An Energy Consumption Optimization Method and System for a Computing Power Server

Abstract

The present invention discloses an energy consumption optimization method and system for a computing power server, which relates to the technical field of server optimization. The method includes: monitoring the operation data of the computing power server in real time to identify tasks and load conditions; constructing a load prediction model to predict the future load of the computing power server, and scheduling tasks according to the predicted load to optimize the energy consumption of the computing power server; continuously optimizing the load prediction model based on the reinforcement learning algorithm; displaying the load condition of the computing power server in real time, and storing the operation data of the computing power server in a database. By collecting the task processing and load data of the computing power server to construct a load prediction model to predict the future load, the present invention improves the accuracy of load prediction, and realizes the load optimization of the computing power server by scheduling task processing according to the predicted future load, effectively realizing the adjustment of the load energy consumption of the tasks processed by the computing power server.

Description

Technical Field

[0001] The present invention relates to the technical field of server optimization, and particularly to an energy consumption optimization method and system for computing power servers. Background Art

[0002] With the rapid development of information technology, computing power servers are increasingly widely used in all walks of life, especially in the fields of cloud computing, big data analysis, artificial intelligence, etc. These applications require a large amount of computing resources. This makes the energy efficiency problem of computing power servers a key problem to be solved urgently. Especially in the context of global energy tension and increasingly serious environmental pollution, how to efficiently utilize computing power resources and reduce energy waste has become an important topic in computer hardware design and data center management. At the same time, with the continuous increase in the load of computing power servers, how to effectively adjust the load to reduce energy consumption and improve energy efficiency has become a core direction for optimizing data center operations.

[0003] Currently, the energy efficiency optimization of computing power servers usually adopts technical means such as dynamic voltage regulation (DVFS), load balancing, and task scheduling. For example, the load balancing technology reduces energy waste by reasonably distributing the computing load of the server to avoid overload and idle. Dynamic voltage and frequency scaling (DVFS) optimizes power by adjusting the voltage and frequency of the server. In recent years, with the continuous progress of artificial intelligence technology, intelligent scheduling algorithms based on machine learning have gradually become an important means for optimizing the energy efficiency of computing power servers. These technologies analyze historical load data, predict future loads, and schedule tasks in advance, reducing unnecessary power consumption.

[0004] However, the existing technologies still have deficiencies. Traditional load prediction methods usually rely on linear regression or simple algorithms based on historical data, and it is difficult to handle the dynamic fluctuations and sudden tasks of server loads, resulting in low prediction accuracy, affecting the effects of task scheduling and power optimization, and being unable to make a quick response according to the real-time changing load requirements. This makes the computing power resources unable to be optimally utilized. Summary of the Invention

[0005] In view of the above existing problems, the present invention is proposed.

[0006] Therefore, the present invention provides an energy consumption optimization method and system for computing power servers, which solves the problems that the existing technologies are difficult to handle the dynamic fluctuations and sudden tasks of server loads and are unable to make a quick response according to the real-time changing load requirements.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] In the first aspect, the present invention provides an energy consumption optimization method for a computing power server, which includes,

[0009] Real-time monitor the operation data of the computing power server and identify tasks and load conditions;

[0010] Build a load prediction model to predict the future load of the computing power server, and schedule tasks according to the predicted load to optimize the energy consumption of the computing power server;

[0011] Continuously optimize the load prediction model based on the reinforcement learning algorithm;

[0012] Display the load condition of the computing power server in real time and store the operation data of the computing power server in the database.

[0013] As a preferred solution of the energy consumption optimization method of the computing power server described in the present invention, wherein: the real-time monitoring of the operation data of the computing power server and identifying tasks and load conditions means installing sensors on the computing nodes of the computing power server to monitor the operation data of the computing power server in real time, and identifying the tasks and load data on the computing power server through the operation data. The load data includes the load required for each task and the comprehensive operation load.

[0014] As a preferred solution of the energy consumption optimization method of the computing power server described in the present invention, wherein: the building of the load prediction model to predict the future load of the computing power server means normalizing the load data, building an LSTM model, setting the input of the LSTM model as the load data, the output as the preliminary predicted load data, and using the mean square error MSE as the original loss function of the LSTM model;

[0015] Synchronously build a support vector machine regression model, set the input of the support vector machine regression model as the preliminary predicted load data, the output as the final predicted load data, and build the loss function of the support vector machine regression model ;

[0016] The loss function of the support vector machine regression model Combine with the original loss function of the LSTM model to form an optimized LSTM model loss function :

[0017] where is the actual load data at time t, is the LSTM model predicted load data at time t, is the prediction duration, is the regularization coefficient, is the i-th slack variable, and n is the number of slack variables;

[0018] Based on the LSTM model training dataset, through the optimized LSTM model loss function Train the LSTM model and output preliminary predicted load data through the trained LSTM model;

[0019] Introduce the regularization term of linear regression into the loss function of the support vector machine regression model to form the loss function of the optimized support vector machine regression model :

[0020] where is the j-th linear regression coefficient, p is the number of linear regression coefficients, is the weight vector of the support vector machine regression model, is the regularization parameter;

[0021] Train the support vector machine regression model based on the training dataset of the support vector machine regression model through the loss function of the optimized support vector machine regression model and obtain the final predicted load data based on the trained support vector machine regression model.

[0022] As a preferred solution of the energy consumption optimization method of the computing power server described in the present invention, wherein: the energy consumption optimization of the computing power server according to the predicted load scheduling task means obtaining the predicted load of each task from the final predicted load data and calculating the priority of each task :

[0023] where is the processing time of the i-th task, is the predicted load of the i-th task, and are weight coefficients;

[0024] Sort the tasks according to the priority of each task and process the tasks in sequence according to the priority sorting. Optimize the predicted load of each task through sliding mode control:

[0025] where is the optimization function, is the adjustment coefficient of the i-th task, and m is the number of tasks;

[0026] Optimize to obtain the optimized predicted load of each task , record the actual load required when processing each task , and further optimize the predicted load of each task using the MPC optimization objective function:

[0027] where is the actual load of the task at time k + j, O is the transpose, is the optimized predicted load at time k+j, and R is the weighting matrix;

[0028] The predicted load of the quadratic optimization task is optimized through the MPC optimization objective function , and the load adjustment amount of the computing power server during task processing is calculated through the coordinated control formula :

[0029] where is the control gain, is a Lipschitz continuous function;

[0030] The load adjustment amount of the computing power server obtained through calculation is used to adjust the load of the computing power server in real time during task processing.

[0031] As a preferred solution of the energy consumption optimization method of the computing power server described in the present invention, wherein: the continuous optimization of the load prediction model based on the reinforcement learning algorithm refers to regularly statistically analyzing the accuracy rate of the load prediction model. When the model accuracy rate is lower than the set threshold, the Q-learning reinforcement learning algorithm is used to optimize and adjust the load prediction model until the model accuracy rate is greater than or equal to the set threshold.

[0032] As a preferred solution of the energy consumption optimization method of the computing power server described in the present invention, wherein: the real-time display of the load situation of the computing power server refers to the visual display of the adjusted load situation of the computing power server, and the task processing situation and the required load of the task are synchronously displayed.

[0033] As a preferred solution of the energy consumption optimization method of the computing power server described in the present invention, wherein: storing the operation data of the computing power server in the database refers to storing the collected operation data of the computing power server in the database. The database regularly detects the integrity and security of the stored data and uploads the stored data to the cloud for storage.

[0034] In a second aspect, the present invention provides an energy consumption optimization system for a computing power server, including

[0035] A data collection module, configured to detect the operation data of the computing power server and identify the processing tasks and load data;

[0036] An energy consumption optimization module, configured to construct a load prediction model to predict the load of the computing power server for processing tasks, and adjust the load of the computing power server according to the predicted load;

[0037] A reinforcement learning module, configured to continuously optimize the load prediction model using a reinforcement learning algorithm;

[0038] A display storage module is used to display the load condition of the computing power server in real time and store the collected operation data in a database.

[0039] In a third aspect, the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the energy consumption optimization method of the computing power server as described in the first aspect of the present invention is implemented.

[0040] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the energy consumption optimization method of the computing power server as described in the first aspect of the present invention is implemented.

[0041] The beneficial effects of the present invention are as follows: by collecting the task processing and load data of the computing power server to construct a load prediction model to predict the future load, the accuracy of load prediction is improved, and the load of the computing power server is optimized by predicting the future load for task processing scheduling, effectively realizing the adjustment of the load energy consumption of the task processed by the computing power server. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0043] Figure 1 It is a flowchart of the energy consumption optimization method of the computing power server in Embodiment 1.

[0044] Figure 2 It is a structural diagram of the energy consumption optimization system of the computing power server in Embodiment 1. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0045] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific embodiments of the present invention with reference to the drawings in the specification.

[0046] Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0047] Second, the "one embodiment" or "embodiment" referred to herein means a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that are mutually exclusive with other embodiments.

[0048] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides an energy consumption optimization method for a computing power server, including the following steps:

[0049] S1. Real-time monitor the operation data of the computing power server to identify tasks and load conditions;

[0050] Specifically, real-time monitoring of the operation data of the computing power server to identify tasks and load conditions means installing sensors on the computing nodes of the computing power server to real-time monitor the operation data of the computing power server, including CPU usage, GPU usage, memory usage, storage usage, and temperature data, and identifying the tasks and load data on the computing power server through the operation data. The load data includes the load required for each task and the comprehensive operation load.

[0051] Real-time monitoring of the operation data of the computing power server by installing sensors, and the collected CPU usage, GPU usage, memory usage, storage usage, and temperature data provide comprehensive basic data for subsequent load prediction and task scheduling. This is a feature not possessed by traditional energy efficiency optimization methods because these methods usually rely on static rules or historical data and fail to reflect the real-time changes of the system in a timely manner. Through real-time monitoring, the system can make dynamic adjustments according to the actual operation conditions of the server, improving the accuracy and real-time response ability of energy efficiency optimization. Identifying tasks and load conditions can accurately identify the resource requirements of the current tasks and the overall load conditions of the system based on the real-time collected data. By real-time monitoring the load requirements of tasks, accurate load prediction can be provided, thus making task scheduling and power management more precise.

[0052] S2. Build a load prediction model to predict the future load of the computing power server, and schedule tasks according to the predicted load to optimize the energy consumption of the computing power server;

[0053] Specifically, building a load prediction model to predict the future load of the computing power server means normalizing the load data and building an LSTM model, including an input layer, a hidden layer, and an output layer. Set the input of the LSTM model as the load data and the output as the preliminary predicted load data, and use the mean square error MSE as the original loss function of the LSTM model;

[0054] Synchronously construct a support vector machine regression model, set the input of the support vector machine regression model as the preliminary predicted load data, and the output as the final predicted load data, and construct the loss function of the support vector machine regression model :

[0055] where is the weight vector of the support vector machine regression model, is the regularization parameter, is the i-th slack variable, and n is the number of slack variables;

[0056] Combine the loss function of the support vector machine regression model with the original loss function of the LSTM model to form the optimized LSTM model loss function , and the introduced one is the loss function regularization term content:

[0057] where is the actual load data at time t, is the predicted load data of the LSTM model at time t, is the prediction duration, is the regularization coefficient, is the i-th slack variable, and n is the number of slack variables;

[0058] Based on the LSTM model training dataset, train the LSTM model through the optimized LSTM model loss function and output the preliminary predicted load data through the trained LSTM model;

[0059] Introduce the regularization term of linear regression into the loss function of the support vector machine regression model to form the optimized loss function of the support vector machine regression model :

[0060] where is the j-th linear regression coefficient, p is the number of linear regression coefficients, is the weight vector of the support vector machine regression model, is the regularization parameter;

[0061] Based on the support vector machine regression model training dataset, train the support vector machine regression model through the optimized loss function of the support vector machine regression model, and obtain the final predicted load data based on the trained support vector machine regression model.

[0062] By normalizing the load data, the dimensional differences between different tasks and load data are eliminated, enabling the model to better capture the patterns of load changes. The normalized data helps the LSTM model to perform more accurate time series predictions and reduce errors caused by data scale differences. The LSTM model itself can effectively handle the long-term dependencies in time series data, which is particularly suitable for predicting the load of computing power servers because server load usually has strong temporal characteristics. After the initial load prediction is output by the LSTM, the SVR model is used for optimization, which can not only further improve the prediction accuracy but also effectively handle the non-linear features in the load data. The LSTM is suitable for processing time series data, while the SVR can capture complex non-linear patterns in the data. The combination of the two makes the final load prediction more accurate and adaptable to complex and dynamically changing server load conditions. By combining the loss function of the SVR model with the original loss function of the LSTM model and introducing a regularization term, not only the training process of the LSTM model is optimized, but also overfitting of the model is prevented. The addition of the regularization term ensures that the model can balance the fitting accuracy and model complexity when facing complex load data, avoiding the problem of overfitting the training data. Introducing the regularization term of linear regression in the SVR model can effectively integrate linear and non-linear features, thereby further improving the prediction accuracy. The introduction of the linear regression coefficient enables the SVR to better balance the linear and non-linear relationships in the load data, and thus optimize the load prediction results.

[0063] Further, optimizing the energy consumption of the computing power server according to the predicted load scheduling task means obtaining the predicted load of each task from the final predicted load data and calculating the priority of each task. :

[0064] where is the processing time of the i-th task, obtained according to the final processing deadline of the task. is the predicted load of the i-th task. and are the weight coefficients;

[0065] Sort the tasks according to the priority of each task and process the tasks in sequence according to the priority sorting. Optimize the predicted load of each task through sliding mode control:

[0066] where is the optimization function. is the adjustment coefficient of the i-th task, and m is the number of tasks;

[0067] Obtain the optimized predicted load of each task , and record the actual load required when processing each task , the predicted load of each task is further optimized using the MPC optimization objective function:

[0068] where is the actual load of the task at time k + j, which can be obtained by linear fitting, O is the transpose, is the optimized predicted load at time k + j, and R are weighting matrices used to adjust the optimized predicted load;

[0069] The quadratic optimized task predicted load is obtained by optimizing the MPC optimization objective function , and the load adjustment amount of the computing power server during task processing is calculated through the coordinated control formula :

[0070] where is the control gain, is the Lipschitz continuous function;

[0071] The load adjustment amount of the computing power server obtained by calculation is used to adjust the load of the computing power server in real time during task processing.

[0072] By calculating the priority of each task and sorting them, the present invention ensures that when the computing power server performs task scheduling, it preferentially processes tasks that need to be executed urgently. The calculation of task priority not only considers the urgency of the task but also introduces predicted load data, which makes task scheduling more scientific and reasonable. Through the optimized priority sorting, the system can automatically allocate more resources to high-priority tasks under high load, avoiding task execution delays or resource waste. As a non-linear control method, sliding mode control can effectively handle the uncertainty in load prediction and the dynamic changes of the system. By using sliding mode control to optimize the predicted load of each task, the response speed of the system to load fluctuations can be improved, reducing the resource waste caused by load fluctuations. At the same time, sliding mode control can provide system stability, avoiding energy efficiency losses caused by excessive load adjustment. The MPC method further optimizes the predicted load of tasks by predicting future load changes. MPC dynamically adjusts the allocation of task loads by optimizing the objective function and combining real-time data, ensuring that task scheduling not only meets real-time requirements but also takes into account future load changes, thereby effectively improving resource utilization rate and energy efficiency. Through quadratic optimization, the present invention can make the load allocation of each task more accurate, avoiding resource waste caused by rough estimation. Through coordinated control, the system can comprehensively consider the mutual relationship between task scheduling and power management, further optimizing the load of the computing power server. Coordinated control can dynamically adjust the server load and task scheduling strategy through a feedback mechanism, enabling the system to respond to load changes in real time and maintain the optimal balance between task execution and resource use.

[0073] S3. Continuously optimize the load prediction model based on the reinforcement learning algorithm;

[0074] Specifically, continuously optimizing the load prediction model based on the reinforcement learning algorithm means regularly counting the accuracy rate of the load prediction model. When the accuracy rate of the model is lower than the set threshold, use the Q-learning reinforcement learning algorithm to optimize and adjust the load prediction model until the accuracy rate of the model is greater than or equal to the set threshold.

[0075] By regularly counting the accuracy rate of the model, the present invention can evaluate the performance of the load prediction model in real time, ensuring that the model always maintains high prediction ability. Different from the fixed optimization period or manual adjustment of the model in the prior art, the present invention can timely detect and correct potential biases of the model by regularly monitoring the accuracy rate. This enables the model to quickly adapt and continue to maintain prediction accuracy when facing long-term operation or changes in the load pattern. When the accuracy rate of the load prediction model is lower than the set threshold, use the Q-learning reinforcement learning algorithm to optimize and adjust the model until its accuracy rate reaches the set threshold. The innovation of this strategy lies in that Q-learning self-adjusts through interaction with the environment (i.e., feedback of load data), does not rely on manual intervention, and can continuously explore the optimal model parameters to optimize the prediction accuracy. Q-learning can adaptively adjust the prediction model according to the load characteristics of the current task, effectively coping with complex and dynamic changes in computing loads. The core of the Q-learning optimization process is to adjust the model through a reward and punishment mechanism, continuously improving the performance of the load prediction model. Specifically, Q-learning adjusts the model parameters according to the error feedback signal between the predicted load and the actual load, thereby achieving an improvement in the load prediction accuracy. This process makes the load prediction not just a static calculation based on historical data, but a dynamic optimization based on real-time feedback of the current task, making task scheduling and energy efficiency optimization more accurate.

[0076] S4. Display the load situation of the computing power server in real time and store the operation data of the computing power server in the database;

[0077] Specifically, displaying the load situation of the computing power server in real time means visually displaying the adjusted load situation of the computing power server, and synchronously displaying the task processing situation and the required load of the task.

[0078] Furthermore, storing the operation data of the computing power server in the database means storing the collected operation data of the computing power server in the database. The database regularly detects the integrity and security of the stored data and uploads the stored data to the cloud for storage.

[0079] This embodiment also provides an energy consumption optimization system for a computing power server, including:

[0080] A data collection module, configured to detect the operation data of the computing power server and identify processing tasks and load data;

[0081] An energy consumption optimization module, configured to build a load prediction model to predict the processing task load of the computing power server, and perform load adjustment on the computing power server according to the predicted load;

[0082] A reinforcement learning module, configured to continuously optimize the load prediction model using a reinforcement learning algorithm;

[0083] A display and storage module, configured to display the load condition of the computing power server in real time, and store the collected operation data in a database.

[0084] This embodiment also provides a computer device, applicable to the case of the energy consumption optimization method for a computing power server, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the energy consumption optimization method for the computing power server as proposed in the above embodiment.

[0085] This computer device may be a terminal. This computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used to communicate with an external terminal in a wired or wireless manner. The wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of this computer device may be a liquid crystal display screen or an electronic ink display screen. The input device of this computer device may be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the housing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0086] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the method for optimizing the energy consumption of a computing power server as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM for short), electrically erasable programmable read-only memory (EEPROM for short), erasable programmable read-only memory (EPROM for short), programmable read-only memory (PROM for short), read-only memory (ROM for short), magnetic memory, flash memory, a magnetic disk or an optical disc.

[0087] In summary, the present invention constructs a load prediction model by collecting the task and load data of the computing power server to predict the future load, improves the accuracy of load prediction, and realizes the load optimization of the computing power server by predicting the future load for task processing scheduling, effectively realizing the adjustment of the load energy consumption of the task processed by the computing power server.

[0088] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A method for optimizing energy consumption of a computing server, characterized in that: include, Monitor computing server operation data in real time and identify tasks and load conditions; Build a load prediction model to predict the future load of the computing server, and optimize the energy consumption of the computing server according to the predicted load scheduling tasks; Continuously optimize the load forecasting model based on reinforcement learning algorithm; Display the load of computing power servers in real time and store the running data of computing power servers in the database; The load prediction model includes an LSTM model and a support vector machine regression model. The loss function of the support vector machine regression model is Combined with the original loss function of the LSTM model to form an optimized LSTM model loss function , the loss function of the support vector machine regression model is The regularization term of linear regression is introduced to form the loss function of the optimized support vector machine regression model , set the LSTM model input as load data and output as preliminary predicted load data, set the support vector machine regression model input as preliminary predicted load data and output as final predicted load data, train the LSTM model and the support vector machine regression model, and obtain the final predicted load data based on the trained support vector machine regression model; Obtain the predicted load of each task from the final predicted load data, calculate the priority of each task, sort the tasks according to the priority of each task, and process the tasks in sequence according to the priority sorting. Optimize the predicted load of each task through sliding mode control to obtain the optimized predicted load of each task. , recording the actual load required to process each task , use the MPC optimization objective function to further optimize the predicted load of each task, and obtain the secondary optimization task predicted load through the MPC optimization objective function optimization , calculate the load adjustment of the computing server when processing tasks through the coordination control formula Real-time adjustment of the load of the computing server when processing tasks; The coordinated control formula is: ; in To control the gain, is a Lipschitz continuous function.

2. The method for optimizing energy consumption of a computing server according to claim 1, characterized in that: The real-time monitoring of the computing server's operating data and identification of tasks and load conditions refers to installing sensors on the computing nodes of the computing server to monitor the computing server's operating data in real time, and identifying tasks and load data on the computing server through the operating data, wherein the load data includes the load required for each task and the comprehensive operating load.

3. The method for optimizing energy consumption of a computing server according to claim 2, characterized in that: The construction of the load prediction model to predict the future load of the computing power server refers to standardizing the load data, building an LSTM model, and using the mean square error MSE as the original loss function of the LSTM model; Simultaneously build a support vector machine regression model and a loss function for the support vector machine regression model ; The loss function of the support vector machine regression model is Combined with the original loss function of the LSTM model to form an optimized LSTM model loss function : ; in is the actual load data at time t, Predict load data for the LSTM model at time t, To predict the duration, is the regularization coefficient, is the i-th slack variable, n is the number of slack variables; Based on the LSTM model training data set, the LSTM model loss function is optimized Train the LSTM model and output preliminary predicted load data through the trained LSTM model; The loss function in the support vector machine regression model The regularization term of linear regression is introduced to form the loss function of the optimized support vector machine regression model : ; in is the jth linear regression coefficient, p is the number of linear regression coefficients, is the weight vector of the support vector machine regression model, is the regularization parameter; Based on the support vector machine regression model training data set, the loss function of the support vector machine regression model is optimized The support vector machine regression model is trained, and the final predicted load data is obtained based on the trained support vector machine regression model.

4. The method for optimizing energy consumption of a computing power server according to claim 3, characterized in that: The optimization of computing server energy consumption according to the predicted load scheduling task refers to obtaining the predicted load of each task from the final predicted load data and calculating the priority of each task. : ; in is the processing time of the ith task, is the predicted load of the ith task, and is the weight coefficient; The tasks are sorted according to their priority, and the tasks are processed in sequence according to the priority. The predicted load of each task is optimized through sliding mode control: ; in To optimize the function, is the adjustment coefficient of the i-th task, and m is the number of tasks; Optimize to get the optimized predicted load for each task , recording the actual load required to process each task , the predicted load of each task is further optimized using the MPC optimization objective function: ; in is the actual load of the task at time k+j, O is the transposition, is the optimized predicted load at time k+j, and R is the weighting matrix; The load prediction of the secondary optimization task is obtained by optimizing the objective function of MPC optimization , calculate the load adjustment of the computing server when processing tasks through the coordination control formula ; The calculated load adjustment of the computing server Real-time adjustment of the load on the computing server when processing tasks.

5. The method for optimizing energy consumption of a computing server according to claim 4, characterized in that: The continuous optimization of the load prediction model based on the reinforcement learning algorithm refers to periodically collecting statistics on the accuracy of the load prediction model. When the model accuracy is lower than a set threshold, the load prediction model is optimized and adjusted using the Q-learning reinforcement learning algorithm until the model accuracy is greater than or equal to the set threshold.

6. The method for optimizing energy consumption of a computing server according to claim 5, characterized in that: The real-time display of the computing power server load situation refers to visually displaying the adjusted computing power server load situation, and synchronously displaying the task processing situation and the load required for the task.

7. The method for optimizing energy consumption of a computing server according to claim 6, characterized in that: Storing the computing power server operation data in a database refers to storing the collected computing power server operation data in a database, the database regularly performs integrity and security checks on the stored data, and uploads the stored data to the cloud for storage.

8. A system for optimizing energy consumption of a computing server, based on the method for optimizing energy consumption of a computing server according to any one of claims 1 to 7, characterized in that: include, Data collection module, used to detect computing server operation data and identify processing tasks and load data; The energy consumption optimization module is used to build a load prediction model to predict the task load of the computing server and adjust the load of the computing server according to the predicted load; Reinforcement learning module, used to continuously optimize the load forecasting model using reinforcement learning algorithm; The display storage module is used to display the load status of the computing server in real time and store the collected operating data in the database.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the energy consumption optimization method of the computing power server described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the energy consumption optimization method of the computing power server described in any one of claims 1 to 7 are implemented.

Images