Energy consumption optimal processor frequency prediction method and system based on support vector machine

Abstract

The application discloses a kind of energy consumption optimal processor frequency prediction method and system based on support vector machine, the present application includes the original data of performance monitoring counter value, running time, processor energy consumption, processor frequency and processor power under different parallelism, different scale, different processor frequency of different programs;For original data, construct data set and target vector according to the preset custom running time and energy consumption constraint condition;Design objective function;Parameter tuning and training evaluation are carried out to support vector machine model;Processor frequency prediction is carried out to target program using trained support vector machine model, and the processor frequency of optimal energy consumption is obtained.The present application aims to deal with the problem that the inappropriate processor frequency setting during program running leads to low energy consumption, through the energy consumption optimal processor frequency prediction based on support vector machine, ensure that application program can obtain its optimal processor frequency before loading, improve program running energy efficiency.

Description

Technical Field

[0001] This invention relates to the field of computer hardware and systems, and specifically to a method and system for predicting processor frequency with optimal energy consumption based on support vector machines. Background Technology

[0002] Processor frequency affects program execution time and power consumption. However, there is room for improvement in reducing power consumption and enhancing performance when running any application at its default frequency. While processor frequency can be adjusted within a certain range, exhaustively searching and comparing various options to find the appropriate frequency for a specific program is too costly and lacks versatility. Existing research indicates that developing processor frequency tuning strategies to find the optimal processor frequency configuration for specific applications is effective in optimizing performance and power consumption. Summary of the Invention

[0003] The technical problem to be solved by this invention is to provide a method and system for predicting the optimal processor frequency based on support vector machines, which addresses the above-mentioned problems in the prior art. This invention aims to solve the problem of low energy efficiency caused by inappropriate processor frequency settings during program execution. By predicting the optimal processor frequency based on support vector machines, the invention ensures that the application can obtain the optimal processor frequency for its operation before loading, thereby improving the energy efficiency of program operation.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0005] A method for predicting the frequency of a processor with optimal energy consumption based on support vector machines includes the following steps:

[0006] S1, respectively acquires the raw data of performance monitoring counter values, running time, processor power consumption, processor frequency and processor power of different programs under different parallelism, different scale and different processor frequency;

[0007] S2, combined with custom runtime and energy consumption constraints, constructs a dataset based on the acquired raw data. and target vector Design the objective function by combining custom constraints on runtime and energy consumption;

[0008] S3, based on the dataset Target vector And the objective function, to fine-tune the parameters of the support vector machine model, and complete the training and evaluation of the support vector machine model;

[0009] S4 uses the trained support vector machine model to predict the processor frequency of the target program, and obtains the energy-efficient processor frequency under the constraints of running time and power consumption.

[0010] Optionally, the raw data obtained in step S1 is of size [missing information]. matrix ,in This represents the total number of program samples. The matrix represents the number of Processor Performance Monitor (PMC) counter values, and 4 represents the four dimensions of the metrics: runtime, processor power consumption, processor frequency, and processor power. middle In the Processor Performance Monitoring Counter (PMC) value, any number of... OK Column elements represent the first The first program The average value of each performance monitoring counter, of which , .

[0011] Optionally, in step S2, the dataset is constructed. and target vector include:

[0012] S2.1, Traversing the matrix For each column of performance monitoring counter attributes, calculate the Pearson correlation coefficient between the performance monitoring counter value and runtime and processor power consumption, respectively. If the absolute values ​​of both Pearson correlation coefficients are greater than the threshold, add the attribute name of that column of performance monitoring counter attributes to the set of valid attributes. In the matrix Remove column attribute names that do not belong to the valid attribute set. The column;

[0013] S2.2, Traversing the matrix For each row of data samples, based on the data samples obtained through traversal If the data sample in this row This row of data samples can simultaneously meet the preset constraints on running time and energy consumption. Add to dataset , , This is the remaining total number of samples, which ultimately yields the dataset. ;

[0014] S2.3, for the obtained dataset , the dataset Programs that differ only in processor frequency but have the same program name, parallelism, and size are grouped together. Each group is iterated through, and the processor frequency with the lowest processor energy consumption value is selected as the optimal processor frequency for all samples in that group. Dataset All the best processor frequencies Construct the target vector Target vector The number of elements is Dataset For size is The matrix, where It is the remaining total number of samples. For a valid set of attributes The number of elements, 4 represents the dimensions of the four indicators: runtime, processor power consumption, processor frequency, and processor power.

[0015] Optionally, the functional expression for the custom runtime and energy consumption constraints in step S2 is:

[0016] ,

[0017] ,

[0018] ,

[0019] in, For processor frequency, For the set of feasible frequencies, , and For the minimum and maximum frequency values, For the program at frequency The processor power consumption obtained during operation For the program at frequency The program execution time obtained by running it below, For the program at the default frequency The program execution time obtained by running it below, For the program at the default frequency The processor power consumption obtained during operation A user-defined allowable loss coefficient for program runtime. A user-defined coefficient for reducing processor power consumption.

[0020] Optionally, the function expression of the objective function designed in step S2 is:

[0021] ,

[0022] in, For the weight vector, For bias, It is a positive slack variable. It is a negative relaxation change. for transpose, For the penalty function, For the sample size, It is a positive slack variable. It is a negative slack variable. and The penalty coefficient is... In order to process the first When there are 10 data samples, the support vector machine model The sample was processed to obtain Required execution time For the first One data sample, For support vector machine model The prediction function, For the program at the default frequency The program execution time obtained by running it below, The degree of time loss allowed for program execution as defined by the user. The degree of improvement in processor power consumption defined by the user. In order to process the first When there are 10 data samples, the support vector machine model The sample was processed to obtain Required energy consumption For the program at the default frequency The processor power consumption obtained during operation.

[0023] Optionally, when performing parameter tuning on the support vector machine model in step S3, the radial basis function kernel function used by the support vector machine model has the following functional expression:

[0024] ,

[0025] in, For radial basis kernel functions, and The first and the One data sample, The number of attributes, The weight of the k-th attribute. For the first The first data sample Each attribute value For the first The first data sample Each attribute value This is the bandwidth parameter corresponding to the k-th attribute.

[0026] Furthermore, the present invention also provides an optimal processor frequency prediction system based on support vector machines, comprising a microprocessor and a memory interconnected thereto, wherein the microprocessor is programmed or configured to execute the optimal processor frequency prediction method based on support vector machines.

[0027] Furthermore, the present invention also provides a computer-readable storage medium storing a computer program or instructions that are programmed or configured to execute the support vector machine-based optimal processor frequency prediction method by a processor.

[0028] Furthermore, the present invention also provides a computer program product, including a computer program or instructions that are programmed or configured to execute the support vector machine-based optimal processor frequency prediction method by a processor.

[0029] Compared with existing technologies, the present invention mainly achieves the following beneficial effects: The present invention first filters the set of valid performance monitoring counter values ​​to ensure a strong correlation between attribute values ​​and performance and power consumption, and then uses custom performance and power consumption constraints to filter out the original data that meets the conditions to construct a dataset. and target vector This invention achieves multi-objective optimization, ensuring program performance remains at a certain level while reducing power consumption. The support vector machine (SVM) model efficiently handles nonlinear relationships between high-dimensional data with low overhead. Furthermore, this invention addresses the problem of low energy efficiency caused by inappropriate processor frequency settings during program execution. By predicting the optimal processor frequency based on SVM, it ensures that the application obtains its optimal processor frequency before loading, thus improving program energy efficiency. Attached Figure Description

[0030] Figure 1 This is a schematic diagram of the basic process of the method in an embodiment of the present invention.

[0031] Figure 2 To construct a dataset in the embodiments of the present invention and target vector A flowchart. Detailed Implementation

[0032] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described in detail below with reference to the accompanying drawings in the embodiments of the present invention.

[0033] like Figure 1 As shown, the optimal processor frequency prediction method based on support vector machines in this embodiment includes the following steps:

[0034] S1, respectively acquires raw data of performance monitoring counter (PMC) values, running time, processor power consumption, processor frequency and processor power for different programs under different parallelism, different scale and different processor frequency;

[0035] S2, combined with custom runtime and energy consumption constraints, constructs a dataset based on the acquired raw data. and target vector Design the objective function by combining custom constraints on runtime and energy consumption;

[0036] S3, based on the dataset Target vector And the objective function, to fine-tune the parameters of the support vector machine model, and complete the training and evaluation of the support vector machine model;

[0037] S4 uses the trained support vector machine model to predict the processor frequency of the target program, and obtains the energy-efficient processor frequency under the constraints of running time and power consumption.

[0038] The raw data obtained in step S1 of this embodiment is of size [missing information]. matrix ,in This represents the total number of program samples. The matrix represents the number of Processor Performance Monitor (PMC) counter values, and 4 represents the four dimensions of the metrics: runtime, processor power consumption, processor frequency, and processor power. middle In the Processor Performance Monitoring Counter (PMC) value, any number of... OK Column elements represent the first The first program The average value of each performance monitoring counter, of which , In this embodiment, multiple sizes matrix The matrix formed As an optional implementation, in this embodiment, the multiple sets of data obtained in step S1 are three sets of data, namely three matrices. Construct a matrix .

[0039] Processor frequency, Processor Performance Monitor (PMC) value, runtime Processor power consumption The mathematical logical relationship between them is as follows:

[0040] ,

[0041] ,

[0042] in, and For mathematical logical relationships, ~ For n processor performance monitoring counter (PMC) values, Where n is the processor frequency, and n is the number of attributes that the performance monitoring counter can monitor, including branch instructions, branch misses, and bus cycles. As an optional implementation, in this embodiment, n = 128. n.

[0043] like Figure 2 As shown, in step S2 of this embodiment, the dataset is constructed. and target vector include:

[0044] S2.1, Traversing the matrix For each column of performance monitoring counter attributes, calculate the Pearson correlation coefficient between the performance monitoring counter value and runtime and processor power consumption, respectively. If the absolute values ​​of both Pearson correlation coefficients are greater than the threshold, add the attribute name of that column of performance monitoring counter attributes to the set of valid attributes. In the matrix Remove column attribute names that do not belong to the valid attribute set. The column;

[0045] S2.2, Traversing the matrix For each row of data samples, based on the data samples obtained through traversal If the data sample in this row This row of data samples can simultaneously meet the preset constraints on running time and energy consumption. Add to dataset , , This is the remaining total number of samples, which ultimately yields the dataset. ;

[0046] S2.3, for the obtained dataset , the dataset Programs that differ only in processor frequency but have the same program name, parallelism, and size are grouped together. Each group is iterated through, and the processor frequency with the lowest processor energy consumption value is selected as the optimal processor frequency for all samples in that group. Dataset All the best processor frequencies Construct the target vector Target vector The number of elements is Dataset For size is The matrix, where It is the remaining total number of samples. For a valid set of attributes The number of elements, 4 represents the dimensions of the four indicators: runtime, processor power consumption, processor frequency, and processor power.

[0047] The functional expression for the custom running time and energy consumption constraints in step S2 of this embodiment is as follows:

[0048] ,

[0049] ,

[0050] ,

[0051] in, For processor frequency, For the set of feasible frequencies, , and For the minimum and maximum frequency values, For the program at frequency The processor power consumption obtained during operation For the program at frequency The program execution time obtained by running it below, For the program at the default frequency The program execution time obtained by running it below, For the program at the default frequency The processor power consumption obtained during operation A user-defined allowable loss coefficient for program runtime. A user-defined coefficient for reducing processor power consumption.

[0052] The function expression of the objective function designed in step S2 of this embodiment is:

[0053] ,

[0054] in, For the weight vector, For bias, It is a positive slack variable. It is a negative relaxation change. for transpose, For the penalty function, For the sample size, It is a positive slack variable. It is a negative slack variable. and These are penalty coefficients (used to control the strength of runtime and energy consumption constraints, respectively. Therefore, additional penalty terms are generated when the predicted runtime or energy consumption violates the constraints). In order to process the first When there are 10 data samples, the support vector machine model The sample was processed to obtain Required execution time For the first One data sample, For support vector machine model The prediction function, For the program at the default frequency The program execution time obtained by running it below, The degree of time loss allowed for program execution as defined by the user. The degree of improvement in processor power consumption defined by the user. In order to process the first When there are 10 data samples, the support vector machine model The sample was processed to obtain Required energy consumption For the program at the default frequency The processor power consumption obtained during operation.

[0055] As an optional implementation, this embodiment designs a new radial basis function (RBF) kernel function based on the characteristics of the data to perform parameter tuning on the support vector machine (SVM) model and conduct a full search of the parameter grid to find the optimal parameter combination. Specifically, in step S3 of this embodiment, when tuning the parameters of the support vector machine model, the function expression of the radial basis function used by the support vector machine model is:

[0056] ,

[0057] in, For radial basis kernel functions, and The first and the One data sample, The number of attributes, The weight of the k-th attribute. For the first The first data sample Each attribute value For the first The first data sample Each attribute value Let be the bandwidth parameter corresponding to the k-th attribute. For comparison, the traditional radial basis function (RBF) kernel function is:

[0058] ,

[0059] in, These represent two sample vectors in the input space. It is bandwidth ( ), A larger value indicates a smoother kernel function and lower model complexity. Compared to the traditional Radial Basis Function (RBF), the new RBF kernel adds weights based on the importance differences of different features. By assigning different weights to each attribute while measuring the similarity between two samples, it enables rapid full search of the parameter grid to find the optimal parameter combination for Support Vector Machine (SVM) model parameter tuning.

[0060] In step S3 of this embodiment, based on the dataset Target vector And when tuning the parameters of the support vector machine model using the objective function, the parameters include the regularization parameters of the support vector machine model. kernel coefficient And the kernel function. In this embodiment, negative mean squared error is used as the scoring metric. The dataset is then subjected to 5-fold cross-validation. The dataset is divided into five subsets. Four subsets are used for training in rotation, and one subset is used for validation. This approach fully evaluates the model's performance on different datasets and ultimately finds the optimal parameter combination. In this embodiment, the regularization parameter... The value of kernel coefficient is generally [0.01, 0.1, 1, 10, 100, 1000]. The values ​​of are generally [0.001, 0.01, 0.1, 1, 10, 100], and the kernel function's value range is generally ['rbf', 'linear']. The process of training and evaluating a support vector machine model is as follows: Model learns from the dataset. The constructed feature matrix X and the set of effective attributes The mapping relationship between the target variable y is established. The trained model is used to predict the test set, and the mean squared error (MSE) between the predicted results and the true values ​​is calculated. Feature extraction and standardization are performed on the entire dataset, and then the saved optimal model is used to predict each sample, yielding the processor frequency value. Running time Processor power consumption For each predicted processor frequency value The nearest discrete frequency value *x* is found as the final predicted value and added to the dataset. The number of samples whose predicted values ​​match the true values ​​is counted, and the model's accuracy is calculated to evaluate its classification performance across the entire dataset. When using the model to directly predict the processor frequency of the target program, a non-dataset is selected. The data used is open-set prediction, which does not use the training and prediction sets from the model training process, to obtain the processor frequency of the target program. .

[0061] As an optional implementation, if the predicted processor frequency... Not in the preset set of feasible frequencies Within, the predicted processor frequency... A linear adjustment is performed to obtain the final optimal processor frequency, ensuring that the processor frequency remains within a preset set of feasible frequencies: The predicted processor frequency is then obtained. Corresponding program execution time ,in This is a function for fitting the program's runtime. The program determines the predicted processor frequency. Corresponding program execution time Violation of constraints:

[0062] ,

[0063] Whether it is valid, among which This is the default frequency. The user-defined allowable program runtime loss level is used, and if it holds true, the predicted processor frequency is adjusted according to the following formula. Perform linear adjustment of program runtime:

[0064] ,

[0065] in, The frequency after linear adjustment of program execution time. Adjust the program runtime coefficient; if this is not met, directly truncate the frequency. Frequency after linear adjustment of program execution time ; Obtain the frequency after linear adjustment of program runtime Corresponding processor power consumption ,in The function is used to fit the processor's energy consumption; the frequency after linear adjustment of the program's runtime is determined. Corresponding processor power consumption Violation of constraints:

[0066] ,

[0067] Whether it is valid, among which This is the default frequency. The degree of increase in processor power consumption defined by the user, if true, is used to linearly adjust the frequency of the program execution time according to the following formula. Perform linear adjustment of processor power consumption:

[0068] ,

[0069] in, This is the frequency after linearly adjusting the processor's energy consumption, which serves as the final optimal processor frequency. This is the processor power consumption adjustment factor.

[0070] In summary, this embodiment's method aims to deeply explore the intrinsic relationship between program performance and energy consumption. It comprehensively collects relevant data such as processor performance monitor counter (PMC) values, runtime, processor energy consumption, processor frequency, and processor power for different programs under various combinations of parallelism, scale, and processor frequency. Then, it sets custom performance and power consumption constraints based on actual needs, constructing a dataset that meets these conditions based on the collected data. Next, it designs a new radial basis function (RBF) kernel function based on the characteristics of the data, and uses a full search method to traverse the parameter grid, performing detailed parameter tuning on the support vector machine (SVM) model to determine the optimal parameter combination. Combining the energy consumption minimization objective with runtime and energy consumption constraints, a new objective function is designed. Finally, it uses the constructed dataset to train and evaluate the SVM model, directly predicting the processor frequency of the target program using the model, and then truncating and linearly adjusting the predicted frequency to obtain the final processor frequency. This allows this embodiment's method to overcome the problem of low energy efficiency caused by inappropriate processor frequency settings when applications are run multiple times.

[0071] Furthermore, this embodiment also provides an optimal processor frequency prediction system based on support vector machines, including a microprocessor and a memory interconnected, wherein the microprocessor is programmed or configured to execute the optimal processor frequency prediction method based on support vector machines.

[0072] Furthermore, this embodiment also provides a computer-readable storage medium storing a computer program or instructions that are programmed or configured to execute the support vector machine-based optimal processor frequency prediction method by a processor.

[0073] Furthermore, this embodiment also provides a computer program product, including a computer program or instructions, which are programmed or configured to execute the support vector machine-based optimal processor frequency prediction method via a processor.

[0074] Those skilled in the art will understand that the technical solutions provided by this invention may take the form of a method, system, or computer program product. Therefore, this invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this invention may take the form of a computer program product embodied on one or more computer-readable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, produce an implementation of the flowchart... Figure 1 One or more processes and / or boxes Figure 1 The computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable apparatus for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0075] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A support vector machine based energy optimal processor frequency prediction method, characterized in that, The method comprises the following steps: S1, obtaining raw data of performance monitoring counter values, running time, processor energy consumption, processor frequency and processor power of different programs under different parallelism, different scale, different processor frequency respectively; S2, combining the self-defined running time and energy consumption constraints, constructing a dataset according to the obtained raw data and target vectors ; designing a target function in combination with the self-defined running time and energy consumption constraints; S3, parameter tuning of the support vector machine model according to the data set , target vector and target function, training and evaluation of the support vector machine model are completed; S4, using the trained support vector machine model to predict the processor frequency of the target program to obtain the processor frequency with optimal energy consumption under the constraints of running time and energy consumption; The function expression of the self-defined running time and energy consumption constraint condition in step S2 is: ， ， ， in, For processor frequency, For the set of feasible frequencies, , and For the minimum and maximum frequency values, For the program at frequency The processor power consumption obtained during operation For the program at frequency The program execution time obtained by running it below, For the program at the default frequency The program execution time obtained by running it below, For the program at the default frequency The processor power consumption obtained during operation A user-defined allowable loss coefficient for program runtime. A user-defined coefficient for reducing processor power consumption; The function expression of the target function designed in step S2 is: ， in, For the weight vector, For bias, It is a positive slack variable. It is a negative relaxation change. for transpose, For the penalty function, For the sample size, It is a positive slack variable. It is a negative slack variable. and The penalty coefficient is... In order to process the first When there are 10 data samples, the support vector machine model The sample was processed to obtain Required execution time For the first One data sample, For Support Vector Machine Model The prediction function, For the program at the default frequency The program execution time obtained by running it below, The degree of time loss allowed for program execution as defined by the user. The degree of improvement in processor power consumption defined by the user. In order to process the first When there are 10 data samples, the support vector machine model The sample was processed to obtain Required energy consumption For the program at the default frequency The processor power consumption obtained during operation.

2. The support vector machine based energy optimal processor frequency prediction method of claim 1, wherein, The raw data obtained in step S1 is of size [size missing]. matrix ,in This represents the total number of program samples. The matrix represents the number of Processor Performance Monitor (PMC) counter values, and 4 represents the four dimensions of the metrics: runtime, processor power consumption, processor frequency, and processor power. middle In the Processor Performance Monitoring Counter (PMC) value, any number of... OK Column elements represent the first The first program The average value of each performance monitoring counter, of which , .

3. The support vector machine based energy optimal processor frequency prediction method of claim 2, wherein, Constructing the dataset in step S2 and target vectors comprises: S2.1, traversing the matrix , for each column performance monitoring counter attribute, calculating the Pearson correlation coefficient between the column performance monitoring counter value and the running time, the processor energy consumption respectively, if the absolute values of the two Pearson correlation coefficients are greater than a threshold, adding the attribute name of the column performance monitoring counter attribute to the valid attribute set ; deleting the columns with attribute names not belonging to the valid attribute set in the matrix ;​ S2.2, traversing each row of data samples in the matrix , and for each traversed data sample , if the data sample satisfies the preset time and energy consumption constraints, adding the data sample to the data set , , is the total number of remaining samples, and finally obtaining the data set ; S2.3, for the obtained dataset , the dataset , only the processor frequency is different, the program name, parallelism, and scale are the same, and a program is a group, each group is traversed, and the corresponding processor frequency with the minimum processor energy consumption value of each group of processors is selected as the optimal processor frequency of all samples in the group , the dataset , all the optimal processor frequencies constitute the target vector , the element number of the target vector ; the dataset is a matrix with a size of , wherein is the total number of remaining samples, is the element number of the effective attribute set , and 4 represents the dimension of the four indicators of running time, processor energy consumption, processor frequency, and processor power.​ 4. The support vector machine based energy optimal processor frequency prediction method of claim 1, wherein, When the support vector machine model is parameter-optimized in step S3, the function expression of the radial basis kernel function used by the support vector machine model is: ， wherein, is a radial basis kernel function, and are the first and the second data samples, is the number of attributes, is the weight of the k-th attribute, is the k-th attribute value of the i-th data sample, is the k-th attribute value of the i-th data sample, is the k-th attribute value of the i-th data sample, is the k-th attribute value of the i-th data sample, is the k-th attribute value of the i-th data sample, is the k-th attribute value of the i-th data sample, is the bandwidth parameter corresponding to the k-th attribute.

5. A support vector machine based energy optimal processor frequency prediction system comprising a microprocessor and a memory interconnected, characterized in that, The microprocessor is programmed or configured to perform the support vector machine-based optimal energy consumption processor frequency prediction method in any one of claims 1-4.

6. A computer-readable storage medium having stored therein a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to perform the support vector machine-based optimal energy consumption processor frequency prediction method in any one of claims 1-4 by the processor.

7. A computer program product comprising computer programs or instructions, characterized in that, The computer program or instructions are programmed or configured to perform the support vector machine-based optimal energy consumption processor frequency prediction method in any one of claims 1-4 by the processor.

Images