Performance information prediction model generation method, prediction method, device, equipment and medium

Abstract

The embodiment of the invention discloses a performance information prediction model generation method and device, a prediction method and device, equipment and a medium. According to one specific embodiment, the method comprises the steps that based on at least one sample in a sample set, the following training steps are executed: the at least one sample is input into a sequence decomposition layer included in an initial performance information prediction model to obtain a sequence decomposition result, and the sample set comprises historical performance data in a preset time period; inputting the sequence decomposition result into a semantic extraction layer to obtain semantic sequence information; inputting the semantic sequence information into a domain conversion layer to obtain converted semantic sequence information; inputting the converted semantic sequence information into a backbone network to obtain performance prediction information corresponding to at least one sample; and training the initial performance information prediction model according to the performance prediction information to obtain a performance information prediction model. The implementation mode is related to performance monitoring, performance prediction can be carried out in advance, and complex change modes in historical performance data can be learned.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to a performance information prediction model generation method, a prediction method, an apparatus, a device, and a medium. Background Art

[0002] Computer performance monitoring helps quickly identify and respond to performance issues, minimizing the impact on business. Currently, computer performance monitoring typically relies on manual inspections, where specialized personnel regularly check system performance; checking system performance using performance alarm thresholds; or predicting performance using time series prediction methods that only consider temporal dependencies.

[0003] However, when using the above methods, the following technical problems often arise: manual inspection methods and performance detection methods that set alarm thresholds have delays. Often, when an alarm is received, the performance anomaly has already lasted for a period of time, which may have unpredictable impacts. Time series prediction methods that only consider time dependence have difficulty learning complex change patterns, resulting in poor accuracy of performance prediction results, many missed performance issues, and more computer system downtime.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the Invention

[0005] The content of this disclosure is used to briefly introduce concepts that will be described in detail in the detailed description section below. The content of this disclosure is not intended to identify key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0006] Some embodiments of the present disclosure propose a performance information prediction model generation method, a performance information prediction method, a performance information prediction model generation device, a performance information prediction device, an electronic device and a computer-readable medium to solve one or more of the technical problems mentioned in the above background technology section.

[0007] In a first aspect, some embodiments of the present disclosure provide a method for generating a performance information prediction model, the method comprising: based on at least one sample in a sample set, performing the following training steps: inputting the at least one sample into a sequence decomposition layer included in an initial performance information prediction model to obtain a sequence decomposition result, wherein the sample set includes historical performance data within a preset time period, and the initial performance information prediction model comprises a sequence decomposition layer, a semantic extraction layer, a domain conversion layer and a backbone network; inputting the sequence decomposition result into the semantic extraction layer to obtain semantic sequence information; inputting the semantic sequence information into the domain conversion layer to obtain converted semantic sequence information; inputting the converted semantic sequence information into the backbone network to obtain performance prediction information corresponding to the at least one sample; and training the initial performance information prediction model based on the performance prediction information to obtain a performance information prediction model.

[0008] Optionally, the method also includes: obtaining various historical performance data within a preset time period; cleaning the above-mentioned historical performance data to obtain various cleaned historical performance data; dividing the above-mentioned cleaned historical performance data to obtain a training set, a verification set and a test set, wherein the above-mentioned sample set includes the above-mentioned training set and the above-mentioned verification set.

[0009] Optionally, the above-mentioned at least one sample is input into the sequence decomposition layer included in the initial performance information prediction model to obtain a sequence decomposition result, including: performing sequence decomposition processing on the above-mentioned at least one sample to obtain an initial trend term, an initial seasonal term and an initial residual term; and normalizing the above-mentioned initial trend term, the above-mentioned initial seasonal term and the above-mentioned initial residual term respectively to obtain a normalized trend term, a normalized seasonal term and a normalized residual term as the sequence decomposition result.

[0010] Optionally, the above-mentioned sequence decomposition result is input into the above-mentioned semantic extraction layer to obtain semantic sequence information, including: mapping the above-mentioned normalized trend item, the above-mentioned normalized season item and the above-mentioned normalized residual item respectively to obtain mapping trend item, mapping season item and mapping residual item; embedding the above-mentioned mapping trend item, the above-mentioned mapping season item and the above-mentioned mapping residual item to obtain trend item semantic sequence, season item semantic sequence and residual item semantic sequence as semantic sequence information.

[0011] Optionally, the above-mentioned semantic sequence information is input into the above-mentioned domain conversion layer to obtain converted semantic sequence information, including: performing domain conversion processing on the trend item semantic sequence and season item semantic sequence included in the above-mentioned semantic sequence information respectively to obtain a converted trend item semantic sequence and a converted season item semantic sequence; and determining the above-mentioned converted trend item semantic sequence, the converted season item semantic sequence and the above-mentioned residual item semantic sequence as converted semantic sequence information.

[0012] Optionally, the above-mentioned conversion semantic sequence information is input into the above-mentioned backbone network to obtain performance prediction information corresponding to the above-mentioned at least one sample, including: generating prompt information based on the above-mentioned conversion semantic sequence information; splicing the above-mentioned prompt information and the above-mentioned conversion semantic sequence information to obtain splicing information; inputting the above-mentioned splicing information into the above-mentioned backbone network to obtain feature sequence information corresponding to the above-mentioned at least one sample; performing inverse conversion processing on the above-mentioned feature sequence information to obtain inverse conversion feature sequence information; inputting the above-mentioned inverse conversion feature sequence information into the output unit included in the above-mentioned backbone network to obtain performance prediction information corresponding to at least one sample.

[0013] Optionally, the above-mentioned historical performance data includes performance information and performance mutation information; the above-mentioned inputting the above-mentioned inverse conversion feature sequence information into the output unit included in the above-mentioned backbone network to obtain performance prediction information corresponding to at least one sample includes: inputting the above-mentioned inverse conversion feature sequence information into the fully connected layer included in the above-mentioned output unit to obtain preliminary performance prediction information; inputting the above-mentioned preliminary performance prediction information into the first fully connected layer and the second fully connected layer included in the above-mentioned output unit respectively to obtain preliminary predicted performance information and preliminary predicted performance mutation information corresponding to the above-mentioned at least one sample; generating predicted performance information corresponding to the above-mentioned at least one sample based on the above-mentioned preliminary predicted performance information; generating predicted performance mutation information corresponding to the above-mentioned at least one sample based on the above-mentioned preliminary predicted performance mutation information; and determining the above-mentioned predicted performance information and the above-mentioned predicted performance mutation information as the performance prediction information corresponding to the above-mentioned at least one sample.

[0014] Optionally, the above-mentioned training of the initial performance information prediction model based on the above-mentioned performance prediction information to obtain the performance information prediction model includes: determining whether the initial performance information prediction model achieves the optimization target based on the above-mentioned performance prediction information and the various historical performance data corresponding to the above-mentioned performance prediction information; in response to determining that the initial performance information prediction model achieves the optimization target, determining the initial performance information prediction model as the performance information prediction model; in response to determining that the initial performance information prediction model does not achieve the optimization target, adjusting the network parameters of the initial performance information prediction model, and using unused samples to form a sample set, and performing the above-mentioned training steps again based on the adjusted initial performance information prediction model.

[0015] Optionally, the above-mentioned determining whether the initial performance information prediction model achieves the optimization target based on the above-mentioned performance prediction information and the various historical performance data corresponding to the above-mentioned performance prediction information includes: generating loss information based on the above-mentioned performance prediction information and the various historical performance data corresponding to the above-mentioned performance prediction information; in response to determining that the above-mentioned loss information meets the preset loss condition, generating training effect information through the verification set and the initial performance information prediction model; in response to determining that the above-mentioned training effect information meets the preset training condition, determining that the initial performance information prediction model achieves the optimization target.

[0016] In a second aspect, some embodiments of the present disclosure provide a performance information prediction method, which includes: obtaining time-series performance data; inputting the above performance data into a pre-trained performance information prediction model to obtain performance prediction information, wherein the above performance information prediction model is generated by the method described in any implementation method of the above first aspect; generating performance alarm information based on the above performance prediction information; and executing a performance alarm operation based on the above performance alarm information.

[0017] In a third aspect, some embodiments of the present disclosure provide a performance information prediction model generation device, the device comprising: an execution unit, configured to perform the following training steps based on at least one sample in a sample set: inputting the at least one sample into a sequence decomposition layer included in an initial performance information prediction model to obtain a sequence decomposition result, wherein the sample set includes historical performance data within a preset time period, and the initial performance information prediction model comprises a sequence decomposition layer, a semantic extraction layer, a domain conversion layer and a backbone network; inputting the sequence decomposition result into the semantic extraction layer to obtain semantic sequence information; inputting the semantic sequence information into the domain conversion layer to obtain converted semantic sequence information; inputting the converted semantic sequence information into the backbone network to obtain performance prediction information corresponding to the at least one sample; and training the initial performance information prediction model based on the performance prediction information to obtain a performance information prediction model.

[0018] Optionally, the prediction model generation apparatus may further include a data acquisition unit, a cleaning unit, and a partitioning unit. The data acquisition unit is configured to acquire various historical performance data within a preset time period. The cleaning unit is configured to perform cleaning processing on the various historical performance data to obtain cleaned historical performance data. The partitioning unit is configured to partition the cleaned historical performance data to obtain a training set, a validation set, and a test set, wherein the sample set includes the training set and the validation set.

[0019] Optionally, the execution unit is further configured to: perform sequence decomposition processing on the above-mentioned at least one sample to obtain an initial trend term, an initial seasonal term and an initial residual term; and perform normalization processing on the above-mentioned initial trend term, the above-mentioned initial seasonal term and the above-mentioned initial residual term respectively to obtain a normalized trend term, a normalized seasonal term and a normalized residual term as a sequence decomposition result.

[0020] Optionally, the execution unit is further configured to: map the above-mentioned normalized trend item, the above-mentioned normalized season item and the above-mentioned normalized residual item respectively to obtain a mapped trend item, a mapped season item and a mapped residual item; embed the above-mentioned mapped trend item, the above-mentioned mapped season item and the above-mentioned mapped residual item to obtain a trend item semantic sequence, a season item semantic sequence and a residual item semantic sequence as semantic sequence information.

[0021] Optionally, the execution unit is further configured to: perform domain conversion processing on the trend item semantic sequence and season item semantic sequence included in the above-mentioned semantic sequence information, respectively, to obtain a converted trend item semantic sequence and a converted season item semantic sequence; and determine the above-mentioned converted trend item semantic sequence, the converted season item semantic sequence and the above-mentioned residual item semantic sequence as the converted semantic sequence information.

[0022] Optionally, the execution unit is further configured to: generate prompt information based on the above-mentioned conversion semantic sequence information; splice the above-mentioned prompt information and the above-mentioned conversion semantic sequence information to obtain splicing information; input the above-mentioned splicing information into the above-mentioned backbone network to obtain feature sequence information corresponding to the above-mentioned at least one sample; perform inverse conversion processing on the above-mentioned feature sequence information to obtain inverse conversion feature sequence information; input the above-mentioned inverse conversion feature sequence information into the output unit included in the above-mentioned backbone network to obtain performance prediction information corresponding to at least one sample.

[0023] Optionally, the above historical performance data includes performance information and performance mutation information.

[0024] Optionally, the execution unit is further configured to: input the above-mentioned inverse conversion feature sequence information into the fully connected layer included in the above-mentioned output unit to obtain preliminary performance prediction information; input the above-mentioned preliminary performance prediction information into the first fully connected layer and the second fully connected layer included in the above-mentioned output unit respectively to obtain preliminary predicted performance information and preliminary predicted performance mutation information corresponding to the above-mentioned at least one sample; generate predicted performance information corresponding to the above-mentioned at least one sample based on the above-mentioned preliminary predicted performance information; generate predicted performance mutation information corresponding to the above-mentioned at least one sample based on the above-mentioned preliminary predicted performance mutation information; determine the above-mentioned predicted performance information and the above-mentioned predicted performance mutation information as performance prediction information corresponding to the above-mentioned at least one sample.

[0025] Optionally, the execution unit is further configured to: determine whether the initial performance information prediction model achieves the optimization target based on the above-mentioned performance prediction information and the various historical performance data corresponding to the above-mentioned performance prediction information; in response to determining that the initial performance information prediction model achieves the optimization target, determine the initial performance information prediction model as the performance information prediction model; in response to determining that the initial performance information prediction model does not achieve the optimization target, adjust the network parameters of the initial performance information prediction model, and adopt unused samples to form a sample set, and execute the above-mentioned training steps again based on the adjusted initial performance information prediction model.

[0026] Optionally, the execution unit is further configured to: generate loss information based on the above-mentioned performance prediction information and each historical performance data corresponding to the above-mentioned performance prediction information; in response to determining that the above-mentioned loss information meets the preset loss condition, generate training effect information through the verification set and the initial performance information prediction model; in response to determining that the above-mentioned training effect information meets the preset training condition, determine that the initial performance information prediction model achieves the optimization goal. In a fourth aspect, some embodiments of the present disclosure provide a performance information prediction device, the device comprising: an acquisition unit, configured to acquire time-series performance data; an input unit, configured to input the above-mentioned performance data into a pre-trained performance information prediction model to obtain performance prediction information, wherein the above-mentioned performance information prediction model is generated by the method described in any implementation method of the above-mentioned first aspect; a generation unit, configured to generate performance alarm information based on the above-mentioned performance prediction information; and an alarm unit, configured to perform a performance alarm operation based on the above-mentioned performance alarm information.

[0027] In a fifth aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by one or more processors, the one or more processors implement the method described in any implementation of the first or second aspect above.

[0028] In a sixth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation of the first or second aspect above is implemented.

[0029] In a seventh aspect, some embodiments of the present disclosure provide a computer program product, including a computer program, which, when executed by a processor, implements the method described in any implementation of the first or second aspect above.

[0030] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: Through the performance information prediction model generation method of some embodiments of the present disclosure, performance prediction can be performed in advance, and complex change patterns in historical performance data can be learned, thereby reducing the impact of performance anomalies and the number of computer system downtimes. Specifically, the reasons for the possibility of unpredictable impacts and the high number of computer system downtimes are: manual inspection methods and performance detection methods that set alarm thresholds have time delays. Often, when an alarm is received, the performance anomaly has already persisted for a period of time, which may have unpredictable impacts; time series prediction methods that only consider time dependencies have difficulty learning complex change patterns, resulting in poor performance prediction results, a large number of missed performance issues, and a large number of computer system downtimes. Based on this, the performance information prediction model generation method of some embodiments of the present disclosure performs the following training steps based on at least one sample in a sample set: First, the at least one sample is input into the sequence decomposition layer included in the initial performance information prediction model to obtain a sequence decomposition result. The sample set includes historical performance data within a preset time period, and the initial performance information prediction model includes a sequence decomposition layer, a semantic extraction layer, a domain conversion layer, and a backbone network. In this way, the structure and characteristics of time series data can be combined to split complex time series data so that the model can be more easily understood and interpreted. The second step is to input the above sequence decomposition results into the above semantic extraction layer to obtain semantic sequence information. In this way, the model can capture the changes in time series more finely. The third step is to input the above semantic sequence information into the above domain conversion layer to obtain converted semantic sequence information. In this way, the time series signal can be decomposed into its component frequencies, which can effectively identify complex patterns in the data. The fourth step is to input the above converted semantic sequence information into the above backbone network to obtain performance prediction information corresponding to the above at least one sample. In this way, performance prediction can be performed through the backbone network. The fifth step is to train the initial performance information prediction model based on the above performance prediction information to obtain a performance information prediction model. In this way, historical performance data and predicted performance data can be used to train the model. Because the trained performance information prediction model can split, extract semantics, and transform domains from time series data through the sequence decomposition layer, semantic extraction layer, and domain transformation layer, it can better understand and capture the complex patterns of performance changes over time. This improves the accuracy of predicted performance information, reduces missed performance issues, and thus reduces the number of computer system downtimes. It also avoids manual inspections and performance testing methods that require setting alarm thresholds, allowing for early performance predictions and reducing the impact of performance anomalies. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The above and other features, advantages, and aspects of the various embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0032] Figure 1 is a flowchart of some embodiments of the method for generating a performance information prediction model according to the present disclosure;

[0033] Figure 2 is a schematic diagram of the structure of a performance information prediction model according to a method for generating a performance information prediction model in accordance with some embodiments of the present disclosure;

[0034] Figure 3 is a flow chart of some embodiments of the performance information prediction method according to the present disclosure;

[0035] Figure 4 is a schematic structural diagram of some embodiments of the performance information prediction model generation device according to the present disclosure;

[0036] Figure 5 is a schematic structural diagram of some embodiments of the performance information prediction device according to the present disclosure;

[0037] Figure 6 is a schematic structural diagram of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0038] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0039] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other.

[0040] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0041] It should be noted that the modifications of "one" and "multiple" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".

[0042] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0043] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0044] Figure 1 The process 100 of some embodiments of the performance information prediction model generation method according to the present disclosure is shown. The performance information prediction model generation method includes the following steps:

[0045] Step 101: Based on at least one sample in the sample set, perform the following training steps:

[0046] Step 1011: input at least one sample into the sequence decomposition layer included in the initial performance information prediction model to obtain a sequence decomposition result.

[0047] In some embodiments, the execution entity (e.g., a server or terminal device) of the performance information prediction model generation method may input the at least one sample into a sequence decomposition layer included in the initial performance information prediction model to obtain a sequence decomposition result. The sample set includes historical performance data within a preset time period. The preset time period may be a pre-set historical time period. The historical performance data may be data related to computer performance indicators. For example, the historical performance data may include, but is not limited to, CPU usage. The sample set may be time series data. For example, CPU usage may represent the average CPU usage per minute. Each CPU usage may be associated with a collection time. Historical performance data may be obtained through a performance monitoring platform. The at least one sample may be a batch of samples during model training. When the training step is first performed, the initial performance information prediction model may be an initialized neural network model without network parameter adjustment. When the training step is subsequently performed, the initial performance information prediction model may be a neural network model after the last network parameter adjustment. The initial performance information prediction model may include a sequence decomposition layer, a semantic extraction layer, a domain conversion layer, and a backbone network. The sequence decomposition layer may be used to perform sequence decomposition on time series data. For example, the sequence decomposition layer can perform global STL (Seasonal and Trend decomposition using Loess) decomposition on time series data. The semantic extraction layer can be used to perform local semantic extraction on the sequence decomposition results. For example, the sequence decomposition results can be input into the embedding layer for local semantic feature extraction. The domain conversion layer can be used to perform domain conversion processing on the local semantic features corresponding to the decomposition results with trend and periodicity. For example, the domain conversion processing can be a frequency domain decomposition processing in Fourier transform. The backbone network can use the complex change pattern of the feature learning performance obtained by the above processing to perform performance prediction on the neural network. For example, the backbone network can be a Transformer. In practice, the above-mentioned execution entity can use global STL decomposition to decompose the above-mentioned at least one sample into trend terms, seasonal terms and residual terms, and use the trend terms, seasonal terms and residual terms as the sequence decomposition results. Each of the trend terms, seasonal terms and residual terms can be time series data decomposed from the above-mentioned at least one sample.

[0048] Optionally, the execution entity may also obtain various historical performance data within a preset time period. Each historical performance data may be time series data. Then, each historical performance data may be cleaned to obtain cleaned historical performance data. For example, historical performance data greater than a preset maximum value or less than a preset minimum value may be deleted from the historical performance data. The preset maximum value and the preset minimum value may be pre-set. Thereafter, the cleaned historical performance data may be divided to obtain a training set, a validation set, and a test set. In practice, the cleaned historical performance data may be divided in sequence using a preset division ratio to obtain a training set, a validation set, and a test set. The training set, the validation set, and the test set may be time series data, respectively. For example, the preset division ratio may be 8:1:1. The sample set may include the training set and the validation set. Specifically, when selecting a batch of samples from the sample set, the samples may be selected from the training set, and the validation set may be used to verify the model effect of the initial performance information prediction model trained for each batch.

[0049] In some optional implementations of some embodiments, the execution entity may perform sequence decomposition on the at least one sample to obtain an initial trend term, an initial seasonal term, and an initial residual term. In practice, sequence decomposition may be performed on the at least one sample using local STL decomposition to obtain the initial trend term, the initial seasonal term, and the initial residual term. Local STL decomposition may use a fixed window size to perform decomposition and introduce learnable parameters to infer each local decomposition term. Using local decomposition can improve computational efficiency.

[0050] Then, the above initial trend term, the above initial seasonal term, and the above initial residual term can be normalized respectively to obtain the normalized trend term, the normalized seasonal term, and the normalized residual term as the sequence decomposition results. In practice, the normalization can be performed using the following formula:

[0051]

[0052] in, Represents the result after normalization, which can represent the normalized trend term, the normalized seasonal term, or the normalized residual term. m represents the initial trend term, initial seasonal term or initial residual term. E[x m ] and Var[x m ] represents the corresponding mean and standard deviation. m and β m Represents the trainable mapping parameters. Therefore, through normalization, the loss caused by distribution shift can be reduced.

[0053] In addition, in order to be close to the global STL decomposition result, the loss function of the sequence decomposition layer can be expressed as:

[0054]

[0055] Among them, f m Represents a with trainable parameter θ m Function that can be used to map Represents the result of global STL decomposition.

[0056] Step 1012: Input the sequence decomposition result into the semantic extraction layer to obtain semantic sequence information.

[0057] In some embodiments, the execution entity may input the sequence decomposition result into the semantic extraction layer to obtain semantic sequence information. In practice, the execution entity may input the trend term, seasonal term, and residual term included in the sequence decomposition result into an embedding layer to obtain a semantic sequence of the trend term, a semantic sequence of the seasonal term, and a semantic sequence of the residual term as the semantic sequence information.

[0058] In some optional implementations of some embodiments, the execution subject may map the normalized trend term, the normalized seasonal term, and the normalized residual term respectively to obtain a mapped trend term, a mapped seasonal term, and a mapped residual term. In practice, the above-mentioned training parameter θ may be used. m The function f m Then, the mapped trend item, the mapped season item, and the mapped residual item can be embedded separately to obtain the semantic sequence of the trend item, the semantic sequence of the season item, and the semantic sequence of the residual item as semantic sequence information. In practice, the mapped trend item, the mapped season item, and the mapped residual item can be input into the embedding layer together to obtain the semantic sequence of the trend item, the semantic sequence of the season item, and the semantic sequence of the residual item as semantic sequence information.

[0059] Step 1013: input the semantic sequence information into the domain conversion layer to obtain converted semantic sequence information.

[0060] In some embodiments, the execution entity may input the semantic sequence information into the domain conversion layer to obtain converted semantic sequence information. In practice, the execution entity may perform domain conversion on the trend item semantic sequence and the season item semantic sequence included in the semantic sequence information to obtain a converted trend item sequence and a converted season item sequence. The converted trend item sequence, the converted season item sequence, and the residual item semantic sequence may then be determined as the converted semantic sequence information.

[0061] In some optional implementations of some embodiments, the execution entity may perform domain conversion processing on the trend item semantic sequence and the season item semantic sequence included in the semantic sequence information to obtain a converted trend item semantic sequence and a converted season item semantic sequence. In practice, the following formula may be used for domain conversion processing:

[0062]

[0063] Where DC(·) represents the frequency domain decomposition function in Fourier transform. m It represents the trend item or season item extracted through local semantics, that is, the semantic sequence of trend item or the semantic sequence of season item.

[0064] Then, the conversion trend term semantic sequence, the conversion season term semantic sequence, and the residual term semantic sequence can be determined as conversion semantic sequence information. Thus, decomposing the time series signal into its component frequencies through Fourier transform can effectively identify complex trends or seasonal patterns in the data.

[0065] Step 1014: input the converted semantic sequence information into the backbone network to obtain performance prediction information corresponding to at least one sample.

[0066] In some embodiments, the execution entity may input the conversion semantic sequence information into the backbone network to obtain performance prediction information corresponding to the at least one sample. The performance prediction information may include at least one predicted performance data, which can be understood as at least one performance data predicted based on the at least one sample. The at least one performance data may constitute time series data.

[0067] In some optional implementations of some embodiments, the execution entity may input the conversion semantic sequence information into the backbone network to obtain performance prediction information corresponding to the at least one sample through the following steps:

[0068] The first step is to generate prompt information based on the above-mentioned conversion semantic sequence information. In practice, the above-mentioned execution entity can first input the conversion semantic sequence information into a preset semi-soft prompt text template to obtain a semi-soft prompt. The preset semi-soft prompt text template may include a fixed part and a variable part. For example, the preset semi-soft prompt text template may be Predict the future time step given the [trend, season, residual]. The fixed part may be "Predict the future time step given the". The variable part may be "[trend, season, residual]". Trend is used to fill in the conversion trend item semantic sequence included in the conversion semantic sequence information. Season is used to fill in the conversion season item semantic sequence included in the conversion semantic sequence information. Residual is used to fill in the residual item semantic sequence included in the conversion semantic sequence information. Each of the trend term, seasonal term, and residual term can correspond to a semi-soft prompt. Specifically, the trend term semantic sequence is converted and input into a preset semi-soft prompt text template to obtain a semi-soft prompt for the corresponding trend term; the season term semantic sequence is converted and input into a preset semi-soft prompt text template to obtain a semi-soft prompt for the corresponding season term; and the residual term semantic sequence is input into a preset semi-soft prompt text template to obtain a semi-soft prompt for the corresponding residual term. The semi-soft prompts can then be converted into word embeddings, and then a linear transformation is performed to obtain a trainable prompt embedding as prompt information. The prompt information can include prompt embeddings for the corresponding trend term, seasonal term, and residual term, respectively.

[0069] In the second step, the above prompt information and the above conversion semantic sequence information are spliced to obtain spliced information. In practice, the prompt embedding of the corresponding trend item and the conversion trend item semantic sequence can be combined into a first sequence. The prompt embedding of the corresponding season item and the conversion season item semantic sequence can be combined into a second sequence. The prompt embedding of the corresponding residual item and the residual item semantic sequence can be combined into a third sequence. Finally, the first sequence, the second sequence, and the third sequence can be spliced in sequence into spliced information. For example, splicing can be performed using the following formula:

[0070]

[0071] Among them, P represents splicing information. Tre Indicates the prompt embedding of the corresponding trend item. Sea Represents the prompt embedding corresponding to the season item. Res represents the prompt embedding of the corresponding residual term. Represents a semantic sequence of conversion trend items. Represents the semantic sequence of transition season terms. Res Represents the semantic sequence of residual terms. Indicates a concatenation operation. ; Indicates a line break operation.

[0072] In the third step, the concatenated information is input into the backbone network to obtain feature sequence information corresponding to the at least one sample. In practice, corresponding position embeddings can be generated based on the concatenated information, or both can be input into the backbone network. The backbone network can include a Transformer. The feature sequence information can be the result of the Transformer's calculations on the concatenated information. The feature sequence information can include feature sequences corresponding to trend terms, seasonal terms, and residual terms.

[0073] The fourth step is to perform a reverse conversion process on the above-mentioned feature sequence information to obtain reverse conversion feature sequence information. In practice, the feature sequence corresponding to the trend item and the feature sequence corresponding to the season item included in the above-mentioned feature sequence information can be subjected to a domain reverse conversion process to obtain a reverse trend item feature sequence and a reverse season item feature sequence. In practice, the domain reverse conversion process can be performed using the following formula:

[0074]

[0075] Where DI(·) represents the frequency aggregation function in Fourier transform.

[0076] Then, the reverse trend item feature sequence, the reverse season item feature sequence, and the feature sequence of the corresponding season item can be determined as the inverse conversion feature sequence information.

[0077] In the fifth step, the inverse conversion feature sequence information is input into the output unit included in the backbone network to obtain performance prediction information corresponding to at least one sample.

[0078] Optionally, the historical performance data may include performance information and performance mutation information. The performance information may be an indicator value of a performance indicator, such as CPU usage. The performance mutation information may be information indicating whether the performance information at a given moment has undergone a mutation. For example, when the CPU usage exceeds a first preset ratio and the difference from the CPU usage at the previous moment exceeds a second preset ratio, the performance mutation information at that moment may be 1; otherwise, the performance mutation information at that moment may be 0. For example, the first preset ratio may be 85%, and the second preset ratio may be 50%.

[0079] In some optional implementations of some embodiments, the execution entity may provide the following steps to input the inverse-converted feature sequence information into the output unit included in the backbone network to obtain performance prediction information corresponding to at least one sample:

[0080] In the first step, the above-mentioned inverse conversion feature sequence information is input into the fully connected layer included in the above-mentioned output unit to obtain preliminary performance prediction information.

[0081] In the second step, the preliminary performance prediction information is input into the first and second fully connected layers included in the output unit, respectively, to obtain preliminary predicted performance information and preliminary predicted performance mutation information corresponding to the at least one sample. The first and second fully connected layers can be fully connected layers for performing feature conversion on the performance information and feature conversion on the performance mutation information, respectively.

[0082] The third step is to generate prediction performance information corresponding to the above-mentioned at least one sample based on the above-mentioned preliminary prediction performance information. In practice, if the decomposed trend term, seasonal term and residual term are all normalized, the preliminary prediction performance information can be first denormalized to obtain denormalized prediction performance information. Denormalization can be an inverse operation of normalization. Then, the denormalized prediction performance information can be mapped to a preset numerical range to obtain prediction performance information corresponding to the above-mentioned at least one sample. The preset numerical range can be (0,1). Specifically, the denormalized prediction performance information can be input into a sigmoid function to map the numerical range of the denormalized prediction performance information to (0,1).

[0083] The fourth step is to generate prediction performance mutation information corresponding to the at least one sample based on the preliminary prediction performance mutation information. In practice, if the decomposed trend term, seasonal term, and residual term are all normalized, the preliminary prediction performance mutation information can be first denormalized to obtain denormalized prediction performance information. Then, the denormalized prediction performance mutation information can be mapped to a preset numerical range to obtain prediction performance mutation information corresponding to the at least one sample. The preset numerical range can be (0, 1). Specifically, the denormalized prediction performance mutation information can be input into a sigmoid function to map the numerical range of the denormalized prediction performance mutation information to (0, 1).

[0084] In a fifth step, the predicted performance information and the predicted performance mutation information are determined as the performance prediction information corresponding to the at least one sample.

[0085] Step 1015: Train the initial performance information prediction model based on the performance prediction information to obtain a performance information prediction model.

[0086] In some embodiments, the execution subject may train the initial performance information prediction model based on the performance prediction information to obtain a performance information prediction model. In practice, the execution subject may use at least one sample of a preset step size after the at least one sample in the sample set as real data, and use at least one performance data included in the performance prediction information as prediction data to generate a prediction accuracy of the initial performance information prediction model, and then in response to determining that the prediction accuracy is greater than or equal to a preset threshold, the initial performance information prediction model may be determined as a trained performance information prediction model. Secondly, in response to determining that the prediction accuracy is less than the preset threshold, the network parameters of the initial performance information prediction model may be adjusted, and unused samples may be used to form a sample set, and based on the adjusted initial performance information prediction model, the training steps may be performed again to finally train and obtain a trained performance information prediction model. The methods for adjusting the network parameters may include, but are not limited to: gradient descent method (such as batch gradient descent, stochastic gradient descent, mini-batch gradient descent), adaptive learning rate method (such as AdaGrad, RMSProp).

[0087] As an example, the model structure of the trained performance information prediction model can be referred to Figure 2 . Figure 2 In [1], the performance information prediction model includes a sequence decomposition layer, a semantic extraction layer, a domain conversion layer, and a backbone network. The backbone network includes position embedding, Transformer, domain inversion, and output units. The input to the sequence decomposition layer is [x t-k , x t-k+1 ,...,x t-1 ] can represent historical time series of performance data. It can represent a performance information in the time series data obtained by trend item after sequence decomposition. It can represent a performance information in the time series data obtained by seasonal terms after sequence decomposition. It can represent a performance information of the time series data obtained by the residual term after sequence decomposition. Tre It can represent the extracted trend item semantic sequence. Sea It can represent the extracted semantic sequence of seasonal terms. Res It can represent the extracted residual semantic sequence. Tre It can represent the semantic sequence of conversion trend items. Sea It can represent the semantic sequence of the conversion season term. It should be noted that the semantic sequence of the residual term P Res Without domain conversion, P Res After the domain conversion layer, it is still P ResPosition embedding can represent the position embedding corresponding to the generated splicing information. 6X means that the Transformer structure is repeatedly established 6 times. LORA (Low-Rank Adaptation) represents the parameter fine-tuning method. Add & Normalization represents Add & Norm processing. Z represents the feature sequence information output by the backbone network. Z Tre Represents the characteristic sequence of the corresponding trend item. Sea Represents the characteristic sequence of the corresponding seasonal term. Res Represents the characteristic sequence of the corresponding residual term. Represents the prediction performance information of step h after the predicted time step k. Indicates the prediction performance mutation information with a step length of h after the prediction time step k.

[0088] In some optional implementations of some embodiments, the execution entity may provide the following steps to train the initial performance information prediction model based on the performance prediction information to obtain the performance information prediction model:

[0089] The first step is to determine whether the initial performance information prediction model has achieved the optimization goal based on the performance prediction information and the historical performance data corresponding to the performance prediction information.

[0090] In the second step, in response to determining that the initial performance information prediction model achieves the optimization goal, the initial performance information prediction model is determined as the performance information prediction model.

[0091] In the third step, in response to determining that the initial performance information prediction model has not achieved the optimization goal, the network parameters of the initial performance information prediction model are adjusted, and unused samples are used to form a sample set, and the above training steps are performed again based on the adjusted initial performance information prediction model.

[0092] In some optional implementations of some embodiments, the execution entity may determine whether the initial performance information prediction model achieves the optimization goal based on the performance prediction information and the historical performance data corresponding to the performance prediction information by performing the following steps:

[0093] The first step is to generate loss information based on the performance prediction information and the historical performance data corresponding to the performance prediction information. The performance prediction information can serve as the predicted performance data at each prediction moment. The historical performance data corresponding to the performance prediction information can serve as the actual performance data at each prediction moment. In practice, the loss value can be determined using a mean square error (MSE) approach. When the performance prediction information includes predicted performance information and predicted performance mutation information, the loss information can include the loss value of the predicted performance information and the loss value of the predicted performance mutation information.

[0094] In the second step, in response to determining that the loss information meets the preset loss condition, training effect information is generated using the validation set and the initial performance information prediction model. The preset loss condition can be that the loss information is less than a preset loss value. When the performance prediction information includes predicted performance information and predicted performance mutation information, the preset loss condition can also be that the average of the two loss values included in the loss information is less than the preset loss value. The preset loss condition is not limited here. In practice, the validation set can be input into the initial performance information prediction model to obtain performance prediction information. Training effect information can then be generated using the performance prediction information and the corresponding real-world performance data. For example, for predicted performance information, the mean absolute error can be used as the first evaluation metric, and for predicted performance mutation information, the harmonic mean of precision and recall can be used as the second evaluation metric. The ratio of the first hyperparameter to the first evaluation metric can be determined as the first fusion evaluation metric, and the product of the second hyperparameter and the second evaluation metric can be determined as the second fusion evaluation metric. Finally, the sum of the first and second fusion evaluation indicators can be determined as the fusion evaluation metric. The value of the fusion evaluation metric represents the training effect information. A larger fusion evaluation metric indicates better prediction effect.

[0095] In a third step, in response to determining that the training effect information satisfies a preset training condition, determining that the initial performance information prediction model has achieved an optimization goal. The preset training condition may be that the training effect information is greater than historical training effect information in a historical training effect information set. The historical training effect information set may be historical training effect information determined during a previous execution of the training step.

[0096] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: Through the performance information prediction model generation method of some embodiments of the present disclosure, performance prediction can be performed in advance, and complex change patterns in historical performance data can be learned, thereby reducing the impact of performance anomalies and the number of computer system downtimes. Specifically, the reasons for the possibility of unpredictable impacts and the high number of computer system downtimes are: manual inspection methods and performance detection methods that set alarm thresholds have time delays. Often, when an alarm is received, the performance anomaly has already persisted for a period of time, which may have unpredictable impacts; time series prediction methods that only consider time dependencies have difficulty learning complex change patterns, resulting in poor performance prediction results, a large number of missed performance issues, and a large number of computer system downtimes. Based on this, the performance information prediction model generation method of some embodiments of the present disclosure performs the following training steps based on at least one sample in a sample set: First, the at least one sample is input into the sequence decomposition layer included in the initial performance information prediction model to obtain a sequence decomposition result. The sample set includes historical performance data within a preset time period, and the initial performance information prediction model includes a sequence decomposition layer, a semantic extraction layer, a domain conversion layer, and a backbone network. In this way, the structure and characteristics of time series data can be combined to split complex time series data so that the model can be more easily understood and interpreted. The second step is to input the above sequence decomposition results into the above semantic extraction layer to obtain semantic sequence information. In this way, the model can capture the changes in time series more finely. The third step is to input the above semantic sequence information into the above domain conversion layer to obtain converted semantic sequence information. In this way, the time series signal can be decomposed into its component frequencies, which can effectively identify complex patterns in the data. The fourth step is to input the above converted semantic sequence information into the above backbone network to obtain performance prediction information corresponding to the above at least one sample. In this way, performance prediction can be performed through the backbone network. The fifth step is to train the initial performance information prediction model based on the above performance prediction information to obtain a performance information prediction model. In this way, historical performance data and predicted performance data can be used to train the model. Because the trained performance information prediction model can split, extract semantics, and transform domains from time series data through the sequence decomposition layer, semantic extraction layer, and domain transformation layer, it can better understand and capture the complex patterns of performance changes over time. This improves the accuracy of predicted performance information, reduces missed performance issues, and thus reduces the number of computer system downtimes. It also avoids manual inspections and performance testing methods that require setting alarm thresholds, allowing for early performance predictions and reducing the impact of performance anomalies.

[0097] Figure 3The process 300 of some embodiments of the performance information prediction method according to the present disclosure is shown. The performance information prediction method includes the following steps:

[0098] Step 301: Acquire time-series performance data.

[0099] In some embodiments, the execution entity (e.g., a server or terminal device) on which the performance information prediction method runs can obtain time-series performance data. Time-series performance data can include various performance information continuously collected within a recent period of time. Time-series performance data can also include performance mutation information.

[0100] Step 302: Input the performance data into a pre-trained performance information prediction model to obtain performance prediction information.

[0101] In some embodiments, the execution subject may input the performance data into a pre-trained performance information prediction model to obtain performance prediction information. Figure 1 The performance prediction information may include the predicted performance information at each prediction moment and the predicted performance mutation information at each prediction moment.

[0102] Step 303: Generate performance warning information based on the performance prediction information.

[0103] In some embodiments, the execution entity may generate performance alert information based on the performance prediction information. In practice, in response to the presence of performance information greater than a preset performance value among the various performance information included in the performance prediction information, first performance alert information may be generated based on the performance information and the corresponding prediction time. For example, the performance information and the corresponding prediction time may be input into a preset first performance alert template to generate the first performance alert information. For example, the preset first performance alert template may be: "CPU usage is expected to be excessively high at time XXX, with a predicted CPU usage of XXX." The first "XXX" may represent the prediction time, and the second "XXX" may represent the predicted performance information at the prediction time. Alternatively, in response to the presence of performance mutation information greater than a preset performance mutation value among the various performance mutation information included in the performance prediction information, second performance alert information may be generated based on the performance mutation information, the corresponding prediction time, and the predicted performance information. For example, the performance mutation information, the corresponding prediction time, and the predicted performance information may be input into a preset second performance alert template to generate the second performance alert information. For example, the preset second performance alarm template may be: The CPU usage is expected to suddenly change at time XXX, with a probability of "XXX" and a predicted CPU usage of XXX. The first "XXX" may represent the predicted time corresponding to the performance mutation information, the second "XXX" may represent the performance mutation information, and the third "XXX" may represent the performance information predicted at the predicted time corresponding to the performance mutation information. The first performance alarm information and / or the second performance alarm information may then be determined as performance alarm information.

[0104] Step 304: Execute a performance alarm operation according to the performance alarm information.

[0105] In some embodiments, the execution entity executes a performance alarm operation based on the performance alarm information. In practice, in response to determining that the performance alarm information includes first performance alarm information, the first performance alarm information can be sent to a pre-bound terminal device. In response to determining that the performance alarm information includes second performance alarm information, the second performance alarm information can be sent to a pre-bound terminal device. After receiving the performance alarm information, the pre-bound terminal device can prompt the corresponding user to review it, thereby enabling early response to predicted performance anomalies and preemptive remediation.

[0106] Further references Figure 4 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a performance information prediction model generation device. These device embodiments are similar to Figure 1 Corresponding to the method embodiments shown, the device can be specifically applied to various electronic devices.

[0107] like Figure 4 As shown, some embodiments of the performance information prediction model generation device 400 include: an execution unit 401. The execution unit 401 is configured to perform the following training steps based on at least one sample in a sample set: inputting the at least one sample into a sequence decomposition layer included in an initial performance information prediction model to obtain a sequence decomposition result, wherein the sample set includes historical performance data within a preset time period, and the initial performance information prediction model includes a sequence decomposition layer, a semantic extraction layer, a domain conversion layer, and a backbone network; inputting the sequence decomposition result into the semantic extraction layer to obtain semantic sequence information; inputting the semantic sequence information into the domain conversion layer to obtain converted semantic sequence information; inputting the converted semantic sequence information into the backbone network to obtain performance prediction information corresponding to the at least one sample; and training the initial performance information prediction model based on the performance prediction information to obtain a performance information prediction model.

[0108] Optionally, the prediction model generation device 400 may further include a data acquisition unit, a cleaning unit, and a partitioning unit (not shown in the figure). The data acquisition unit is configured to acquire various historical performance data within a preset time period. The cleaning unit is configured to perform a cleaning process on the various historical performance data to obtain the cleaned historical performance data. The partitioning unit is configured to partition the cleaned historical performance data to obtain a training set, a validation set, and a test set, wherein the sample set includes the training set and the validation set.

[0109] Optionally, the execution unit 401 can be further configured to: perform sequence decomposition processing on the above-mentioned at least one sample to obtain an initial trend term, an initial seasonal term and an initial residual term; and perform normalization processing on the above-mentioned initial trend term, the above-mentioned initial seasonal term and the above-mentioned initial residual term respectively to obtain a normalized trend term, a normalized seasonal term and a normalized residual term as the sequence decomposition result.

[0110] Optionally, the execution unit 401 can be further configured to: map the above-mentioned normalized trend item, the above-mentioned normalized season item and the above-mentioned normalized residual item respectively to obtain a mapped trend item, a mapped season item and a mapped residual item; embed the above-mentioned mapped trend item, the above-mentioned mapped season item and the above-mentioned mapped residual item to obtain a trend item semantic sequence, a season item semantic sequence and a residual item semantic sequence as semantic sequence information.

[0111] Optionally, the execution unit 401 can be further configured to: perform domain conversion processing on the trend item semantic sequence and the season item semantic sequence included in the above-mentioned semantic sequence information, respectively, to obtain a converted trend item semantic sequence and a converted season item semantic sequence; and determine the above-mentioned converted trend item semantic sequence, the converted season item semantic sequence and the above-mentioned residual item semantic sequence as the converted semantic sequence information.

[0112] Optionally, the execution unit 401 can be further configured to: generate prompt information based on the above-mentioned conversion semantic sequence information; splice the above-mentioned prompt information and the above-mentioned conversion semantic sequence information to obtain splicing information; input the above-mentioned splicing information into the above-mentioned backbone network to obtain feature sequence information corresponding to the above-mentioned at least one sample; perform inverse conversion processing on the above-mentioned feature sequence information to obtain inverse conversion feature sequence information; input the above-mentioned inverse conversion feature sequence information into the output unit included in the above-mentioned backbone network to obtain performance prediction information corresponding to at least one sample.

[0113] Optionally, the above historical performance data includes performance information and performance mutation information.

[0114] Optionally, the execution unit 401 can be further configured to: input the above-mentioned inverse conversion feature sequence information into the fully connected layer included in the above-mentioned output unit to obtain preliminary performance prediction information; input the above-mentioned preliminary performance prediction information into the first fully connected layer and the second fully connected layer included in the above-mentioned output unit respectively to obtain preliminary predicted performance information and preliminary predicted performance mutation information corresponding to the above-mentioned at least one sample; generate predicted performance information corresponding to the above-mentioned at least one sample based on the above-mentioned preliminary predicted performance information; generate predicted performance mutation information corresponding to the above-mentioned at least one sample based on the above-mentioned preliminary predicted performance mutation information; determine the above-mentioned predicted performance information and the above-mentioned predicted performance mutation information as performance prediction information corresponding to the above-mentioned at least one sample.

[0115] Optionally, the execution unit 401 can be further configured to: determine whether the initial performance information prediction model achieves the optimization target based on the above-mentioned performance prediction information and the various historical performance data corresponding to the above-mentioned performance prediction information; in response to determining that the initial performance information prediction model achieves the optimization target, determine the initial performance information prediction model as the performance information prediction model; in response to determining that the initial performance information prediction model does not achieve the optimization target, adjust the network parameters of the initial performance information prediction model, and adopt unused samples to form a sample set, and based on the adjusted initial performance information prediction model, perform the above-mentioned training steps again.

[0116] Optionally, the execution unit 401 can be further configured to: generate loss information based on the above-mentioned performance prediction information and the various historical performance data corresponding to the above-mentioned performance prediction information; in response to determining that the above-mentioned loss information meets the preset loss condition, generate training effect information through the verification set and the initial performance information prediction model; in response to determining that the above-mentioned training effect information meets the preset training condition, determine that the initial performance information prediction model achieves the optimization goal.

[0117] It is understandable that the units recorded in the performance information prediction model generation device 400 and the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the performance information prediction model generation device 400 and the units included therein, and will not be described in detail here.

[0118] Further references Figure 5 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a performance information prediction device. These device embodiments are similar to Figure 3 Corresponding to the method embodiments shown, the device can be specifically applied to various electronic devices.

[0119] like Figure 5 As shown, the performance information prediction device 500 of some embodiments includes: an acquisition unit 501, an input unit 502, a generation unit 503 and an alarm unit 504. The acquisition unit 501 is configured to acquire time-series performance data; the input unit 502 is configured to input the above performance data into a pre-trained performance information prediction model to obtain performance prediction information, wherein the above performance information prediction model is based on Figure 1 The corresponding embodiments are generated by the methods described; the generating unit 503 is configured to generate performance warning information according to the above performance prediction information; the warning unit 504 is configured to perform a performance warning operation according to the above performance warning information.

[0120] It is understandable that the units recorded in the performance information prediction device 500 are similar to those in the reference Figure 3 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the performance information prediction device 500 and the units included therein, and will not be described in detail here.

[0121] Reference below Figure 6, which shows a schematic diagram of the structure of an electronic device 600 (e.g., a server or terminal device) suitable for implementing some embodiments of the present disclosure. The electronic devices in some embodiments of the present disclosure may include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 6 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0122] like Figure 6 As shown, the electronic device 600 may include a processing device 601 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. Various programs and data required for the operation of the electronic device 600 are also stored in the RAM 603. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0123] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device 600 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 6 The electronic device 600 is shown with various devices, but it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed instead. Figure 6 Each block shown in the figure may represent one device, or may represent multiple devices as needed.

[0124] In particular, according to some embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from a network via the communication device 609, or installed from the storage device 608, or installed from the ROM 602. When the computer program is executed by the processing device 601, the above-mentioned functions defined in the method of some embodiments of the present disclosure are performed.

[0125] It should be noted that the computer-readable medium described in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, device, or device. In some embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0126] In some embodiments, the client and server can communicate using any currently known or future developed network protocol, such as HTTP (HyperText Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0127] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device performs the following training steps based on at least one sample in the sample set: inputting the at least one sample into the sequence decomposition layer included in the initial performance information prediction model to obtain a sequence decomposition result, wherein the sample set includes historical performance data within a preset time period, and the initial performance information prediction model includes a sequence decomposition layer, a semantic extraction layer, a domain conversion layer, and a backbone network; inputting the sequence decomposition result into the semantic extraction layer to obtain semantic sequence information; inputting the semantic sequence information into the domain conversion layer to obtain converted semantic sequence information; inputting the converted semantic sequence information into the backbone network to obtain performance prediction information corresponding to the at least one sample; and training the initial performance information prediction model based on the performance prediction information to obtain a performance information prediction model.

[0128] Or the electronic device: obtains time-series performance data; inputs the performance data into a pre-trained performance information prediction model to obtain performance prediction information, wherein the performance information prediction model is based on Figure 1 The corresponding embodiments are generated by the methods described above; based on the above performance prediction information, performance alarm information is generated; based on the above performance alarm information, a performance alarm operation is executed.

[0129] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages such as Java, Smalltalk, C++, and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0130] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0131] The units described in some embodiments of the present disclosure may be implemented in software or in hardware. The units described may also be provided in a processor. For example, they may be described as follows: a processor includes an acquisition unit, an input unit, a generation unit, and an alarm unit. The names of these units do not, in some cases, constitute limitations on the units themselves. For example, the acquisition unit may also be described as a "unit for acquiring time-series performance data."

[0132] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0133] Some embodiments of the present disclosure further provide a computer program product, including a computer program, which implements any of the above-mentioned performance information prediction model generation methods or performance information prediction methods when executed by a processor.

[0134] The above description is only an illustration of some preferred embodiments of the present disclosure and the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present disclosure.

Claims

1. A method for generating a performance information prediction model, comprising: Based on at least one sample in the sample set, perform the following training steps: Inputting the at least one sample into a sequence decomposition layer included in an initial performance information prediction model to obtain a sequence decomposition result, wherein the sample set includes historical performance data within a preset time period, and the initial performance information prediction model includes a sequence decomposition layer, a semantic extraction layer, a domain conversion layer, and a backbone network; Inputting the sequence decomposition result into the semantic extraction layer to obtain semantic sequence information; Inputting the semantic sequence information into the domain conversion layer to obtain converted semantic sequence information; Inputting the converted semantic sequence information into the backbone network to obtain performance prediction information corresponding to the at least one sample; An initial performance information prediction model is trained according to the performance prediction information to obtain a performance information prediction model.

2. The method according to claim 1, wherein The method further comprises: Obtain various historical performance data within a preset time period; Cleaning the historical performance data to obtain cleaned historical performance data; The cleaned historical performance data are divided into a training set, a validation set and a test set, wherein the sample set includes the training set and the validation set.

3. The method according to claim 1, wherein Inputting the at least one sample into a sequence decomposition layer included in the initial performance information prediction model to obtain a sequence decomposition result includes: Performing sequence decomposition processing on the at least one sample to obtain an initial trend term, an initial seasonal term, and an initial residual term; The initial trend term, the initial seasonal term and the initial residual term are respectively normalized to obtain a normalized trend term, a normalized seasonal term and a normalized residual term as a sequence decomposition result.

4. The method according to claim 3, wherein: Inputting the sequence decomposition result into the semantic extraction layer to obtain semantic sequence information includes: Mapping the normalized trend term, the normalized seasonal term, and the normalized residual term respectively to obtain a mapped trend term, a mapped seasonal term, and a mapped residual term; The mapping trend item, the mapping season item and the mapping residual item are embedded to obtain a trend item semantic sequence, a season item semantic sequence and a residual item semantic sequence as semantic sequence information.

5. The method according to claim 3, wherein: The step of inputting the semantic sequence information into the domain conversion layer to obtain converted semantic sequence information includes: Performing domain conversion processing on the trend item semantic sequence and the season item semantic sequence included in the semantic sequence information to obtain a converted trend item semantic sequence and a converted season item semantic sequence; The conversion trend item semantic sequence, the conversion season item semantic sequence and the residual item semantic sequence are determined as conversion semantic sequence information.

6. The method according to claim 1, wherein Inputting the converted semantic sequence information into the backbone network to obtain performance prediction information corresponding to the at least one sample includes: generating prompt information according to the converted semantic sequence information; splicing the prompt information and the converted semantic sequence information to obtain splicing information; Inputting the splicing information into the backbone network to obtain characteristic sequence information corresponding to the at least one sample; Performing inverse conversion processing on the characteristic sequence information to obtain inverse conversion characteristic sequence information; The inverse conversion feature sequence information is input into the output unit included in the backbone network to obtain performance prediction information corresponding to at least one sample.

7. The method according to claim 6, wherein: The historical performance data includes performance information and performance mutation information; the inputting the inverse conversion feature sequence information into the output unit included in the backbone network to obtain performance prediction information corresponding to at least one sample includes: Inputting the inverse conversion feature sequence information into the fully connected layer included in the output unit to obtain preliminary performance prediction information; Inputting the preliminary performance prediction information into the first fully connected layer and the second fully connected layer included in the output unit respectively, to obtain preliminary predicted performance information and preliminary predicted performance mutation information corresponding to the at least one sample; generating, based on the preliminary predicted performance information, predicted performance information corresponding to the at least one sample; Generating predicted performance mutation information corresponding to the at least one sample according to the preliminary predicted performance mutation information; The predicted performance information and the predicted performance mutation information are determined as performance prediction information corresponding to the at least one sample.

8. The method according to claim 1, wherein The step of training an initial performance information prediction model based on the performance prediction information to obtain a performance information prediction model includes: Determining whether the initial performance information prediction model achieves an optimization goal based on the performance prediction information and each historical performance data corresponding to the performance prediction information; In response to determining that the initial performance information prediction model achieves the optimization goal, determining the initial performance information prediction model as the performance information prediction model; In response to determining that the initial performance information prediction model does not achieve the optimization goal, the network parameters of the initial performance information prediction model are adjusted, and unused samples are used to form a sample set, and the training step is performed again based on the adjusted initial performance information prediction model.

9. The method according to claim 8, wherein The determining, based on the performance prediction information and each historical performance data corresponding to the performance prediction information, whether the initial performance information prediction model achieves the optimization goal includes: generating loss information based on the performance prediction information and each historical performance data corresponding to the performance prediction information; In response to determining that the loss information satisfies a preset loss condition, generating training effect information using a validation set and an initial performance information prediction model; In response to determining that the training effect information satisfies a preset training condition, it is determined that the initial performance information prediction model achieves an optimization goal.

10. A performance information prediction method, comprising: Obtaining time-series performance data; Inputting the performance data into a pre-trained performance information prediction model to obtain performance prediction information, wherein the performance information prediction model is generated by the method of any one of claims 1 to 9; generating performance warning information based on the performance prediction information; Perform a performance alarm operation according to the performance alarm information.

11. A device for generating a performance information prediction model, comprising: The execution unit is configured to perform the following training steps based on at least one sample in the sample set: inputting the at least one sample into a sequence decomposition layer included in an initial performance information prediction model to obtain a sequence decomposition result, wherein the sample set includes historical performance data within a preset time period, and the initial performance information prediction model includes a sequence decomposition layer, a semantic extraction layer, a domain conversion layer and a backbone network; inputting the sequence decomposition result into the semantic extraction layer to obtain semantic sequence information; inputting the semantic sequence information into the domain conversion layer to obtain converted semantic sequence information; inputting the converted semantic sequence information into the backbone network to obtain performance prediction information corresponding to the at least one sample; and training the initial performance information prediction model based on the performance prediction information to obtain a performance information prediction model.

12. A performance information prediction device, comprising: an acquisition unit, configured to acquire time-series performance data; an input unit configured to input the performance data into a pre-trained performance information prediction model to obtain performance prediction information, wherein the performance information prediction model is generated by the method according to any one of claims 1 to 9; a generating unit configured to generate performance warning information according to the performance prediction information; The alarm unit is configured to perform a performance alarm operation according to the performance alarm information.

13. An electronic device comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 9 or 10.

14. A computer-readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the method according to any one of claims 1 to 9 or 10 is implemented.

15. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 9 or 10.

Images