possible) and computing task continuity information (dependencies and continuity requirements between tasks).

[0025] The system will prioritize high-urgency tasks and ensure that computing tasks are completed during low-energy or low-cost periods, while avoiding high-cost periods for non-urgent tasks, thereby optimizing the energy efficiency of the computing center.

[0026] In summary, the method proposed in this application, by acquiring server load information, internal environment information, power grid operation information, and external environment information in real time, enables the system to dynamically adjust the scheduling of computing tasks under different environmental conditions and power grid statuses, thereby optimizing energy use. This significantly improves the flexibility and accuracy of energy management. This solution utilizes energy consumption prediction models and energy cost models to predict and identify energy consumption and cost information in different time periods, allowing the computing center to execute non-urgent tasks during low-cost periods and avoid high-cost periods, maximizing energy savings. The system comprehensively considers the urgency and continuity of tasks, prioritizing urgent tasks through intelligent scheduling algorithms to ensure tasks are completed efficiently within appropriate time periods, avoiding conflicts and resource waste between tasks. By monitoring and adjusting the server's internal environment, the system can prevent excessive energy consumption and operational instability under high load and harsh environments, thereby extending server lifespan and improving overall energy efficiency. In conclusion, this solution, by integrating multi-dimensional real-time information and intelligent prediction models, achieves comprehensive optimization of energy regulation in computing centers, not only reducing energy costs but also improving task scheduling efficiency and computing resource utilization, providing important technical support for the sustainable development of large-scale computing centers.

[0027] In some examples, the server load information mentioned above includes CPU utilization, memory utilization, disk I / O, network loan utilization, number and status of processes, and average load; The aforementioned internal server environment information includes server humidity information, server temperature information, power status information, and cooling fan speed information; The energy consumption prediction model includes a first input layer, a first feature extraction layer, a first algorithm core layer, and a first output layer. The first feature extraction layer includes a time-series feature extraction module and a frequency domain analysis module. The first algorithm core layer includes an ARIMA module and an LSTM module. The first output layer is used to output energy consumption prediction information, which includes short-term energy consumption prediction information, long-term energy consumption prediction information, and uncertainty assessment information.

[0028] For example, server load information includes several key metrics that reflect the server's operating status. Specifically, this information includes: CPU utilization: This indicates the usage of the server's processors and reflects the current computing task's demand for computing resources.

[0029] Memory usage: Displays the server's memory usage, indicating the memory consumption level of the current computing task.

[0030] Disk I / O: Monitor the frequency and speed of read and write operations on the server disk, which is closely related to the data access load.

[0031] Network bandwidth utilization: This indicates the usage of the server's network interface and reflects the current network data transmission load.

[0032] Process count and status: Counts the number of processes running on the server and their current status (e.g., running, suspended, etc.).

[0033] Average load: Calculates the average workload of the server over a period of time to help determine whether the server is overloaded.

[0034] Simultaneously, the system also acquires information about the server's internal environment. This information is used to assess the server's operating environment and adjust the energy consumption prediction model. Specific information includes: Server humidity information: Monitors the humidity level inside the server. Humidity that is too high or too low may affect the operation of the equipment.

[0035] Server temperature information: Monitor the internal temperature of the server. Temperature changes may affect energy consumption and heat dissipation efficiency.

[0036] Power status information: Monitors the voltage, current, and other statuses of the server power supply to ensure the stability of the power supply.

[0037] Cooling fan speed information: Monitor the speed of the server's cooling fans. The working status of the cooling fans directly affects the server's heat dissipation effect and energy consumption.

[0038] To predict the energy consumption of the target computing center, this proposal suggests an energy consumption prediction model. This model consists of the following layers: The first input layer receives server load information and internal server environment information. This input data forms the basis for the model's energy consumption prediction.

[0039] First feature extraction layer: Responsible for extracting key features from the input data, including: Time series feature extraction module: Analyzes the time series characteristics of the input data and extracts important features that change over time. These features are crucial for energy consumption prediction.

[0040] Frequency domain analysis module: Performs frequency domain analysis on the input data to capture periodic or frequency characteristics in server load and environmental information, which helps to identify patterns in energy consumption changes.

[0041] First core layer of the algorithm: This layer is the core part of the model, responsible for processing and predicting the extracted features, and contains two main modules: The ARIMA module is used for time-series analysis and short-term forecasting of server energy consumption, and is particularly suitable for linear and trending energy consumption data.

[0042] LSTM module: It uses a Long Short-Term Memory (LSTM) network to process energy consumption data. It is particularly good at handling nonlinear time series with long-term dependencies and can make more complex long-term energy consumption predictions.

[0043] First output layer: This layer is responsible for outputting the final prediction result, i.e., energy consumption prediction information. Specifically, it includes: Short-term energy consumption forecast information: Provides the server's energy consumption forecast results in the short term to help the system perform real-time energy management.

[0044] Long-term energy consumption forecast information: Provides the server's energy consumption forecast results over a longer period of time, which can be used to formulate long-term energy strategies.

[0045] Uncertainty assessment information: assesses the uncertainty of energy consumption forecast results, provides confidence intervals or risk analysis of the forecast results, so that possible biases or risks can be taken into account when making decisions.

[0046] This application's method predicts the energy consumption of a target computing center by acquiring server load and internal environment information and utilizing a multi-layered energy consumption prediction model. The prediction model includes an input layer, a feature extraction layer, an algorithm core layer, and an output layer. The temporal feature extraction module and frequency domain analysis module in the feature extraction layer are responsible for extracting key features, while the ARIMA and LSTM modules in the algorithm core layer perform short-term and long-term energy consumption predictions. Finally, the output layer provides energy consumption prediction information, including short-term predictions, long-term predictions, and uncertainty assessments, thereby providing data support for energy regulation and task scheduling in the computing center.

[0047] In some examples, the aforementioned operational information based on the target power grid includes real-time power supply information, forecasts of other users' power demand, electricity price information, power supply stability information, and backup power information; The aforementioned external environmental information includes weather conditions, natural disaster early warning information, seasonal factors, and solar radiation intensity information; The aforementioned energy consumption cost model includes a second input layer, a second feature extraction layer, a second algorithm core layer, and a second output layer. The second feature extraction layer includes a time feature extraction module, an environmental feature extraction module, and a power supply and demand matching feature extraction module. The second algorithm core layer includes a cost prediction module and a risk assessment module. The cost prediction module is used to evaluate cost prediction information under normal circumstances, and the risk assessment module is used to simulate the distribution of energy consumption costs under external environment and power grid operation conditions to assess cost risks under different conditions and to assess the impact of uncertainties caused by power price fluctuations or supply fluctuations on overall costs.

[0048] For example, the operational information of the target power grid includes the following key data: Real-time power supply information: Reflects the current power supply situation of the power grid and indicates the total power that the power grid can provide.

[0049] Other user electricity demand forecast information: used to predict the electricity demand of other users in the power grid, thereby determining the balance between electricity supply and demand.

[0050] Electricity price information: including real-time electricity prices or time-of-use prices, especially peak prices during periods of power shortage and off-peak prices during off-peak periods.

[0051] Power supply stability information: This includes voltage fluctuations and frequency stability of the power grid, reflecting the reliability and quality of the power supply.

[0052] Backup power information: This includes the status, capacity, and availability of backup power generation equipment in the power grid, in order to cope with power shortages or emergencies.

[0053] The system also collects external environmental information, which has a significant impact on power grid operation and energy cost prediction. Specifically, this includes: Weather conditions, such as temperature, humidity, wind speed, and precipitation, directly affect electricity demand and supply efficiency (e.g., air conditioning load and wind power generation efficiency).

[0054] Natural disaster early warning information: such as warnings of natural disasters such as typhoons and earthquakes. These events may have a significant impact on the operation of the power grid and may even lead to power outages.

[0055] Seasonal factors: Consider the impact of seasonal variations on electricity demand, such as peak electricity demand in summer and winter.

[0056] Solar radiation intensity information: This is especially important for the grid portion that relies on solar power generation, as solar radiation intensity determines the efficiency of photovoltaic power generation.

[0057] To predict the energy consumption cost of the target computing center, this proposal suggests an energy consumption cost model. This model consists of the following layers: The second input layer receives operational information from the target power grid and information from the external environment. This input data forms the basis for the model's cost prediction and risk assessment.

[0058] The second feature extraction layer is responsible for extracting key features from the input data, including: Time Feature Extraction Module: Analyzes the time series features of power grid operation information and environmental information to extract time-related change patterns, such as the periodic features of electricity price fluctuations and supply and demand changes.

[0059] Environmental Feature Extraction Module: Processes external environmental information and extracts environmental factors related to power supply and demand, such as the impact of temperature and wind speed on power demand.

[0060] Power supply and demand matching feature extraction module: Analyzes the power grid supply and demand balance, extracts the matching features between power supply and demand, and thus judges possible power shortage or surplus.

[0061] The second core layer of the algorithm: This layer is the core part of the model, responsible for processing and predicting the extracted features, and contains two main modules: Cost Prediction Module: This module evaluates cost prediction information under typical conditions. Based on extracted features, it calculates energy consumption costs over different time periods and outputs the prediction results.

[0062] Risk Assessment Module: This module simulates the distribution of energy costs under external environmental and grid operating conditions. It identifies and quantifies cost risks under different scenarios by assessing fluctuations in electricity prices or supply, and evaluates the uncertainty of overall energy costs.

[0063] The second output layer: This layer is responsible for outputting the final prediction result, namely the energy cost prediction information for the target computing center. Specifically, it includes: Typical cost forecasting information: Energy cost forecasting under normal grid operating conditions helps the computing center formulate daily energy management strategies.

[0064] Risk assessment information: Based on the risk assessment results of changes in the external environment and power grid fluctuations, it provides an analysis of energy consumption cost risks under the conditions of electricity price fluctuations or insufficient power supply.

[0065] This application embodiment predicts and evaluates the energy consumption cost of a target computing center by acquiring operational information of the target power grid and external environmental information, and utilizing a multi-layered energy cost model. The model includes an input layer, a feature extraction layer, an algorithm core layer, and an output layer. The feature extraction layer's time feature extraction module, environmental feature extraction module, and power supply and demand matching feature extraction module are responsible for extracting key features. The algorithm core layer's cost prediction module and risk assessment module predict and evaluate energy consumption costs under both normal and abnormal conditions. Finally, the output layer provides cost prediction information and risk assessment information under normal conditions, thereby providing data support for the computing center's energy regulation and task scheduling, and optimizing energy use.

[0066] In some examples, the cost prediction module is based on linear regression, LASSO regression, and seasonal ARIMA models, while the risk assessment module is based on Monte Carlo and Copula models.

[0067] For example, the cost prediction module is built based on the following models: Linear regression model: used to establish a linear relationship between electricity costs and input characteristics, suitable for cost prediction under simplified conditions.

[0068] LASSO regression model: Based on linear regression, a regularization term is introduced to select the most important features and avoid overfitting, making it suitable for cost prediction of high-dimensional feature data.

[0069] Seasonal ARIMA model: Used to analyze and predict time-series characteristics of electricity costs, especially suitable for handling electricity cost data with seasonal fluctuations. This model can capture trend and periodic changes in the data, thus providing more accurate cost forecasts.

[0070] Risk assessment module: In some examples, the risk assessment module is built based on the following methods: Monte Carlo method: This method assesses the volatility and uncertainty of electricity costs by simulating the distribution of electricity costs under different environmental conditions through extensive random sampling. It can simulate various possibilities under conditions of fluctuating electricity prices or insufficient electricity supply and calculate the probability distribution of these risks.

[0071] Copula model: Used to establish dependencies between different risk factors and assess cost risk under the combined effect of multiple risk factors through joint distribution. This model can more accurately capture the nonlinear dependencies between multiple variables, thus providing a more comprehensive risk assessment.

[0072] This solution acquires operational information from the target power grid and external environmental data, and utilizes a multi-layered energy cost model to predict and assess the energy consumption costs of the target computing center. The cost prediction module is built upon linear regression, LASSO regression, and seasonal ARIMA models, effectively predicting energy costs under normal conditions. The risk assessment module, based on the Monte Carlo method and the Copula model, simulates cost distributions under different environmental and power grid conditions, assessing the risks posed by electricity price fluctuations or power supply fluctuations. This prediction and assessment information provides data support for energy regulation and task scheduling in the computing center, thereby optimizing energy use and minimizing risks.

[0073] In some examples, the above-mentioned adjustments to the computing task waiting queue of the target computing center based on the aforementioned energy consumption prediction information, energy cost prediction information, computing task urgency information, and computing task continuity information include: Based on the above energy consumption prediction information, energy cost prediction information, computational task urgency information, and computational task continuity information, determine the total energy cost function, optimization objective function, time of minimizing objective time function, continuity constraint formula, priority weighted total cost function, task scheduling time function, task resource cost function, and optimization problem, and obtain the waiting queue of computational tasks to be adjusted. The waiting queue for computing tasks in the target computing center is adjusted based on the aforementioned waiting queue for computing tasks to be adjusted.

[0074] In some examples, the above-mentioned adjustments to the computing task waiting queue of the target computing center based on the aforementioned energy consumption prediction information, energy cost prediction information, computing task urgency information, and computing task continuity information include: The total energy cost is calculated using the following formula. TC i (t) : ; Define the optimization objective function J(t) : ; ; ; Solving for the time point of minimizing the objective function : ; Define the continuity constraint formula: ; Define the priority-weighted total cost function J p (t) : ; ; Task scheduling time is obtained by solving the following formula: ; Task T i Resources allocated R j The cost function is: ; The optimal allocation strategy obtains the waiting queue of computational tasks to be adjusted by solving the following optimization problem: ; Pi It is a task T i priority, E i Indicates task T i The predicted energy consumption value, C(t), represents the electricity cost when performing the task at time t. U i Indicates task T i The urgency weight, L i Indicates task T i The continuity weight N represents the total number of tasks. x ij It is a task T i Resources allocated R j Binary decision variables.

[0075] For example, energy consumption forecast information and energy cost forecast information are used to define the total energy cost function. This function measures the total energy cost required to perform all computational tasks, taking into account changes in energy prices and the energy requirements of the tasks over different time periods.

[0076] Based on the requirements of energy consumption cost and computational efficiency, an optimization objective function is defined. This function aims to minimize the total energy consumption cost while ensuring the efficiency and timely completion of the computational task.

[0077] Determine the optimal time to execute the task that minimizes total energy costs. This involves evaluating electricity costs and server energy efficiency over different time periods to find the lowest-cost execution time.

[0078] For tasks that need to be executed continuously or have specific time dependencies, define a continuity constraint formula. This ensures that these tasks are processed continuously within the necessary time window without interruption.

[0079] By considering the priorities of computational tasks, a priority-weighted total cost function is defined. This function optimizes the scheduling priority of important tasks by weighting tasks based on their urgency and importance.

[0080] Define a task scheduling time function. This function is used to determine the optimal start time for each task, taking into account the urgency, continuity requirements, and energy cost efficiency of the task.

[0081] Evaluate the cost of running each task on specific resources, including energy consumption and potential runtime latency. This function helps determine the optimal resource allocation strategy to minimize operating costs.

[0082] Based on the functions and constraints defined above, an optimization problem is formed. By solving this problem, the system can find the optimal task scheduling and resource allocation strategy to minimize costs and meet performance requirements.

[0083] Based on the solution to the optimization problem, a queue of computational tasks to be adjusted is generated. This queue is sorted according to task priority, predetermined start time, and resource allocation.

[0084] Based on the generated waiting queue of computing tasks to be adjusted, adjustments are made to the waiting queue of computing tasks in the target computing center. This includes updating task priorities, adjusting task start times, and reassigning tasks to different servers or computing resources.

[0085] By employing a series of sophisticated mathematical models and algorithms to optimize task scheduling in computing centers, the solution minimizes energy costs and maximizes computational efficiency. By integrating energy consumption prediction information, energy cost prediction information, and information on task urgency and continuity, the solution can dynamically adjust the task queue, thereby optimizing the use of computing resources and energy consumption. This approach not only improves the operational efficiency of computing centers but also helps reduce energy costs, making it significant for achieving green computing.

[0086] In some examples, it also includes: If the energy consumption prediction information exceeds the energy prediction information provided by the target power grid, the task urgency information exceeds the preset threshold information, and the task continuity information exceeds the continuity threshold information, then determine the energy prediction difference information between the energy consumption prediction information and the energy prediction information provided by the target power grid. The operating power information of the hydrogen-powered UPS is determined based on the above energy prediction difference information.

[0087] For example, when the predicted energy consumption exceeds the predicted energy provided by the target power grid, the difference in energy predictions between the two is first determined. This step involves calculating the difference between the total energy consumption expected by the target computing center and the energy predicted to be provided by the target power grid.

[0088] The energy consumption prediction information is calculated based on server load information, server internal environment information, and energy consumption prediction model, while the energy prediction information provided by the target power grid is estimated based on the current operating status of the power grid and external environment information.

[0089] When this difference occurs, it indicates that the supply from the target power grid is insufficient to meet the energy consumption needs of the computing center, especially when the urgency of the task exceeds the preset threshold and the continuity of the task reaches the continuity threshold.

[0090] Based on the calculated energy prediction difference, the operating power information of the hydrogen-powered UPS is then determined. A hydrogen-powered UPS (Uninterruptible Power Supply) is an uninterruptible power supply system that uses hydrogen energy as its energy source, providing necessary power when the grid supply is insufficient.

[0091] Determining the operating power information of a hydrogen-powered UPS involves calculating how much power the system needs to provide to make up for energy shortages. This calculation is based on the energy efficiency, storage capacity, and conversion capability of the hydrogen system to ensure the continued operation of the computing center when grid capacity is insufficient, especially for urgent and continuous computing tasks.

[0092] If the computing center's energy demand suddenly increases (potentially due to a large number of urgent and continuity-critical tasks being processed simultaneously), and the target power grid cannot provide sufficient power for various reasons (such as maintenance, weather conditions, or unstable supply), the hydrogen-powered UPS will activate as a critical backup system, providing the necessary additional power. This not only ensures task continuity and the operational efficiency of the computing center but also provides a reliable energy solution, reducing the risk of potential outages due to insufficient power supply.

[0093] The method provided in this application ensures the stable operation of computing centers when facing severe power supply challenges by intelligently integrating energy consumption forecasting and grid energy supply information, coupled with efficient backup power management. The introduction of hydrogen-powered UPS not only improves the flexibility of energy utilization but also enhances the system's adaptability to unstable external power supply conditions, representing an innovative application in the energy management strategy of modern large-scale computing centers.

[0094] In some examples, the above-mentioned determination of the operating power information of the hydrogen-powered UPS based on the energy prediction difference information includes: Based on the operating power information of the aforementioned hydrogen energy UPS, the internal environmental information of the aforementioned server, and the external environmental information, the output parameters for controlling the solid hydrogen storage system using physical adsorption materials and the solid hydrogen storage system using chemical hydrogen storage materials are determined respectively.

[0095] For example, Information about the server's internal environment includes its temperature, humidity, and cooling fan speed. This information directly impacts the operation of hydrogen energy systems, particularly in controlling the thermal management and safety of hydrogen storage systems.

[0096] External environmental information includes weather-related data such as temperature, humidity, and air pressure, all of which are important factors affecting the efficiency and safety of hydrogen energy systems. For example, changes in external temperature can affect the efficiency of hydrogen absorption and release.

[0097] Solid-state hydrogen storage systems using physical adsorption materials store hydrogen through physical adsorption. The output parameters of these systems, such as pressure and temperature, need to be adjusted based on the operating power information of the hydrogen UPS and environmental data to optimize the hydrogen release and absorption process.

[0098] Solid-state hydrogen storage systems using chemical hydrogen storage materials store and release hydrogen through chemical reactions. The output parameters of these systems, particularly those related to the chemical reaction rate and temperature, need to be precisely controlled according to the operational requirements and environmental conditions of the hydrogen-powered UPS to ensure a rapid supply of the required hydrogen during peak demand periods.

[0099] First, adjust the operating power of the hydrogen-powered UPS based on the energy prediction difference information to ensure sufficient power output to meet the needs of the computing center.

[0100] At the same time, the system monitors internal and external environmental information, which will be used to adjust the parameters of the hydrogen storage system in real time to ensure efficient storage and utilization of hydrogen.

[0101] For solid-state hydrogen storage systems using physical adsorption materials, the temperature and pressure parameters of the system are adjusted to optimize the adsorption and release efficiency of hydrogen, in order to meet the immediate energy needs of the computing center.

[0102] For solid-state hydrogen storage systems using chemical hydrogen storage materials, controlling the conditions of the chemical reaction, such as catalyst activity and reaction temperature, is crucial for the rapid release of hydrogen, especially when the power grid supply is unstable.

[0103] This application provides a flexible and reliable backup power solution for computing centers through a highly integrated hydrogen-powered UPS system, especially in emergencies when the power grid supply is insufficient. By precisely controlling the output parameters of the physical adsorption and chemical hydrogen storage systems, the solution not only optimizes the utilization efficiency of hydrogen energy but also enhances the stability and safety of the system under various environmental conditions. This approach effectively supports the continuous operation of the computing center, ensuring uninterrupted data processing and services.

[0104] In some examples, it also includes: When the internal temperature of the server is higher than the preset internal temperature, solid-state hydrogen storage systems using chemical hydrogen storage materials should be used first. When the internal temperature of the server is less than or equal to the above-mentioned preset internal temperature, the solid hydrogen storage system with physical adsorption materials should be used first. When the ambient temperature is higher than the preset ambient temperature, solid-state hydrogen storage systems using chemical hydrogen storage materials should be used preferentially. When the ambient temperature is less than or equal to the preset ambient temperature, solid hydrogen storage systems using physical adsorption materials should be used preferentially.

[0105] For example, if the server's internal temperature exceeds a preset threshold, the system will preferentially use a solid-state hydrogen storage system made of chemical hydrogen storage materials. This is because the performance of chemical hydrogen storage systems is not significantly affected at higher temperatures, and they can effectively release hydrogen.

[0106] When the ambient temperature exceeds the preset ambient temperature threshold, the solid-state hydrogen storage system using chemical hydrogen storage materials will be prioritized. Higher ambient temperatures are conducive to chemical reactions, thereby enhancing the system's output efficiency.

[0107] When the server's internal temperature is lower than or equal to the preset temperature, the system prioritizes using a solid-state hydrogen storage system with physical adsorption materials. Physical adsorption systems are more stable at lower temperatures and can more effectively adsorb and release hydrogen.

[0108] When the ambient temperature is lower than or equal to the preset ambient temperature threshold, the solid-state hydrogen storage system using physical adsorption materials will be used preferentially. At lower ambient temperatures, the efficiency of physical adsorption is improved, which helps maintain the continuous operation and stable output of the system.

[0109] This temperature-based system selection strategy not only improves the operational efficiency of the hydrogen energy system but also ensures the stability and security of the computing center's energy supply under various environmental conditions. This strategy allows the computing center to automatically adjust its energy management strategy in response to different temperature variations, thereby ensuring optimal energy use and reliable system operation. This approach is of great significance for improving the energy efficiency and environmental adaptability of large-scale computing centers.

[0110] Secondly, this application proposes an intelligent dynamic energy regulation system for large-scale computing centers, comprising: The first acquisition unit 21 is used to acquire server load information and server internal environment information of the target computing center; The first determining unit 22 is used to determine the energy consumption prediction information of the target computing center based on the server load information, the server internal environment information and the energy consumption prediction model. The second acquisition unit 23 is used to acquire the operation information and external environment information of the target power grid; The second determining unit 24 is used to determine the energy consumption cost prediction information of the target power grid based on the operation information of the target power grid, the external environment information and the energy consumption cost model. The adjustment unit 25 is used to adjust the computing task waiting queue of the target computing center based on the above-mentioned energy consumption prediction information, energy consumption cost prediction information, computing task urgency information and computing task continuity information.

[0111] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for intelligent dynamic energy conditioning of a large scale computing center, comprising: include: Obtain server load information and internal server environment information of the target computing center; Based on the server load information, the server internal environment information, and the energy consumption prediction model, the energy consumption prediction information of the target computing center is determined; Acquire operational information and external environmental information of the target power grid; Based on the operation information of the target power grid, the external environment information, and the energy consumption cost model, the energy consumption cost prediction information of the target power grid is determined; The waiting queue for computing tasks in the target computing center is adjusted based on the energy consumption prediction information, the energy cost prediction information, the urgency information of computing tasks, and the continuity information of computing tasks.

2. The intelligent dynamic energy regulation method for large-scale computing centers according to claim 1, characterized in that, The server load information includes CPU utilization, memory utilization, disk I / O, network loan utilization, number and status of processes, and average load. The server's internal environmental information includes server humidity information, server temperature information, power status information, and cooling fan speed information; The energy consumption prediction model includes a first input layer, a first feature extraction layer, a first algorithm core layer, and a first output layer. The first feature extraction layer includes a time-series feature extraction module and a frequency domain analysis module. The first algorithm core layer includes an ARIMA module and an LSTM module. The first output layer is used to output energy consumption prediction information, which includes short-term energy consumption prediction information, long-term energy consumption prediction information, and uncertainty assessment information.

3. The intelligent dynamic energy regulation method for large-scale computing centers according to claim 1, characterized in that, The operational information based on the target power grid includes real-time power supply information, forecast information on other users' power demand, electricity price information, power supply stability information, and backup power information. The external environmental information includes weather conditions, natural disaster early warning information, seasonal factors, and solar radiation intensity information; The energy consumption cost model includes a second input layer, a second feature extraction layer, a second algorithm core layer, and a second output layer. The second feature extraction layer includes a time feature extraction module, an environmental feature extraction module, and a power supply and demand matching feature extraction module. The second algorithm core layer includes a cost prediction module and a risk assessment module. The cost prediction module is used to evaluate cost prediction information under normal circumstances. The risk assessment module is used to simulate the distribution of energy consumption costs under external environment and power grid operation conditions to assess cost risks under different conditions and to assess the impact of uncertainties caused by power price fluctuations or supply fluctuations on overall costs.

4. The intelligent dynamic energy regulation method for large-scale computing centers according to claim 3, characterized in that, The cost prediction module is based on linear regression, LASSO regression and seasonal ARIMA models, while the risk assessment module is based on Monte Carlo method and Copula model.

5. The intelligent dynamic energy regulation method for large-scale computing centers according to claim 3, characterized in that, The adjustment of the computing task waiting queue of the target computing center based on the energy consumption prediction information, the energy cost prediction information, the computing task urgency information, and the computing task continuity information includes: Based on the energy consumption prediction information, the energy consumption cost prediction information, the computational task urgency information, and the computational task continuity information, determine the total energy consumption cost function, the optimization objective function, the time for minimizing the objective time function, the continuity constraint formula, the priority-weighted total cost function, the task scheduling time function, the task resource cost function, and the optimization problem, and obtain the waiting queue of computational tasks to be adjusted; The computing task waiting queue of the target computing center is adjusted according to the computing task waiting queue to be adjusted.

6. The intelligent dynamic energy regulation method for large-scale computing centers according to claim 3, characterized in that, The adjustment of the computing task waiting queue of the target computing center based on the energy consumption prediction information, the energy cost prediction information, the computing task urgency information, and the computing task continuity information includes: The total energy cost is calculated using the following formula. TC i (t) : ； Define the optimization objective function J(t) : ； ； ； Solving for the time point of minimizing the objective function : ； Define the continuity constraint formula: ； Define the priority-weighted total cost function J p (t) : ； ； Task scheduling time is obtained by solving the following formula: ； Task T i Resources allocated R j The cost function is: ； The optimal allocation strategy obtains the waiting queue of computational tasks to be adjusted by solving the following optimization problem: ； P i It is a task T i priority, E i Indicates task T i The predicted energy consumption value, C(t), represents the electricity cost when performing the task at time t. U i Indicates task T i The urgency weight, L i Indicates task T i The continuity weight N represents the total number of tasks. x ij It is a task T i Resources allocated R j Binary decision variables.

7. The intelligent dynamic energy regulation method for large-scale computing centers according to claim 1, characterized in that, Also includes: If the energy consumption prediction information exceeds the energy prediction information provided by the target power grid, the task urgency information exceeds the preset threshold information, and the task continuity information exceeds the continuity threshold information, then determine the energy prediction difference information between the energy consumption prediction information and the energy prediction information provided by the target power grid. The operating power information of the hydrogen-powered UPS is determined based on the energy prediction difference information.

8. The intelligent dynamic energy regulation method for large-scale computing centers according to claim 7, characterized in that, The process of determining the operating power information of the hydrogen-powered UPS based on the energy prediction difference information includes: Based on the operating power information of the hydrogen energy UPS, the internal environment information of the server, and the external environment information, the output parameters of the solid hydrogen storage system using physical adsorption materials and the solid hydrogen storage system using chemical hydrogen storage materials are determined respectively.

9. The intelligent dynamic energy regulation method for large-scale computing centers according to claim 8, characterized in that, Also includes: When the internal temperature of the server is higher than the preset internal temperature, solid-state hydrogen storage systems using chemical hydrogen storage materials should be used first. When the internal temperature of the server is less than or equal to the preset internal temperature, a solid hydrogen storage system using physical adsorption materials should be used preferentially. When the ambient temperature is higher than the preset ambient temperature, solid-state hydrogen storage systems using chemical hydrogen storage materials should be used preferentially. When the ambient temperature is less than or equal to the preset ambient temperature, solid hydrogen storage systems using physical adsorption materials should be used preferentially.

10. An intelligent dynamic energy regulation system for a large-scale computing center, characterized in that, include: The first acquisition unit is used to acquire server load information and server internal environment information of the target computing center. The first determining unit is used to determine the energy consumption prediction information of the target computing center based on the server load information, the server internal environment information and the energy consumption prediction model. The second acquisition unit is used to acquire the target power grid's operating information and external environment information; The second determining unit is used to determine the energy cost prediction information of the target power grid based on the operation information of the target power grid, the external environment information, and the energy cost model. The adjustment unit is used to adjust the computing task waiting queue of the target computing center based on the energy consumption prediction information, the energy cost prediction information, the computing task urgency information, and the computing task continuity information.

Images