Model training method and device, energy efficiency prediction method and device, medium and program product

Abstract

The invention discloses a model training method and device, an energy efficiency prediction method and device, a medium and a program product in the technical field of computers. According to the method, the model training data is constructed according to the partial configuration combination and the non-full-load energy efficiency test result of the partial configuration combination, the energy efficiency prediction model is obtained through training, the energy efficiency prediction model can be used for achieving server energy efficiency prediction, full-load testing does not need to be conducted on the server after server BIOS options are configured every time, computer resources and test time are saved, and the server energy efficiency prediction efficiency is improved. The precision of a test result can be improved by the energy efficiency prediction model; model training data for training the energy efficiency prediction model is obtained based on partial configuration combinations in a plurality of configuration combinations formed by to-be-tested configuration items of the basic input and output system and a non-full-load test, the selection of the partial configuration combinations and the non-full-load test can reduce the data processing amount, the model training efficiency is improved, the computer resources are saved, and the cost is reduced. And the server energy efficiency test efficiency and accuracy are further improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a model training, energy efficiency prediction method, device, medium and program product. Background Art

[0002] At present, server energy efficiency evaluation tools are generally used to test servers, which can measure server energy efficiency by comprehensively analyzing performance and energy consumption indicators. Using server energy efficiency evaluation tools requires full load testing of the server after configuring the server BIOS (Basic Input Output System) options each time, which takes a lot of time, and the accuracy of the test results depends on the accuracy of data collection.

[0003] Therefore, how to improve the efficiency and accuracy of server energy efficiency testing is a problem that technical personnel in this field need to solve. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a model training, energy efficiency prediction method, device, medium and program product to improve the efficiency and accuracy of server energy efficiency testing. The specific scheme is as follows:

[0005] In a first aspect, the present invention provides a model training method, which is applied to a test control system, comprising:

[0006] Determine multiple configuration combinations formed by configuration items to be tested of the basic input and output system;

[0007] Selecting some configuration combinations from the multiple configuration combinations, calling an energy efficiency test tool to perform a non-full load test on the selected configuration combinations, and collecting corresponding energy efficiency test results output by the energy efficiency test tool; the degree of similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold;

[0008] The selected configuration combinations and corresponding energy efficiency test results are constructed into the same data group to obtain multiple data groups;

[0009] Model training data is constructed based on adjacent data groups in the multiple data groups, and an energy efficiency prediction model is trained using the model training data.

[0010] Optionally, determining a plurality of configuration combinations formed by configuration items to be tested of the basic input / output system includes:

[0011] Determine each configuration item to be tested of the basic input / output system and the configuration value corresponding to each configuration item to be tested;

[0012] The configuration values ​​corresponding to the configuration items to be tested are respectively combined without duplication to obtain the multiple configuration combinations.

[0013] Optionally, selecting some configuration combinations from the multiple configuration combinations includes:

[0014] Selecting some configuration combinations from the multiple configuration combinations according to a preset ratio; the preset ratio is not less than fifty percent.

[0015] Optionally, selecting some configuration combinations from the multiple configuration combinations includes:

[0016] Sorting the multiple configuration combinations to obtain a target sequence;

[0017] Configuration combinations are selected at intervals in the target sequence.

[0018] Optionally, selecting configuration combinations at intervals in the target sequence includes:

[0019] The configuration combination arranged at the even-numbered positions in the target sequence is taken as the selected configuration combination; or the configuration combination arranged at the odd-numbered positions in the target sequence is taken as the selected configuration combination.

[0020] Optionally, call the energy efficiency test tool to perform a non-full load test on the selected configuration combination, including:

[0021] Determine, according to the task configuration file of the energy efficiency test tool, various test tasks corresponding to the full load test, and select some test tasks from the various test tasks;

[0022] The energy efficiency test tool is called to run the selected test task to perform a non-full load test on the selected configuration combination.

[0023] Optionally, some test tasks are selected from the various test tasks, including:

[0024] Some test tasks are selected from the various test tasks according to the non-full load test table preset in the task configuration file.

[0025] Optionally, some test tasks are selected from the various test tasks, including:

[0026] According to the random selection strategy configured in the task configuration file, randomly select a test task from the various test tasks as a task to be selected;

[0027] Calling the energy efficiency test tool to run the selected task to perform a non-full load test on the selected configuration combination, and collecting corresponding energy efficiency test results output by the energy efficiency test tool;

[0028] Selecting, from the obtained energy efficiency test results, an energy efficiency test result whose similarity to the energy efficiency test result of the full load test is greater than the preset threshold as an optional result;

[0029] The unselected task corresponding to the optional result is taken as the selected test task.

[0030] Optionally, the selected configuration combination and the corresponding energy efficiency test results are constructed into the same data group to obtain multiple data groups, including:

[0031] The selected configuration combination is combined with the corresponding energy efficiency test results to obtain corresponding multiple data groups.

[0032] Optionally, constructing model training data based on adjacent data groups among the multiple data groups includes:

[0033] Set the unit data size;

[0034] Determining the number of data groups included in the adjacent data group according to the unit data size;

[0035] A plurality of the model training data are constructed in the plurality of data groups with the number of the data groups as a unit.

[0036] Optionally, the number of data groups is 3;

[0037] Accordingly, in the plurality of data groups, a plurality of the model training data are constructed with the number of the data groups as a unit, including:

[0038] Every three adjacent data groups among the multiple data groups are spliced ​​in sequence.

[0039] Optionally, splicing every three adjacent data groups in the multiple data groups in sequence comprises:

[0040] Each data group in the multiple data groups is converted into a one-hot code, and the one-hot codes of every three adjacent data groups are concatenated in sequence.

[0041] Optionally, using the model training data to train an energy efficiency prediction model includes:

[0042] Inputting the model training data into a deep learning model so that the deep learning model outputs corresponding training energy efficiency results;

[0043] Calculating a loss value between the training energy efficiency result and a preset energy efficiency label of the model training data;

[0044] If the loss value meets the preset model convergence condition, the current deep learning model is used as the energy efficiency prediction model.

[0045] Optionally, if the loss value does not meet the model convergence condition, the model parameters of the current deep learning model are iteratively updated using the loss value until the loss value meets the model convergence condition.

[0046] Optionally, after the energy efficiency prediction model is obtained by training the model training data, the method further includes:

[0047] Model application data is constructed based on the multiple configuration combinations.

[0048] In a second aspect, the present invention provides an energy efficiency prediction method, comprising:

[0049] Determine multiple configuration combinations formed by configuration items to be tested of the basic input and output system;

[0050] Selecting some configuration combinations from the multiple configuration combinations, calling an energy efficiency test tool to perform a non-full load test on the selected configuration combinations, and collecting corresponding energy efficiency test results output by the energy efficiency test tool; the degree of similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold;

[0051] Using the energy efficiency prediction model, performing energy efficiency prediction on the configuration combinations for which there are no corresponding energy efficiency test results among the multiple configuration combinations, and collecting corresponding energy efficiency prediction results;

[0052] Wherein, the energy efficiency prediction model is obtained based on any of the model training methods described above.

[0053] Optionally, it also includes:

[0054] Summarizing and analyzing the multiple configuration combinations and their corresponding energy efficiency prediction results to determine the influence relationship between the configuration values ​​corresponding to the configuration items to be tested in the multiple configuration combinations and the corresponding energy efficiency prediction results;

[0055] The multiple configuration combinations and their corresponding energy efficiency prediction results are visualized using curves.

[0056] In a third aspect, the present invention provides an electronic device, comprising:

[0057] Memory for storing computer programs;

[0058] The processor is used to execute the computer program to implement the method disclosed above.

[0059] In a fourth aspect, the present invention provides a non-volatile storage medium for storing a computer program, wherein the computer program implements the aforementioned disclosed method when executed by a processor.

[0060] In a fifth aspect, the present invention provides a computer program product, comprising a computer program / instruction, which implements the steps of the corresponding method disclosed above when executed by a processor.

[0061] In a sixth aspect, the present invention provides a model training device, which is applied to test a control system, comprising:

[0062] A first determination module is used to determine a plurality of configuration combinations formed by various configuration items to be tested of the basic input and output system;

[0063] A first test module is used to select some configuration combinations from the multiple configuration combinations, call an energy efficiency test tool to perform a non-full load test on the selected configuration combinations, and collect corresponding energy efficiency test results output by the energy efficiency test tool; the similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold;

[0064] A construction module, used to construct the selected configuration combination and the corresponding energy efficiency test results into the same data group to obtain multiple data groups;

[0065] The training module is used to construct model training data based on adjacent data groups in the multiple data groups, and use the model training data to train an energy efficiency prediction model.

[0066] Optionally, the determination module is specifically used for:

[0067] Determine each configuration item to be tested of the basic input / output system and the configuration value corresponding to each configuration item to be tested;

[0068] The configuration values ​​corresponding to the configuration items to be tested are respectively combined without duplication to obtain the multiple configuration combinations.

[0069] Optionally, the test module is specifically used to:

[0070] Selecting some configuration combinations from the multiple configuration combinations according to a preset ratio; the preset ratio is not less than fifty percent.

[0071] Optionally, the test module is specifically used to:

[0072] Sorting the multiple configuration combinations to obtain a target sequence;

[0073] Configuration combinations are selected at intervals in the target sequence.

[0074] Optionally, the test module is specifically used to:

[0075] The configuration combination arranged at the even-numbered positions in the target sequence is taken as the selected configuration combination; or the configuration combination arranged at the odd-numbered positions in the target sequence is taken as the selected configuration combination.

[0076] Optionally, the test module is specifically used to:

[0077] Determine, according to the task configuration file of the energy efficiency test tool, various test tasks corresponding to the full load test, and select some test tasks from the various test tasks;

[0078] The energy efficiency test tool is called to run the selected test task to perform a non-full load test on the selected configuration combination.

[0079] Optionally, the test module is specifically used to:

[0080] Some test tasks are selected from the various test tasks according to the non-full load test table preset in the task configuration file.

[0081] Optionally, the test module is specifically used to:

[0082] According to the random selection strategy configured in the task configuration file, randomly select a test task from the various test tasks as a task to be selected;

[0083] Calling the energy efficiency test tool to run the selected task to perform a non-full load test on the selected configuration combination, and collecting corresponding energy efficiency test results output by the energy efficiency test tool;

[0084] Selecting, from the obtained energy efficiency test results, an energy efficiency test result whose similarity to the energy efficiency test result of the full load test is greater than the preset threshold as an optional result;

[0085] The unselected task corresponding to the optional result is taken as the selected test task.

[0086] Optionally, the building block is specifically used to:

[0087] The selected configuration combination is combined with the corresponding energy efficiency test results to obtain corresponding multiple data groups.

[0088] Optionally, the building block is specifically used to:

[0089] Set the unit data size;

[0090] Determining the number of data groups included in the adjacent data group according to the unit data size;

[0091] A plurality of the model training data are constructed in the plurality of data groups with the number of the data groups as a unit.

[0092] Optionally, the number of data groups is 3; accordingly, the construction module is specifically used for:

[0093] Every three adjacent data groups among the multiple data groups are spliced ​​in sequence.

[0094] Optionally, the building block is specifically used to:

[0095] Each data group in the multiple data groups is converted into a one-hot code, and the one-hot codes of every three adjacent data groups are concatenated in sequence.

[0096] Optionally, the training module is specifically used for:

[0097] Inputting the model training data into a deep learning model so that the deep learning model outputs corresponding training energy efficiency results;

[0098] Calculating a loss value between the training energy efficiency result and a preset energy efficiency label of the model training data;

[0099] If the loss value meets the preset model convergence condition, the current deep learning model is used as the energy efficiency prediction model.

[0100] Optionally, the training module is specifically used to: if the loss value does not meet the model convergence condition, use the loss value to iteratively update the model parameters of the current deep learning model until the loss value meets the model convergence condition.

[0101] Optionally, it also includes:

[0102] Another construction module is used to obtain model application data based on the multiple configuration combinations after the energy efficiency prediction model is trained using the model training data.

[0103] In a seventh aspect, the present invention provides an energy efficiency prediction device, comprising:

[0104] A second determination module is used to determine a plurality of configuration combinations formed by various configuration items to be tested of the basic input and output system;

[0105] A second test module is used to select some configuration combinations from the multiple configuration combinations, call the energy efficiency test tool to perform a non-full load test on the selected configuration combinations, and collect corresponding energy efficiency test results output by the energy efficiency test tool; the similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold;

[0106] A prediction module, configured to use an energy efficiency prediction model to perform energy efficiency prediction on configuration combinations among the multiple configuration combinations for which there are no corresponding energy efficiency test results, and collect corresponding energy efficiency prediction results;

[0107] Wherein, the energy efficiency prediction model is obtained based on any of the model training methods described above.

[0108] Optionally, it also includes:

[0109] An analysis module, used to summarize and analyze the multiple configuration combinations and their corresponding energy efficiency prediction results, so as to determine the influence relationship between the configuration values ​​corresponding to the configuration items to be tested in the multiple configuration combinations and the corresponding energy efficiency prediction results;

[0110] The display module is used to visualize the multiple configuration combinations and their corresponding energy efficiency prediction results using curves.

[0111] It can be seen from the above scheme that the present invention provides a model training method, including: determining multiple configuration combinations formed by various configuration items to be tested of the basic input and output system; selecting some configuration combinations from the multiple configuration combinations, calling the energy efficiency test tool to perform non-full load testing on the selected configuration combinations, and collecting the corresponding energy efficiency test results output by the energy efficiency test tool; the similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold; constructing the selected configuration combination and the corresponding energy efficiency test results into the same data group to obtain multiple data groups; constructing model training data based on adjacent data groups in the multiple data groups, and using the model training data to train an energy efficiency prediction model.

[0112] It can be seen that the present invention constructs model training data with partial configuration combinations and their non-full load energy efficiency test results, and trains to obtain an energy efficiency prediction model. This energy efficiency prediction model can be used to realize server energy efficiency prediction without the need to perform a full load test on the server after each configuration of the server BIOS option, thereby saving computer resources and testing time, and the energy efficiency prediction model can improve the accuracy of the test results; the model training data for training the energy efficiency prediction model is obtained based on partial configuration combinations and non-full load tests in multiple configuration combinations formed by each configuration item to be tested of the basic input and output system. The selection of partial configuration combinations and non-full load tests can reduce the amount of data processing, improve the model training efficiency, save computer resources, and further improve the efficiency and accuracy of server energy efficiency testing.

[0113] Correspondingly, a model training, energy efficiency prediction device, equipment, medium and program product provided by the present invention also have the above-mentioned technical effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0114] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0115] Figure 1 A flow chart of a model training method disclosed in the present invention;

[0116] Figure 2 A flow chart of an energy efficiency prediction method disclosed in the present invention;

[0117] Figure 3 A schematic diagram of an energy efficiency test scheme disclosed in the present invention;

[0118] Figure 4 A schematic diagram of energy efficiency prediction using a model disclosed in the present invention;

[0119] Figure 5 A schematic diagram of a BIOS combination and a corresponding energy efficiency curve disclosed in the present invention;

[0120] Figure 6 A schematic diagram of the structure of an energy efficiency prediction model disclosed in the present invention;

[0121] Figure 7 A schematic diagram of an electronic device disclosed in the present invention;

[0122] Figure 8 A server structure diagram provided by the present invention;

[0123] Fig. 9 A terminal structure diagram provided by the present invention. DETAILED DESCRIPTION

[0124] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other examples obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0125] At present, server energy efficiency evaluation tools are generally used to test servers, which can measure server energy efficiency by comprehensively analyzing performance and energy consumption indicators. Using server energy efficiency evaluation tools requires full load testing of the server after configuring the server BIOS options each time, which takes a lot of time, and the accuracy of the test results depends on the accuracy of data collection. To this end, the present invention provides a model training and energy efficiency prediction scheme, which can improve the efficiency and accuracy of server energy efficiency testing.

[0126] See also Figure 1 As shown, an embodiment of the present invention discloses a model training method, which is applied to test a control system, comprising:

[0127] S101, determining a plurality of configuration combinations formed by various configuration items to be tested of a basic input / output system.

[0128] It should be noted that the test control system includes: a program implementing the solution of this embodiment, which can interact with the basic input and output system and the energy efficiency test tool, for example: reading the BIOS options and configuration values ​​of the basic input and output system, calling the energy efficiency test tool, etc.

[0129] In one example, each configuration item to be tested of the basic input-output system may include: Enabled, MinFrequency, Max Frequency, and Custom; the configuration values ​​corresponding to these four configuration items may be Disabled or Enabled, then each configuration item to be tested and its corresponding configuration value may include: Enabled_Disabled, Enabled_Enabled, Min Frequency_Disabled, Min Frequency_Enabled, Max Frequency_Disabled, Max Frequency_Enabled, Custom_Disabled, and Custom_Enabled, and a total of 8 configuration combinations may be obtained. Therefore, in one embodiment, determining multiple configuration combinations formed by each configuration item to be tested of the basic input-output system includes: determining each configuration item to be tested of the basic input-output system and the configuration values ​​corresponding to each configuration item to be tested; performing non-repetitive combinations on the configuration values ​​corresponding to each configuration item to be tested to obtain multiple configuration combinations.

[0130] In one embodiment, selecting some configuration combinations from the plurality of configuration combinations includes: selecting some configuration combinations from the plurality of configuration combinations according to a preset ratio; the preset ratio is not less than 50%. For example: randomly selecting 4 configuration combinations from the 8 configuration combinations in the above example.

[0131] In one embodiment, selecting some configuration combinations from a plurality of configuration combinations includes: sorting the plurality of configuration combinations to obtain a target sequence; and selecting configuration combinations at intervals in the target sequence. Among them, selecting configuration combinations at intervals in the target sequence includes: selecting configuration combinations arranged at even positions in the target sequence as the selected configuration combinations; or selecting configuration combinations arranged at odd positions in the target sequence as the selected configuration combinations. For example: after arranging the 8 configuration combinations in the above example, select configuration combinations arranged at positions 1, 3, 5, and 7 or configuration combinations arranged at positions 2, 4, 6, and 8.

[0132] S102. Select some configuration combinations from multiple configuration combinations, call an energy efficiency test tool to perform a non-full load test on the selected configuration combinations, and collect corresponding energy efficiency test results output by the energy efficiency test tool; the similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold.

[0133] In this embodiment, full load testing means that during the test process, all test tasks corresponding to the full load test are executed by the server at the same time; correspondingly, non-full load testing means that during the test process, some test tasks corresponding to the full load test are executed by the server at the same time, while other test tasks do not participate in the test process. For example, if the full load test corresponds to 11 test tasks in total, if all these 11 test tasks participate in the test, it is a full load test; otherwise, it is a non-full load test.

[0134] Among them, the test process can be implemented with the help of energy efficiency evaluation tools. The specific process is: change the BIOS configuration from options such as the BIOS panel and the energy efficiency evaluation tool, and select the parameter configuration with the best performance-to-power ratio under all possible parameter configurations; the performance-to-power ratio is calculated based on 11 different workloads (100%, 90%, 80%...10%, idle, and the system utilization rate also decreases in turn). First, the performance calculation under all loads is summed ∑ssj_ops. Secondly, the power meter connected to the system power supply will record the system power status power in real time, calculate the power consumption of all loads divided by ∑power, and finally, the system will divide the accumulated performance and power to obtain the performance-to-power ratio ∑ssj_ops divided by ∑power=Performance to Power Ratio as the performance-to-power ratio of the server.

[0135] In one implementation, calling the energy efficiency test tool to perform a non-full load test on the selected configuration combination includes: determining the various test tasks corresponding to the full load test according to the task configuration file of the energy efficiency test tool, and selecting some test tasks from the various test tasks; calling the energy efficiency test tool to run the selected test tasks to perform a non-full load test on the selected configuration combination. The task configuration file can be read and written through the task configuration function of the energy efficiency test tool. Specifically, the user can modify the task configuration file on the management page of the energy efficiency test tool, and the non-full load test table, task selection strategy and other information can be written into the file.

[0136] In one embodiment, selecting some test tasks from the various test tasks includes: selecting some test tasks from the various test tasks according to a non-full load test table preset in the task configuration file. The preset non-full load test table can record in advance: M test tasks whose similarity to the energy efficiency test results of the full load test is greater than a preset threshold, where M is less than the total number of all test tasks corresponding to the full load test.

[0137] In one embodiment, some test tasks are selected from the various test tasks, including: randomly selecting test tasks from the various test tasks as tasks to be selected according to a random selection strategy (a task selection strategy) configured in the task configuration file; calling the energy efficiency test tool to run the tasks to be selected to perform a non-full load test on the selected configuration combination, and collecting corresponding energy efficiency test results output by the energy efficiency test tool; selecting, from the obtained energy efficiency test results, energy efficiency test results whose degree of similarity with the energy efficiency test results of the full load test is greater than the preset threshold as optional results; and using the tasks to be selected corresponding to the optional results as selected test tasks.

[0138] S103: construct the selected configuration combination and the corresponding energy efficiency test results into the same data group to obtain multiple data groups.

[0139] In one implementation, the selected configuration combination and the corresponding energy efficiency test results are constructed into the same data group to obtain multiple data groups, including: splicing the selected configuration combination and the corresponding energy efficiency test results to obtain corresponding multiple data groups.

[0140] S104: constructing model training data based on adjacent data groups in the multiple data groups, and training an energy efficiency prediction model using the model training data.

[0141] In one embodiment, model training data is constructed based on adjacent data groups in multiple data groups, including: setting a unit data size; determining the number of data groups included in the adjacent data groups according to the unit data size; and constructing multiple model training data in the multiple data groups with the number of data groups as a unit.

[0142] In one embodiment, the number of data groups is 3; accordingly, in the multiple data groups, the multiple model training data are constructed with the number of data groups as a unit, including: splicing every three adjacent data groups in the multiple data groups in sequence. For example: after the 8 data groups are arranged in sequence, the data groups at the arrangement positions 1, 2, and 3 are spliced ​​in sequence into one model training data, the data groups at the arrangement positions 2, 3, and 4 are spliced ​​in sequence into one model training data, the data groups at the arrangement positions 3, 4, and 5 are spliced ​​in sequence into one model training data, and the rest are spliced ​​in sequence and so on.

[0143] In one embodiment, every three adjacent data groups in the multiple data groups are sequentially spliced, including: converting each data group in the multiple data groups into a one-hot encoding, and splicing the one-hot encodings of every three adjacent data groups in sequence.

[0144] In one embodiment, the energy efficiency prediction model is obtained by training the model training data, including: inputting the model training data into the deep learning model so that the deep learning model outputs the corresponding training energy efficiency result; calculating the loss value between the training energy efficiency result and the preset energy efficiency label of the model training data; if the loss value meets the preset model convergence condition, the current deep learning model is used as the energy efficiency prediction model. If the loss value does not meet the model convergence condition, the model parameters of the current deep learning model are iteratively updated using the loss value until the loss value meets the model convergence condition. Among them, calculating the loss value between the training energy efficiency result and the preset energy efficiency label of the model training data includes: using the formula Calculate the loss value between the training energy efficiency result and the preset energy efficiency label of the model training data; represents the loss value, represents the training efficiency result, Represents the preset energy efficiency label, and N is the total number of samples participating in the training. When the number of data groups is 3, the preset energy efficiency label of the model training data is: the energy efficiency test result of the data group located in the middle position of the model training data.

[0145] In one embodiment, after the energy efficiency prediction model is obtained by training the model training data, the method further includes: constructing model application data based on configuration combinations other than the selected configuration combination in the multiple configuration combinations. The model application data is input into the energy efficiency prediction model so that the energy efficiency prediction model outputs the corresponding energy efficiency prediction results. In this way, it is not necessary to actually test all configuration combinations formed by the configuration items to be tested of the basic input and output system. For example, assuming that there are 8 configuration combinations in total, 4 of them can be used to construct model training data. After the energy efficiency prediction model is obtained by training, the energy efficiency prediction model is used to predict the remaining 4 configuration combinations for which there are no energy efficiency test results, and the corresponding energy efficiency prediction results are collected; thus, the energy efficiency test of all configuration combinations can be completed. Among them, the process of constructing model application data includes: arranging multiple configuration combinations to obtain a sequence; in the sequence, a plurality of model application data are constructed with a set number of combinations (the number of combinations should be equal to the number of data groups) as a unit. When the set number of combinations is 3, each three adjacent configuration combinations in the sequence are spliced ​​in sequence to obtain a plurality of model application data. That is, each three adjacent configuration combinations in the sequence constitute a model application data. It can be seen that the model application data does not contain energy efficiency test results, while the model training data contains energy efficiency test results; and when the energy efficiency prediction model performs energy efficiency prediction for a model application data, it refers to the energy efficiency test results of the adjacent configuration combinations of the predicted configuration combination in the current model application data. For example: if every three adjacent configuration combinations constitute a model application data, then the predicted configuration combination is located in the middle position, and its energy efficiency prediction results are predicted with reference to the energy efficiency test results of the two configuration combinations before and after the middle position. In other words: in a model application data, including a predicted configuration combination, other configuration combinations except this predicted configuration combination already have energy efficiency test results. For another example: if every five adjacent configuration combinations constitute a model application data, then the predicted configuration combination is located in the middle position, and its energy efficiency prediction results are predicted with reference to the energy efficiency test results of the four configuration combinations before and after the middle position. If every four adjacent configuration combinations constitute a model application data, then when the predicted configuration combination is in the second position, its energy efficiency prediction result refers to the energy efficiency test results of the previous and next two configuration combinations; when the predicted configuration combination is in the third position, its energy efficiency prediction result refers to the energy efficiency test results of the previous and next two configuration combinations (specifically, the performance-energy consumption ratio) for prediction.

[0146] It can be seen that this embodiment constructs model training data with partial configuration combinations and their non-full load energy efficiency test results, and trains to obtain an energy efficiency prediction model. This energy efficiency prediction model can be used to realize server energy efficiency prediction without the need to perform a full load test on the server after each configuration of the server BIOS option, thereby saving computer resources and testing time, and the energy efficiency prediction model can improve the accuracy of the test results; the model training data for training the energy efficiency prediction model is obtained based on partial configuration combinations and non-full load tests in multiple configuration combinations formed by various configuration items to be tested of the basic input and output system. The selection of partial configuration combinations and non-full load tests can reduce the amount of data processing, improve the model training efficiency, save computer resources, and further improve the efficiency and accuracy of server energy efficiency testing.

[0147] An energy efficiency prediction method provided by an embodiment of the present invention is introduced below. The energy efficiency prediction method described below can be referenced to other embodiments described in this document.

[0148] See also Figure 2 As shown, an embodiment of the present invention discloses an energy efficiency prediction method, comprising:

[0149] S201: Determine a plurality of configuration combinations formed by configuration items to be tested of a basic input / output system.

[0150] S202. Select some configuration combinations from multiple configuration combinations, call the energy efficiency test tool to perform non-full load test on the selected configuration combinations, and collect corresponding energy efficiency test results output by the energy efficiency test tool; the similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold.

[0151] S203: Use the energy efficiency prediction model to perform energy efficiency prediction on configuration combinations for which there are no corresponding energy efficiency test results among the multiple configuration combinations, and collect corresponding energy efficiency prediction results.

[0152] In this embodiment, the energy efficiency prediction model is obtained based on the model training method described in any embodiment.

[0153] In one embodiment, multiple configuration combinations and their corresponding energy efficiency prediction results are summarized and analyzed to determine the influence relationship between the configuration values ​​corresponding to each configuration item to be tested in the multiple configuration combinations and the corresponding energy efficiency prediction results; and multiple configuration combinations and their corresponding energy efficiency prediction results are visualized using curves.

[0154] It can be seen that this embodiment uses the prediction model to improve the accuracy of the test results, save computer resources, and improve the efficiency and accuracy of the server energy efficiency test.

[0155] See also Figure 3,A server energy efficiency testing solution includes functional modules such as parameter ,configuration, evaluation tool startup, data and model training, and model ,intelligent decision-making.

[0156] 1. Parameter configuration.

[0157] The user enters the pre-tested BIOS option file, which is in JSON format. For example, the user enters the following JSON format content:

[0158] { " UncoreFreqScaling ": [" Enabled ", " Min Frequency", " MaxFrequency", "Custom"]

[0159] "CoreC-States": ["Disabled", "Enabled"]

[0160] }

[0161] The backend will traverse all combinations of the input options, which are 4×2=8 combinations in total, namely: Enabled_Disabled, Enabled_Enabled, Min Frequency_Disabled, Min Frequency_Enabled, Max Frequency_Disabled, Max Frequency_Enabled, Custom_Disabled, and Custom_Enabled. All combinations are placed in a file and encoded in one-hot form. For example, Enabled_Disabled is encoded in one-hot as a vector with a length of 8, the first element is 1, and the other elements are 0; Enabled_Enabled is encoded in one-hot as a vector with a length of 8, the second element is 1, and the other elements are 0, so as to facilitate subsequent traversal.

[0162] 2. Start the benchmark tool.

[0163] Filter some BIOS combinations from the 8 combinations in the above example, such as selecting 4 of them, setting the BIOS panel parameters accordingly according to the parameter values ​​of each combination option, and starting the SPECpower evaluation tool respectively. Only 4 BIOS combinations are executed here to accumulate training data for subsequent model training. This method can reduce the time for searching for the optimal combination. In addition, the 4 BIOS combinations are screened according to the principle of adjacent intervals, such as sorting the 8 combinations from 1 to 8, and selecting 1 / 3 / 5 / 7 or 2 / 4 / 6 / 8 to execute the SPECpower evaluation tool. The purpose is to ensure that when predicting the remaining combinations, adjacent trained combinations can be used for prediction, which can ensure the accuracy of the prediction.

[0164] like Figure 4 As shown, the SPECpower evaluation tool is executed on the four BIOS option combinations 1 / 3 / 5 / 7. When predicting the performance-energy-consumption ratio of the second BIOS option combination, the one-hot encoding of the three BIOS option combinations 1 / 2 / 3 can be used as the input data of the deep learning model. Figure 4 The five digits after 1, 3, 5, and 7 are the performance-energy ratio of the corresponding BIOS option combination. In addition, when executing SPECpower, it does not execute all 11 workloads (i.e., test tasks), but selects the workload combination with the most consistent performance-energy ratio change trend under the 11 workloads to participate in the actual SPECpower evaluation, such as Figure 5 As shown, Figure 5 Among the 60 BIOS combinations shown, the energy efficiency curve corresponding to 9_8_5_2_0 has the highest consistency in change trend with the energy efficiency curves corresponding to the 11 full loads.

[0165] 3. Data and model training.

[0166] The Specpower evaluation tool was used to evaluate the four BIOS parameter combinations 1 / 3 / 5 / 7, and the corresponding evaluation data, including the server's performance, power consumption, load level and other values, were obtained. Based on this, the performance-to-energy ratio was further calculated. Here, the performance, power consumption and other parameters need to be normalized. The result of the normalization is ,in, is the original value of the parameter, The load level parameter will be preprocessed in the form of one-hot encoding for the maximum value of the corresponding parameter. Similarly, the performance-energy-consumption ratio value obtained by the evaluation will also be normalized in the same way.

[0167] Furthermore, the one-hot codes of the BIOS option combination whose energy efficiency test results need to be predicted and its adjacent combinations are concatenated together and preprocessed as the input of the deep learning-based neural network (the initial model that needs to be trained). Considering that the energy efficiency of neighboring BIOS options may be related, concatenating adjacent combinations can improve accuracy and enable the model to mine the energy efficiency relationship between neighboring combinations.

[0168] In one example, the structure of the model can be as follows Figure 6 As shown, it includes: an input layer, two hidden layers and an output layer. Among them, the loss function adopts the cross entropy loss function: ;in, is the true value of the performance-to-energy ratio, The performance-to-energy ratio predicted by the model.

[0169] 4. Model output and decision making.

[0170] After the model training is completed, the energy efficiency prediction model can be obtained, and then the energy efficiency prediction model can be used to predict the energy efficiency of the remaining 4 BIOS parameter combinations. In this example, the energy efficiency test results of the 4 BIOS option combinations 1 / 3 / 5 / 7 are known (SPECpower real evaluation data), so the model is used to predict the energy efficiency of the 4 combinations 2 / 4 / 6 / 8, thereby determining the combination with the highest performance-to-energy ratio from the 8 combinations as the optimal BIOS parameter combination in this test.

[0171] Furthermore, after obtaining all the performance-energy-consumption ratios evaluated by the SPECpower tool and the performance-energy-consumption ratios predicted by the model, the parameter combinations of these BIOS options and their corresponding performance-energy-consumption ratios can be automatically analyzed to determine which BIOS option values ​​have a greater impact on performance, energy consumption, and performance-energy-consumption ratio. And through a visual interface, it can help users understand the impact trend of changes in a certain BIOS parameter value on performance, energy consumption, and performance-energy-consumption ratio. It provides users with a more convenient and efficient analysis method to understand BIOS characteristics.

[0172] In this embodiment, in order to improve the efficiency of SPECpower test screening the optimal BIOS parameter combination, the model training samples are sampled and the number of loads tested is also screened. Adjacent BIOS option encoding information is also added to the model input to help the model more accurately predict the performance-energy-consumption ratio of the BIOS options to be predicted. Finally, the parameter combinations of all options and their corresponding performance-energy-consumption ratios are automatically analyzed to determine the BIOS option values ​​that have a greater impact on performance, energy consumption, and performance-energy-consumption ratio.

[0173] A model training device provided in an embodiment of the present invention is introduced below. The model training device described below can be referenced to other embodiments described in this document.

[0174] The embodiment of the present invention discloses a model training device, which is applied to test a control system, including:

[0175] A first determination module is used to determine a plurality of configuration combinations formed by various configuration items to be tested of the basic input and output system;

[0176] A first test module is used to select some configuration combinations from multiple configuration combinations, call an energy efficiency test tool to perform a non-full load test on the selected configuration combination, and collect corresponding energy efficiency test results output by the energy efficiency test tool; the similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold;

[0177] A construction module, used to construct the selected configuration combination and the corresponding energy efficiency test results into the same data group to obtain multiple data groups;

[0178] The training module is used to construct model training data based on adjacent data groups in multiple data groups, and use the model training data to train an energy efficiency prediction model.

[0179] In one implementation, the determination module is specifically configured to:

[0180] Determine each configuration item to be tested of the basic input and output system and the configuration value corresponding to each configuration item to be tested;

[0181] The configuration values ​​corresponding to the configuration items to be tested are combined without duplication to obtain multiple configuration combinations.

[0182] In one embodiment, the test module is specifically used to:

[0183] Select a portion of the configuration combinations from the multiple configuration combinations according to a preset ratio; the preset ratio is not less than fifty percent.

[0184] In one embodiment, the test module is specifically used to:

[0185] Sort multiple configuration combinations to obtain a target sequence;

[0186] Interval selection of configuration combinations in the target sequence.

[0187] In one embodiment, the test module is specifically used to:

[0188] The configuration combination arranged at the even-numbered position in the target sequence is taken as the selected configuration combination; or the configuration combination arranged at the odd-numbered position in the target sequence is taken as the selected configuration combination.

[0189] In one embodiment, the test module is specifically used to:

[0190] Determine, according to the task configuration file of the energy efficiency test tool, various test tasks corresponding to the full load test, and select some test tasks from the various test tasks;

[0191] The energy efficiency test tool is called to run the selected test task to perform a non-full load test on the selected configuration combination.

[0192] Optionally, the test module is specifically used to:

[0193] Some test tasks are selected from the various test tasks according to the non-full load test table preset in the task configuration file.

[0194] Optionally, the test module is specifically used to:

[0195] According to the random selection strategy configured in the task configuration file, randomly select a test task from the various test tasks as a task to be selected;

[0196] Calling the energy efficiency test tool to run the selected task to perform a non-full load test on the selected configuration combination, and collecting corresponding energy efficiency test results output by the energy efficiency test tool;

[0197] Selecting, from the obtained energy efficiency test results, an energy efficiency test result whose similarity to the energy efficiency test result of the full load test is greater than the preset threshold as an optional result;

[0198] The unselected task corresponding to the optional result is taken as the selected test task.

[0199] In one embodiment, the building blocks are specifically used to:

[0200] The selected configuration combination is combined with the corresponding energy efficiency test results to obtain corresponding multiple data groups.

[0201] In one embodiment, the building blocks are specifically used to:

[0202] Set the unit data size;

[0203] Determine the number of data groups included in the adjacent data group according to the unit data size;

[0204] In the multiple data groups, multiple model training data are constructed based on the number of data groups.

[0205] In one implementation, the number of data groups is 3; accordingly, the construction module is specifically used for:

[0206] Every three adjacent data groups among the multiple data groups are concatenated in sequence.

[0207] In one embodiment, the building blocks are specifically used to:

[0208] Each data group in the multiple data groups is converted into a one-hot code, and the one-hot codes of every three adjacent data groups are concatenated in sequence.

[0209] In one embodiment, the training module is specifically used to:

[0210] Inputting model training data into the deep learning model so that the deep learning model outputs corresponding training energy efficiency results;

[0211] Calculate the loss value between the training energy efficiency result and the preset energy efficiency label of the model training data;

[0212] If the loss value meets the preset model convergence condition, the current deep learning model will be used as the energy efficiency prediction model.

[0213] In one embodiment, the training module is specifically used to: if the loss value does not meet the model convergence conditions, then use the loss value to iteratively update the model parameters of the current deep learning model until the loss value meets the model convergence conditions.

[0214] In one embodiment, it further includes:

[0215] Another construction module is used to construct model application data based on configuration combinations other than the selected configuration combination among multiple configuration combinations after the energy efficiency prediction model is trained using the model training data.

[0216] Among them, for more specific working processes of each module and unit in this embodiment, reference can be made to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.

[0217] It can be seen that this embodiment provides a model training device, which constructs model training data with partial configuration combinations and their non-full load energy efficiency test results, and trains to obtain an energy efficiency prediction model. This energy efficiency prediction model can be used to realize server energy efficiency prediction without the need to perform a full load test on the server after each configuration of the server BIOS option, thereby saving computer resources and testing time, and the energy efficiency prediction model can improve the accuracy of the test results; the model training data for training the energy efficiency prediction model is obtained based on partial configuration combinations and non-full load tests in multiple configuration combinations formed by each configuration item to be tested of the basic input and output system. The selection of partial configuration combinations and non-full load tests can reduce the amount of data processing, improve the model training efficiency, save computer resources, and further improve the efficiency and accuracy of server energy efficiency testing.

[0218] An energy efficiency prediction device provided by an embodiment of the present invention is introduced below. The energy efficiency prediction device described below can be referenced to other embodiments described in this document.

[0219] The embodiment of the present invention discloses an energy efficiency prediction device, comprising:

[0220] A second determination module is used to determine a plurality of configuration combinations formed by various configuration items to be tested of the basic input and output system;

[0221] The second test module is used to select some configuration combinations from multiple configuration combinations, call the energy efficiency test tool to perform a non-full load test on the selected configuration combination, and collect the corresponding energy efficiency test results output by the energy efficiency test tool; the similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold;

[0222] A prediction module, used to use an energy efficiency prediction model to perform energy efficiency prediction on configuration combinations for which there are no corresponding energy efficiency test results among multiple configuration combinations, and to collect corresponding energy efficiency prediction results;

[0223] Among them, the energy efficiency prediction model is obtained based on the model training method described in any embodiment.

[0224] In one embodiment, it further includes:

[0225] An analysis module is used to summarize and analyze multiple configuration combinations and their corresponding energy efficiency prediction results to determine the influence relationship between the configuration values ​​corresponding to each configuration item to be tested in the multiple configuration combinations and the corresponding energy efficiency prediction results;

[0226] The display module is used to visualize multiple configuration combinations and their corresponding energy efficiency prediction results using curves.

[0227] Among them, for more specific working processes of each module and unit in this embodiment, reference can be made to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.

[0228] It can be seen that this embodiment provides an energy efficiency prediction device, which improves the accuracy of test results by using a prediction model, saves computer resources, and improves the efficiency and accuracy of server energy efficiency testing.

[0229] An electronic device provided by an embodiment of the present invention is introduced below. The electronic device described below can be referenced to other embodiments described in this document.

[0230] See also Figure 7 As shown, an embodiment of the present invention discloses an electronic device, including:

[0231] Memory 701, used for storing computer programs;

[0232] The processor 702 is used to execute the computer program to implement the method disclosed in any of the above embodiments.

[0233] In this embodiment, when the processor executes the computer program stored in the memory, the following steps can be specifically implemented: determining multiple configuration combinations formed by the configuration items to be tested of the basic input and output system; selecting some configuration combinations from the multiple configuration combinations, calling the energy efficiency test tool to perform a non-full load test on the selected configuration combinations, and collecting the corresponding energy efficiency test results output by the energy efficiency test tool; the similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold; constructing the selected configuration combination and the corresponding energy efficiency test results into the same data group to obtain multiple data groups; constructing model training data based on adjacent data groups in the multiple data groups, and using the model training data to train an energy efficiency prediction model.

[0234] In this embodiment, when the processor executes the computer program stored in the memory, the following steps can be specifically implemented: determining each configuration item to be tested of the basic input and output system and the configuration value corresponding to each configuration item to be tested; performing non-repetitive combinations of the configuration values ​​corresponding to each configuration item to be tested to obtain multiple configuration combinations.

[0235] In this embodiment, when the processor executes the computer program stored in the memory, the following steps may be specifically implemented: selecting some configuration combinations from multiple configuration combinations according to a preset ratio; the preset ratio is not less than fifty percent.

[0236] In this embodiment, when the processor executes the computer program stored in the memory, the following steps may be specifically implemented: sorting multiple configuration combinations to obtain a target sequence; and selecting configuration combinations at intervals in the target sequence.

[0237] In this embodiment, when the processor executes the computer program stored in the memory, the following steps can be specifically implemented: taking the configuration combination arranged at the even-numbered positions in the target sequence as the selected configuration combination; or taking the configuration combination arranged at the odd-numbered positions in the target sequence as the selected configuration combination.

[0238] In this embodiment, when the processor executes the computer program stored in the memory, the following steps can be specifically implemented: determining each test task corresponding to the full load test; selecting some test tasks from each test task; and using the selected test tasks to call the energy efficiency test tool to perform non-full load testing on the selected configuration combination.

[0239] In this embodiment, when the processor executes the computer program stored in the memory, the following steps may be specifically implemented: selecting some test tasks from various test tasks according to a preset non-full load test table.

[0240] In this embodiment, when the processor executes the computer program stored in the memory, the following steps can be specifically implemented: randomly selecting a test task from each test task as a candidate task; using the candidate task to call the energy efficiency test tool to perform a non-full load test on the selected configuration combination, and collecting the corresponding energy efficiency test results output by the energy efficiency test tool; selecting an energy efficiency test result whose similarity with the energy efficiency test result of the full load test is greater than a preset threshold from the obtained energy efficiency test results as an optional result; and using the candidate task corresponding to the optional result as the selected test task.

[0241] In this embodiment, when the processor executes the computer program stored in the memory, the following steps may be specifically implemented: splicing the selected configuration combination with the corresponding energy efficiency test results to obtain corresponding multiple data groups.

[0242] In this embodiment, when the processor executes the computer program stored in the memory, the following steps can be specifically implemented: setting the unit data size; determining the number of data groups included in adjacent data groups according to the unit data size; and constructing multiple model training data in multiple data groups based on the number of data groups.

[0243] In this embodiment, when the processor executes the computer program stored in the memory, the following steps may be specifically implemented: splicing every three adjacent data groups among the multiple data groups in sequence.

[0244] In this embodiment, when the processor executes the computer program stored in the memory, the following steps can be specifically implemented: converting each data group in the multiple data groups into a one-hot code, and splicing the one-hot codes of every three adjacent data groups in sequence.

[0245] In this embodiment, when the processor executes the computer program stored in the memory, the following steps can be specifically implemented: inputting model training data into the deep learning model so that the deep learning model outputs corresponding training energy efficiency results; calculating the loss value between the training energy efficiency results and the preset energy efficiency labels of the model training data; if the loss value meets the preset model convergence conditions, the current deep learning model is used as an energy efficiency prediction model.

[0246] In this embodiment, when the processor executes the computer program stored in the memory, the following steps can be specifically implemented: if the loss value does not meet the model convergence conditions, the model parameters of the current deep learning model are iteratively updated using the loss value until the loss value meets the model convergence conditions.

[0247] In this embodiment, when the processor executes the computer program stored in the memory, the following steps may be specifically implemented: constructing model application data based on multiple configuration combinations.

[0248] In this embodiment, when the processor executes the computer program stored in the memory, the following steps may be specifically implemented: determining multiple configuration combinations formed by various configuration items to be tested of the basic input and output system; selecting some configuration combinations from the multiple configuration combinations, calling the energy efficiency test tool to perform a non-full load test on the selected configuration combination, and collecting the corresponding energy efficiency test results output by the energy efficiency test tool; the degree of similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold; using the energy efficiency prediction model to perform energy efficiency prediction on the configuration combinations for which there are no corresponding energy efficiency test results in the multiple configuration combinations, and collecting the corresponding energy efficiency prediction results;

[0249] In this embodiment, when the processor executes the computer program stored in the memory, the following steps can be specifically implemented: summarizing and analyzing multiple configuration combinations and their corresponding energy efficiency prediction results to determine the influence relationship between the configuration values ​​corresponding to each configuration item to be tested in the multiple configuration combinations and the corresponding energy efficiency prediction results; using curves to visualize the multiple configuration combinations and their corresponding energy efficiency prediction results.

[0250] Furthermore, an embodiment of the present invention also provides an electronic device. The electronic device can be Figure 8 The server shown can also be Fig. 9 Terminal shown. Figure 8 and Fig. 9 All of them are structural diagrams of electronic devices according to an exemplary embodiment, and the contents in the diagrams cannot be regarded as any limitation on the scope of application of the present invention.

[0251] Figure 8 A schematic diagram of the structure of a server provided in an embodiment of the present invention. The server may specifically include: at least one processor, at least one memory, a power supply, a communication interface, an input / output interface, and a communication bus. The memory is used to store a computer program, and the computer program is loaded and executed by the processor to implement the relevant steps in the model training disclosed in any of the aforementioned embodiments.

[0252] In this embodiment, the power supply is used to provide working voltage for each hardware device on the server; the communication interface can create a data transmission channel between the server and external devices, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present invention, and is not specifically limited here; the input and output interface is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0253] In addition, the memory as a carrier for resource storage can be a read-only memory, random access memory, disk or CD, etc. The resources stored thereon include operating system, computer programs and data, etc. The storage method can be temporary storage or permanent storage.

[0254] The operating system is used to manage and control the hardware devices and computer programs on the server to realize the operation and processing of the data in the memory by the processor, which can be Windows Server, Netware, Unix, Linux, etc. In addition to computer programs that can be used to complete the model training method disclosed in any of the aforementioned embodiments, computer programs can also further include computer programs that can be used to complete other specific tasks. In addition to data such as application update information, data can also include data such as application developer information.

[0255] Fig. 9 A schematic diagram of the structure of a terminal provided in an embodiment of the present invention, wherein the terminal may specifically include but is not limited to a smart phone, a tablet computer, a laptop computer or a desktop computer.

[0256] Generally, the terminal in this embodiment includes: a processor and a memory.

[0257] Among them, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0258] The memory may include one or more computer non-volatile storage media, which may be non-transitory. The memory may also include high-speed random access memory, and non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory is used to store at least the following computer program, wherein, after the computer program is loaded and executed by the processor, it can implement the relevant steps in the model training and / or energy efficiency prediction method performed by the terminal side disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory may also include an operating system and data, etc., and the storage method may be short-term storage or permanent storage. Among them, the operating system may include Windows, Unix, Linux, etc. The data may include, but is not limited to, update information of the application.

[0259] In some embodiments, the terminal may also include a display screen, an input and output interface, a communication interface, a sensor, a power supply, and a communication bus.

[0260] Those skilled in the art will understand that Fig. 9 The structure shown in the figure does not constitute a limitation on the terminal, and may include more or fewer components than those shown in the figure.

[0261] A nonvolatile storage medium provided in an embodiment of the present invention is introduced below. The nonvolatile storage medium described below and other embodiments described herein may be cross-referenced.

[0262] A non-volatile storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the model training and / or energy efficiency prediction method disclosed in the above-mentioned embodiment. The non-volatile storage medium is a computer-readable non-volatile storage medium, which, as a carrier for resource storage, may be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon include an operating system, a computer program, and data, etc., and the storage method may be temporary storage or permanent storage.

[0263] A computer program product provided by an embodiment of the present invention is introduced below. The computer program product described below can be referenced to other embodiments described in this document.

[0264] A computer program product includes a computer program / instruction, which, when executed by a processor, implements the steps of the aforementioned disclosed model training and / or energy efficiency prediction method.

[0265] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0266] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of non-volatile storage medium known in the art.

[0267] Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A model training method, characterized in that: Used in testing control systems, including: Determine multiple configuration combinations formed by configuration items to be tested of the basic input and output system; Selecting some configuration combinations from the multiple configuration combinations, calling an energy efficiency test tool to perform a non-full load test on the selected configuration combinations, and collecting corresponding energy efficiency test results output by the energy efficiency test tool; the degree of similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold; The selected configuration combinations and corresponding energy efficiency test results are constructed into the same data group to obtain multiple data groups; Model training data is constructed based on adjacent data groups in the multiple data groups, and an energy efficiency prediction model is trained using the model training data.

2. The method according to claim 1, characterized in that Determine multiple configuration combinations formed by various configuration items to be tested of the basic input and output system, including: Determine each configuration item to be tested of the basic input / output system and the configuration value corresponding to each configuration item to be tested; The configuration values ​​corresponding to the configuration items to be tested are respectively combined without duplication to obtain the multiple configuration combinations.

3. The method according to claim 1, characterized in that Selecting some configuration combinations from the multiple configuration combinations includes: Selecting some configuration combinations from the multiple configuration combinations according to a preset ratio; the preset ratio is not less than fifty percent.

4. The method according to claim 1, characterized in that: Selecting some configuration combinations from the multiple configuration combinations includes: Sorting the multiple configuration combinations to obtain a target sequence; Configuration combinations are selected at intervals in the target sequence.

5. The method according to claim 4, characterized in that Interval selection configuration combinations in the target sequence include: The configuration combination arranged at the even-numbered positions in the target sequence is taken as the selected configuration combination; or the configuration combination arranged at the odd-numbered positions in the target sequence is taken as the selected configuration combination.

6. The method according to claim 1, characterized in that Call the energy efficiency test tool to perform a non-full load test on the selected configuration combination, including: Determine, according to the task configuration file of the energy efficiency test tool, various test tasks corresponding to the full load test, and select some test tasks from the various test tasks; The energy efficiency test tool is called to run the selected test task to perform a non-full load test on the selected configuration combination.

7. The method according to claim 6, characterized in that Select some test tasks from the various test tasks, including: Some test tasks are selected from the various test tasks according to the non-full load test table preset in the task configuration file.

8. The method according to claim 6, characterized in that Select some test tasks from the various test tasks, including: According to the random selection strategy configured in the task configuration file, randomly select a test task from the various test tasks as a task to be selected; Calling the energy efficiency test tool to run the selected task to perform a non-full load test on the selected configuration combination, and collecting corresponding energy efficiency test results output by the energy efficiency test tool; Selecting, from the obtained energy efficiency test results, an energy efficiency test result whose similarity to the energy efficiency test result of the full load test is greater than the preset threshold as an optional result; The unselected task corresponding to the optional result is taken as the selected test task.

9. The method according to claim 1, characterized in that: The selected configuration combinations and corresponding energy efficiency test results are constructed into the same data group, and multiple data groups are obtained, including: The selected configuration combination is combined with the corresponding energy efficiency test results to obtain corresponding multiple data groups.

10. The method according to any one of claims 1 to 9, characterized in that: Constructing model training data based on adjacent data groups in the multiple data groups includes: Set the unit data size; Determining the number of data groups included in the adjacent data group according to the unit data size; A plurality of the model training data are constructed in the plurality of data groups with the number of the data groups as a unit.

11. The method according to claim 10, characterized in that The number of data groups is 3; Accordingly, in the plurality of data groups, a plurality of the model training data are constructed with the number of the data groups as a unit, including: Every three adjacent data groups among the multiple data groups are spliced ​​in sequence.

12. The method according to claim 11, characterized in that Sequentially splicing every three adjacent data groups in the plurality of data groups, comprising: Each data group in the multiple data groups is converted into a one-hot code, and the one-hot codes of every three adjacent data groups are concatenated in sequence.

13. The method according to any one of claims 1 to 9, characterized in that: The energy efficiency prediction model is obtained by training the model training data, including: Inputting the model training data into a deep learning model so that the deep learning model outputs corresponding training energy efficiency results; Calculating a loss value between the training energy efficiency result and a preset energy efficiency label of the model training data; If the loss value meets the preset model convergence condition, the current deep learning model is used as the energy efficiency prediction model.

14. The method according to claim 13, characterized in that If the loss value does not meet the model convergence condition, the model parameters of the current deep learning model are iteratively updated using the loss value until the loss value meets the model convergence condition.

15. The method according to any one of claims 1 to 9, characterized in that: After the energy efficiency prediction model is obtained by training the model training data, the method further includes: Model application data is constructed based on the multiple configuration combinations.

16. An energy efficiency prediction method, characterized in that: include: Determine multiple configuration combinations formed by configuration items to be tested of the basic input and output system; Selecting some configuration combinations from the multiple configuration combinations, calling an energy efficiency test tool to perform a non-full load test on the selected configuration combinations, and collecting corresponding energy efficiency test results output by the energy efficiency test tool; the degree of similarity between the energy efficiency test results of the non-full load test and the full load test is greater than a preset threshold; Using the energy efficiency prediction model, performing energy efficiency prediction on the configuration combinations for which there are no corresponding energy efficiency test results among the multiple configuration combinations, and collecting corresponding energy efficiency prediction results; Wherein, the energy efficiency prediction model is obtained based on the model training method described in any one of claims 1 to 15.

17. The method according to claim 16, characterized in that Also includes: Summarizing and analyzing the multiple configuration combinations and their corresponding energy efficiency prediction results to determine the influence relationship between the configuration values ​​corresponding to the configuration items to be tested in the multiple configuration combinations and the corresponding energy efficiency prediction results; The multiple configuration combinations and their corresponding energy efficiency prediction results are visualized using curves.

18. An electronic device, characterized in that: include: Memory for storing computer programs; A processor, configured to execute the computer program to implement the method according to any one of claims 1 to 17.

19. A non-volatile storage medium, characterized in that: Used to store a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 17 is implemented.

20. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instructions are executed by a processor, the method according to any one of claims 1 to 17 is implemented.

Images