Waveform Recognition Method and Device for Performance Index Time-Series Data

Abstract

The present invention provides a waveform recognition method and device for time-series data of performance indicators. The method includes: collecting operation data of a target system within a target time period and determining time-series data of at least one performance indicator; based on a preset waveform recognition algorithm, performing waveform feature recognition on the determined time-series data of each performance indicator; wherein the waveform recognition algorithm includes a sudden rise and fall type recognition algorithm, a periodic type recognition algorithm, a trend type recognition algorithm, and a step type recognition algorithm; inputting the recognized waveform features into a classification model to obtain a prediction result of the waveform form of each performance indicator within the target time period output by the classification model; wherein the classification model is trained based on the waveform features of the time-series data of performance indicator samples and their labeled waveform forms. The present invention can achieve accurate prediction of the waveform form of time-series data of performance indicators.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular, to a method and device for waveform recognition of time series data of performance indicators. Background Art

[0002] Most of the existing waveform recognition methods are applied to the field of signal processing. In the field of signal processing, since the waveforms of signals are mostly periodic, the waveform recognition methods usually only apply to periodic waveforms. For example, methods such as curve fitting, wavelet analysis, and Fourier transform are used.

[0003] Among them, for waveform recognition by the method of curve fitting, the current waveform shape is fitted by a preset curve equation. This method has a high accuracy for periodic waveforms with obvious features, but has a low accuracy for other types of waveforms. For waveform recognition by the methods of wavelet analysis and Fourier transform, most of the analyzed waveforms are also periodic waveforms, which have a high accuracy for periodic waveforms and a low accuracy for other types of waveforms.

[0004] Due to the complex changes and numerous types of time series data of performance indicators, the existing waveform recognition methods applied to the field of signal processing cannot meet the requirements for waveform recognition of time series data of performance indicators. Summary of the Invention

[0005] The present invention provides a method and device for waveform recognition of time series data of performance indicators, which are used to solve the defect that the prior art lacks a waveform recognition method for time series data of performance indicators, and can achieve accurate prediction of the waveform shape of time series data of performance indicators.

[0006] In a first aspect, the present invention provides a method for waveform recognition of time series data of performance indicators, including:

[0007] Collect the operation data of the target system within the target time period, and determine the time series data of at least one performance indicator;

[0008] Based on a preset waveform recognition algorithm, perform waveform feature recognition on the time series data of each determined performance indicator; wherein, the waveform recognition algorithm includes a sudden rise and fall type recognition algorithm, a periodic type recognition algorithm, a trend type recognition algorithm, and a stepped type recognition algorithm;

[0009] Input the recognized waveform features into a classification model, and obtain the prediction result of the waveform shape of each performance indicator within the target time period output by the classification model; wherein, the classification model is trained based on the waveform features of the time series data of performance indicator samples and their labeled waveform shapes.

[0010] According to the waveform recognition method for performance index time-series data provided by the present invention, based on a preset waveform recognition algorithm, waveform feature recognition is performed on the time-series data of each determined performance index, including:

[0011] Based on a preset sudden rise and fall type recognition algorithm, sudden rise and fall type waveform feature recognition is performed on the time-series data of each performance index;

[0012] Based on a preset periodic type recognition algorithm, periodic type waveform feature recognition is performed on the time-series data of each performance index;

[0013] Based on a preset trend type recognition algorithm, trend type waveform feature recognition is performed on the time-series data of each performance index;

[0014] Based on a preset step type recognition algorithm, step type waveform feature recognition is performed on the time-series data of each performance index;

[0015] The features obtained by performing sudden rise and fall type waveform feature recognition, periodic type waveform feature recognition, trend type waveform feature recognition, and step type waveform feature recognition on the time-series data of each performance index are aggregated to obtain the waveform features of the time-series data of each performance index.

[0016] According to the waveform recognition method for performance index time-series data provided by the present invention, based on a preset sudden rise and fall type recognition algorithm, sudden rise and fall type waveform feature recognition is performed on the time-series data of each performance index, including:

[0017] Determine the extreme values in the time-series data of each performance index, and determine the average value of the time-series data of a target number near each extreme value;

[0018] Based on the determined extreme values and the average value of the time-series data nearby, determine the peak value in the time-series data of each performance index, and determine the proportion of each peak value in the target time period;

[0019] If the determined peak proportion is within a preset peak proportion threshold range, it is determined that the time-series data of the corresponding performance index has sudden rise and fall type waveform features; and / or,

[0020] Determine the average value of the time-series data of each performance index, and determine the standard deviation of the time-series data of each performance index;

[0021] Based on three times the standard deviation of the determined standard deviation and the corresponding average value, determine the range of the time-series data of each performance index;

[0022] If there is data in the time-series data of the performance index that is not within the range of the time-series data of the corresponding determined performance index, it is determined that the time-series data of the corresponding performance index has a sudden rise and fall waveform characteristic.

[0023] According to the waveform recognition method of the performance index time-series data provided by the present invention, the periodic waveform characteristic recognition of the time-series data of each performance index based on a preset periodic recognition algorithm includes:

[0024] Based on the magnitude relationship between each data in the time-series data of each performance index and its front and rear data, the peaks in the time-series data of each performance index are determined;

[0025] If the number of peaks in the time-series data of the performance index is greater than a preset first quantity threshold, the increasing subsequence, decreasing subsequence, and / or constant subsequence in the time-series data of the corresponding performance index are determined;

[0026] If the difference of the determined subsequences of the same type is a fixed value, based on the number of times the data appears in the subsequence being greater than a preset number threshold, it is determined that the time-series data of the corresponding performance index has a periodic waveform characteristic; and / or,

[0027] Based on the time-series data of each performance index, the waveform of the time-series data of the corresponding performance index is constructed, and a fast Fourier transform is performed on the constructed waveform of the time-series data of each performance index to obtain the amplitude spectrum of the time-series data of each performance index;

[0028] If the number of inflection points in the amplitude spectrum of the time-series data of the performance index is less than a preset second quantity threshold, it is determined that the time-series data of the corresponding performance index has a periodic waveform characteristic;

[0029] Based on a preset time window, the autocorrelation coefficient of the waveform of the time-series data of the performance index with the periodic waveform characteristic in different time windows is determined, and the period of the time-series data of the performance index with the periodic waveform characteristic is obtained.

[0030] According to the waveform recognition method of the performance index time-series data provided by the present invention, the trend-type waveform characteristic recognition of the time-series data of each performance index based on a preset trend-type recognition algorithm includes:

[0031] Perform a first-order polynomial fitting on the time-series data of each performance index to obtain a straight line corresponding to the time-series data of each performance index;

[0032] Determine the sum of the maximum distances by which the time-series data of each performance index deviates from the determined corresponding straight line on both sides of the straight line;

[0033] If the sum of the maximum distances by which the timing data of the determined performance metrics deviate from the corresponding straight line is less than a preset distance threshold, it is determined that the timing data of the corresponding performance metrics have a trend-type waveform feature.

[0034] According to the waveform recognition method for performance metric timing data provided by the present invention, the step-type waveform feature recognition of the timing data of each performance metric based on a preset step-type recognition algorithm includes:

[0035] Sampling the timing data of each performance metric based on a preset time window;

[0036] If the number of types of data in the sampled data is less than a preset third quantity threshold, it is determined that the timing data of the corresponding performance metric have a step-type waveform feature.

[0037] According to the waveform recognition method for performance metric timing data provided by the present invention, the convergence of the features obtained by performing sudden rise and fall type waveform feature recognition, periodic waveform feature recognition, trend-type waveform feature recognition, and step-type waveform feature recognition on the timing data of each performance metric to obtain the waveform feature of the timing data of each performance metric includes:

[0038] Converging the features obtained by performing sudden rise and fall type waveform feature recognition, periodic waveform feature recognition, trend-type waveform feature recognition, and step-type waveform feature recognition on the timing data of each performance metric, and converting them into numerical features to obtain the waveform feature vector of the timing data of each performance metric;

[0039] The inputting the recognized waveform features into a classification model to obtain the prediction result of the waveform form of each performance metric in the target time period output by the classification model includes:

[0040] Inputting the waveform feature vector of the timing data of each performance metric into the classification model to obtain the prediction result of the waveform form of each performance metric in the target time period output by the classification model.

[0041] In a second aspect, the present invention also provides a waveform recognition device for performance metric timing data, including:

[0042] An acquisition module for acquiring the operation data of the target system in the target time period and determining the timing data of at least one performance metric;

[0043] A recognition module for performing waveform feature recognition on the timing data of each determined performance metric based on a preset waveform recognition algorithm; wherein, the waveform recognition algorithm includes a sudden rise and fall type recognition algorithm, a periodic recognition algorithm, a trend-type recognition algorithm, and a step-type recognition algorithm;

[0044] A prediction module, configured to input the recognized waveform features into a classification model to obtain prediction results of the waveform morphologies of each of the performance metrics within the target time period output by the classification model; wherein, the classification model is trained based on the waveform features of the time series data of the performance metric samples and their labeled waveform morphologies.

[0045] In a third aspect, the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps of the waveform recognition method for the time series data of the performance metrics as described in any one of the above are implemented.

[0046] In a fourth aspect, the invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the waveform recognition method for the time series data of the performance metrics as described in any one of the above are implemented.

[0047] In a fifth aspect, the invention further provides a computer program product, on which a computer program is stored. When the computer program is executed by a processor, the steps of the waveform recognition method for the time series data of the performance metrics as described in any one of the above are implemented.

[0048] The waveform recognition method and device for the time series data of the performance metrics provided by the present invention, based on the characteristics of the time series data of the performance metrics, jointly recognize the waveform features of the time series data of the performance metrics by combining waveform recognition algorithms of multiple waveform types, and can accurately predict the waveform morphologies of the time series data of the performance metrics. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0050] Figure 1 is a schematic flowchart of the waveform recognition method for the time series data of the performance metrics provided by the present invention;

[0051] Figure 2A is a schematic flowchart of a recognition algorithm for a sudden rise and fall type provided by the present invention;

[0052] Figure 2B is a schematic diagram for determining the extreme values and the average values in the vicinity thereof of the time series data of the performance metrics provided by the present invention;

[0053] Figure 3It is a schematic flow chart of another sudden-rise-and-fall type recognition algorithm provided by the present invention;

[0054] Figure 4A It is a schematic flow chart of a periodic type recognition algorithm provided by the present invention;

[0055] Figure 4B It is a schematic diagram of determining the peak of the timing data of the performance index provided by the present invention;

[0056] Figure 4C It is a schematic diagram of determining the increasing subsequence of the timing data of the performance index provided by the present invention;

[0057] Figure 4D It is a schematic diagram of determining the number of occurrences of data in the timing data of the performance index provided by the present invention;

[0058] Figure 5A It is a schematic flow chart of another periodic type recognition algorithm provided by the present invention;

[0059] Figure 5B It is a schematic diagram of constructing the waveform of the timing data of the performance index provided by the present invention;

[0060] Figure 5C It is Figure 5B A schematic diagram of the waveform of the corresponding amplitude spectrum;

[0061] Figure 5D It is a schematic diagram of the ACF test provided by the present invention;

[0062] Figure 6A It is a schematic flow chart of the trend type recognition algorithm provided by the present invention;

[0063] Figure 6B It is a schematic diagram of the straight line corresponding to the first-order polynomial fitting provided by the present invention;

[0064] Figure 7A It is a schematic flow chart of the step type recognition algorithm provided by the present invention;

[0065] Figure 7B It is a schematic diagram of sampling the step type waveform provided by the present invention;

[0066] Figure 8 It is a schematic flow chart of the application field of the first method for waveform recognition of the timing data of the performance index provided by the present invention;

[0067] Figure 9 It is a schematic diagram of the composition structure of the waveform recognition device for the timing data of the performance index provided by the present invention;

[0068] Figure 10 It is a schematic diagram of the composition structure of the electronic device provided by the present invention. Detailed implementation manners

[0069] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some but not all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0070] The following combines Figures 1 - 8 to describe a waveform recognition method for performance index time series data of the present invention.

[0071] Please refer to Figure 1 , Figure 1 which is a flowchart of a waveform recognition method for performance index time series data provided by the present invention. Figure 1 The waveform recognition method for performance index time series data shown can be executed by a waveform recognition device for performance index time series data. The waveform recognition device for performance index time series data can be set in a server. For example, the server can be a physical server including an independent host, a virtual server hosted by a host cluster, a cloud server, etc. The embodiments of the present invention do not limit this. As Figure 1 shown, the waveform recognition method for performance index time series data at least includes:

[0072] 101. Collect the operation data of the target system within the target time period and determine the time series data of at least one performance index.

[0073] In the embodiments of the present invention, the target system can be a system that needs to obtain index data reflecting its performance, such as an IT information system. The embodiments of the present invention do not limit the type of the target system. The time series data of the performance index of the target system within the target time period can be obtained by collecting the operation data of the components in the target system within the target time period. Among them, the number and type of the components in the target system for collecting operation data, the target time period for collecting operation data, and the type of the performance index of the target system obtained based on the collected operation data can be set according to the purpose of obtaining the performance index. The embodiments of the present invention do not limit this. For example, when performing anomaly detection on an IT information system, the operation data of all running hardware components and software components in the IT information system for one day can be collected, and the time series data of the performance indexes of all running hardware components and software components within one day can be determined based on the collected operation data. Among them, the performance indexes of the hardware components can include the CPU occupancy rate, the number of processes, and the memory usage rate of the host, etc., and the performance indexes of the software components can include the compatibility, security, and maintainability of the software, etc.

[0074] The embodiments of the present invention do not limit the implementation method of collecting the operation data of the target system within the target time period. For example, the operation data can be collected from the hardware components and software components in the IT information system through the Agent technology. The embodiments of the present invention do not limit the implementation method of determining the time-series data of the performance indicators of the target system based on the collected operation data of the target system within the target time period. For example, the operation data collected from the hardware components and software components in the IT information system through the Agent technology can be stored in a data warehouse, and the data can be processed and aggregated to obtain the time-series data of the performance indicators of the hardware components and software components in the IT information system. Among them, the method of processing and aggregating the data when determining the data of the performance indicators can be implemented by using existing methods according to the type of the performance indicators.

[0075] 102. Based on a preset waveform recognition algorithm, perform waveform feature recognition on the time-series data of each determined performance indicator; wherein, the waveform recognition algorithm includes a sudden rise and fall type recognition algorithm, a periodic type recognition algorithm, a trend type recognition algorithm, and a stepped type recognition algorithm.

[0076] In the embodiments of the present invention, since different types of waveforms usually have different characteristics, various waveform recognition algorithms can be designed by analyzing the characteristics of various types of waveforms. Through multiple waveform recognition algorithms, the waveform features of the time-series data of each performance indicator of the target system within the target time period can be jointly recognized, and more features of the waveforms of the time-series data of each performance indicator of the target system within the target time period can be obtained. Among them, the waveform recognition algorithm can mainly include a sudden rise and fall type recognition algorithm, a periodic type recognition algorithm, a trend type recognition algorithm, a stepped type recognition algorithm, etc. The embodiments of the present invention do not limit the types and quantities of the preset waveform recognition algorithms. Among them, the sudden rise and fall type can refer to the characteristic that the time-series data has no obvious fluctuation in most time and shows a sudden increase or sudden decrease at some time points; the periodic type can refer to the characteristic that one or more types of data in the time-series data appear multiple times with the change of time and show a periodic characteristic; the trend type can refer to the characteristic that the time-series data shows a gentle upward or gentle downward trend with the increase of time; the stepped type can refer to the characteristic that the time-series data shows a stepped shape with the increase of time. The embodiments of the present invention can design corresponding waveform recognition algorithms based on the above characteristics of the sudden rise and fall type, periodic type, trend type, and stepped type waveforms. The embodiments of the present invention do not limit the implementation manners of designing the sudden rise and fall type recognition algorithm, periodic type recognition algorithm, trend type recognition algorithm, and stepped type recognition algorithm.

[0077] In some alternative examples, according to multiple pre-set waveform recognition algorithms, waveform feature recognition can be performed on the time-series data of each performance metric of the target system within the target time period, and the features obtained for the time-series data of each performance metric through multiple waveform recognition algorithms can be aggregated to obtain the waveform features of the time-series data of each performance metric. For example, the pre-set waveform recognition algorithms can be sudden rise and fall type recognition algorithm, periodic type recognition algorithm, trend type recognition algorithm, and stepped type recognition algorithm; based on the pre-set sudden rise and fall type recognition algorithm, sudden rise and fall type waveform feature recognition can be performed on the time-series data of each performance metric; based on the pre-set periodic type recognition algorithm, periodic type waveform feature recognition can be performed on the time-series data of each performance metric; based on the pre-set trend type recognition algorithm, trend type waveform feature recognition can be performed on the time-series data of each performance metric; based on the pre-set stepped type recognition algorithm, stepped type waveform feature recognition can be performed on the time-series data of each performance metric; finally, the features obtained from the sudden rise and fall type waveform feature recognition, periodic type waveform feature recognition, trend type waveform feature recognition, and stepped type waveform feature recognition of the time-series data of each performance metric can be aggregated to obtain the waveform features of the time-series data of each performance metric.

[0078] 103, input the recognized waveform features into the classification model to obtain the prediction results of the waveform morphology of each performance metric within the target time period output by the classification model; wherein, the classification model is trained based on the waveform features of the time-series data of the performance metric samples and their labeled waveform morphologies.

[0079] In the embodiments of the present invention, after performing waveform feature recognition on the time-series data of each determined performance metric according to the pre-set waveform recognition algorithms, the waveform features of the time-series data of each recognized performance metric can be input into the classification model, and the classification model can be used to recognize the waveform morphology of each performance metric within the target time period through the waveform features, so as to obtain the prediction results of the waveform morphology of each performance metric within the target time period output by the classification model. Among them, the prediction results of the waveform morphology output by the classification model can be the type of the waveform, which can be one of the waveform types corresponding to the pre-set waveform recognition algorithms. For example, the waveform recognition algorithms include sudden rise and fall type recognition algorithm, periodic type recognition algorithm, trend type recognition algorithm, and stepped type recognition algorithm, and the waveform morphology of each performance metric within the target time period output by the classification model can be one of the following waveform types: sudden rise and fall type, periodic type, trend type, and stepped type.

[0080] In an embodiment of the present invention, a classification model for identifying the waveform morphology of a performance indicator may be obtained through supervised training using the waveform features of the time series data of the performance indicator sample and the waveform morphology it annotates. The embodiment of the present invention does not limit the type of classification model and the implementation method for training the classification model. For example, the classification model may select existing models such as xgboost, random forest, and catboost. The embodiment of the present invention does not limit the method for obtaining performance indicator samples. For example, performance indicator samples may be obtained by precipitating historical data and manually annotating the historical data.

[0081] In some optional examples, the features obtained by performing sudden rise and fall waveform feature recognition, periodic waveform feature recognition, trend waveform feature recognition and step waveform feature recognition on the time series data of each performance indicator can be aggregated and converted into numerical features to obtain a waveform feature vector of the time series data of each performance indicator. The waveform feature vector of the time series data of each performance indicator can be input into a classification model to obtain a prediction result of the waveform shape of each performance indicator output by the classification model within the target time period. The embodiment of the present invention does not limit the implementation method of converting the features aggregated by each performance indicator into numerical features and the form of the obtained waveform feature vector. For example, one-hot encoding can be used to convert the features aggregated by each performance indicator into numerical features, thereby obtaining a waveform feature vector composed of 0 and 1 for each performance indicator.

[0082] The waveform recognition method for performance indicator time series data provided in an embodiment of the present invention is based on the characteristics of the time series data of the performance indicator. By combining waveform recognition algorithms of multiple waveform types, the waveform characteristics of the time series data of the performance indicator are jointly identified, thereby realizing accurate prediction of the waveform morphology of the time series data of the performance indicator.

[0083] See also Figure 2A , Figure 2A FIG. 1 is a flow chart of a sudden rise and fall type recognition algorithm provided by the present invention, such as Figure 2A As shown, based on a preset sudden rise and fall type recognition algorithm, the sudden rise and fall type waveform feature recognition of the time series data of each performance indicator at least includes:

[0084] 201, determine the extreme value in the time series data of each performance indicator, and determine the average value of the time series data of the target quantity near each extreme value.

[0085] 202 , based on the determined extreme value and the average value of the time series data near it, determine the peak value in the time series data of each performance indicator, and determine the proportion of each peak value in the target time period.

[0086] 203. If the determined peak ratio is within the pre-set peak ratio threshold range, it is determined that the time series data of the corresponding performance metric has a sudden rise and fall waveform characteristic.

[0087] In the embodiments of the present invention, first, by screening the time series data of each performance metric, all extreme values, i.e., local maximum values, in the time series data of each performance metric can be determined. Then, according to the determined extreme values, N time series data are selected near each extreme value, and the average value of the time series data near the extreme value is calculated based on the selected N time series data, as Figure 2B shown. After that, according to the extreme value and its nearby average value, the ratio of the extreme value to the average value can be calculated. If the ratio of the extreme value to the average value is greater than K or less than 1 / K, the extreme value can be determined as the peak in the time series data of the corresponding performance metric, where it is greater than K when the peak is upward and less than 1 / K when the peak is downward. If there are peaks in the time series data of the performance metric, the time ratio of each peak in the target time period can be calculated. By comparing the time ratio of each peak with the pre-set peak ratio threshold range, when the time ratio of the peak is within the pre-set peak ratio threshold range, it can be considered that the time series data of the performance metric corresponding to the peak has a sudden rise and fall waveform characteristic.

[0088] Among them, the method for screening extreme values in the time series data of each performance metric and the method for calculating the time ratio of each peak in the target time period can be implemented by selecting appropriate methods, which are not limited in the embodiments of the present invention. The values of N and K, and the range of the pre-set peak ratio threshold range, can be set according to experience and are not limited in the embodiments of the present invention. For example, N can be taken as 6, K can be taken as 5, and the pre-set peak ratio threshold range can be [0.01, 0.05].

[0089] Please refer to Figure 3 , Figure 3 which is a schematic flowchart of another sudden rise and fall type recognition algorithm provided by the present invention, as Figure 3 shown. Based on the pre-set sudden rise and fall type recognition algorithm, the recognition of the sudden rise and fall waveform characteristic of the time series data of each performance metric at least includes:

[0090] 301. Determine the average value of the time series data of each performance metric and determine the standard deviation of the time series data of each performance metric.

[0091] 302. Based on three times the standard deviation of the determined standard deviation and the corresponding average value, determine the range of the time series data of each performance metric.

[0092] 303. If there is data in the time-series data of a performance metric that is not within the range of the time-series data of the corresponding performance metric determined, it is determined that the time-series data of the corresponding performance metric has a sudden rise and fall waveform characteristic.

[0093] In an embodiment of the present invention, it can be first assumed that the time-series data of each performance metric follows a normal distribution. Thus, according to the time-series data of each performance metric, the mean μ and standard deviation σ of the time-series data of each performance metric are calculated. Then, based on the mean μ and three times the standard deviation σ of the time-series data of each performance metric, the range of the time-series data of each performance metric is calculated, i.e., [μ - 3σ, μ + 3σ]. After that, the time-series data of each performance metric is compared with the calculated range of the time-series data of this performance metric. When there is data in the time-series data of a performance metric that is not within the calculated range of the time-series data of this performance metric, it can be considered that the time-series data of this performance metric has a sudden rise and fall waveform characteristic.

[0094] Please refer to Figure 4A , Figure 4A which is a schematic flowchart of a periodic identification algorithm provided by the present invention. As Figure 4A shown, based on a pre-set periodic identification algorithm, the periodic waveform characteristic identification of the time-series data of each performance metric at least includes:

[0095] 401. Based on the magnitude relationship between each data in the time-series data of each performance metric and its adjacent data before and after, the peaks in the time-series data of each performance metric are determined.

[0096] 402. If the number of peaks in the time-series data of a performance metric is greater than a pre-set first quantity threshold, the increasing subsequence, decreasing subsequence, and / or constant subsequence in the time-series data of the corresponding performance metric are determined.

[0097] 403. If the difference of the determined subsequences of the same type is a fixed value, based on the number of times the data appears in the subsequence being greater than a pre-set number threshold, it is determined that the time-series data of the corresponding performance metric has a periodic waveform characteristic.

[0098] In an embodiment of the present invention, it can be first determined, according to the fact that if the waveform of the time-series data of a performance metric is periodic, the time-series data of this performance metric must have the characteristics of peaks and valleys, the peaks in the time-series data of each performance metric are determined by calculating the magnitude relationship between each data in the time-series data of each performance metric and its adjacent data before and after. When the number of peaks in the time-series data of a performance metric is greater than M, it can be preliminarily determined that the time-series data of this performance metric has a periodic waveform characteristic, as Figure 4B shown.

[0099] Then, for the time series data of the performance metric initially determined to have a periodic waveform feature, each data in the time series data can be compared with the subsequent data. By judging the magnitude relationship between each data and the subsequent data, it can be determined whether there are increasing subsequences, decreasing subsequences, and / or constant subsequences in the time series data. When there are increasing subsequences, decreasing subsequences, and / or constant subsequences in the time series data, the increasing subsequences, decreasing subsequences, and / or constant subsequences in the time series data can be extracted, as Figure 4C shown.

[0100] After that, the data at both ends of each subsequence extracted from the same time series data can be differentiated. Since if the waveform of the time series data of a performance metric is periodic, the result of differentiating the data at both ends of each increasing subsequence extracted from the time series data of the performance metric is a fixed value, the result of differentiating the data at both ends of each decreasing subsequence is a fixed value, and / or the result of differentiating the data at both ends of each constant subsequence is a fixed value. Therefore, according to the result of differentiating the increasing subsequence, the result of differentiating the decreasing subsequence, and / or the result of differentiating the constant subsequence, the judgment window of the increasing subsequence, the judgment window of the decreasing subsequence, and / or the judgment window of the constant subsequence can be determined respectively, as Figure 4D shown. Thus, it can be determined whether the number of times a data appears in the determined judgment window of the increasing subsequence, the judgment window of the decreasing subsequence, or the judgment window of the constant subsequence is greater than a preset number threshold. When the number of times a data appears in the determined judgment window of the increasing subsequence, the judgment window of the decreasing subsequence, or the judgment window of the constant subsequence is greater than the preset number threshold, it can be considered that the time series data of the performance metric has a periodic waveform feature.

[0101] Among them, the value of M and the value of the preset number threshold can be set according to experience, and the embodiments of the present invention do not limit this. For example, M can be taken as 4, and the preset number threshold can also be taken as 4.

[0102] Please refer to Figure 5A , Figure 5A which is a schematic flowchart of another periodicity recognition algorithm provided by the present invention, as Figure 5A shown. Based on the preset periodicity recognition algorithm, the periodic waveform feature recognition of the time series data of each performance metric at least includes:

[0103] 501. Based on the time series data of each performance metric, construct the waveform of the corresponding performance metric time series data, and perform a fast Fourier transform on the constructed waveform of each performance metric time series data to obtain the amplitude spectrum of the time series data of each performance metric.

[0104] 502. If the number of inflection points in the amplitude spectrum of the time-series data of a performance metric is less than a preset second quantity threshold, it is determined that the time-series data of the corresponding performance metric has a periodic waveform characteristic.

[0105] 503. Based on a preset time window, determine the autocorrelation coefficient of the waveform of the time-series data of the performance metric with a periodic waveform characteristic in different time windows, and obtain the period of the time-series data of the performance metric with a periodic waveform characteristic.

[0106] Since any continuous periodic signal can be composed of a set of appropriate sine curves, although a sine curve cannot be combined into a signal with sharp corners, it can be very closely approximated by a sine curve, and the approximation can be made such that there is no energy difference between the two representations. The Fourier transform can represent a certain function that satisfies certain conditions as a trigonometric function, such as a sine and / or cosine function, or a linear combination of their integrals, and the Fourier transform can transform a signal in the time domain to the frequency domain, and the frequency of the signal can be directly obtained through the frequency-domain image.

[0107] Therefore, in the embodiments of the present invention, first, the waveform of the time-series data of each performance metric can be constructed according to the time-series data of each performance metric, as Figure 5B shown, and the fast Fourier transform is performed on the waveform of the time-series data of each constructed performance metric to obtain the amplitude spectrum of the time-series data of each performance metric, as Figure 5C shown. Then, the number of inflection points in the amplitude spectrum can be determined through the amplitude spectrum of the time-series data of each performance metric. In Figure 5C , the frequencies corresponding to the inflection points are 10, 14, 18, and 22, indicating that Figure 5B the waveform in is composed of the superposition of sine functions with frequencies of 10, 14, 18, and 22. When the number of inflection points in the amplitude spectrum is less than a preset second quantity threshold, it can be preliminarily considered that the time-series data of the performance metric has a periodic waveform characteristic. Among them, the value of the preset second quantity threshold can be set according to experience, and the embodiments of the present invention do not limit this. For example, the preset second quantity threshold can be taken as 3.

[0108] Since the coefficient of the autocorrelation function (Autocorrelation Function, abbreviated as ACF) can reflect the autocorrelation characteristics of a sequence by calculating the correlation degree of the sequence in different windows, and a periodic sequence has a high degree of autocorrelation. Therefore, in the embodiments of the present invention, after preliminarily considering that the time-series data of the performance metric has a periodic waveform characteristic, the ACF can be used to further verify the periodic waveform characteristic of the time-series data of the performance metric. As Figure 5D shown, the ACF test can be realized by calculating the correlation of the waveform of the time-series data of the performance metric in different time windows w-1 and w-2.

[0109] Since the periodic sequence has the same periodic characteristics as the waveform itself, autocorrelation can help determine the existence of a period; meanwhile, if the periodic waveform contains a significant sine wave, the period of this sine wave is the period of the periodic waveform, and the frequency of this sine wave can be found through the fast Fourier transform. The larger the Fourier coefficient, the more likely the period of the corresponding sine wave is the period of the periodic waveform. Therefore, the embodiments of the present invention can more accurately determine that the timing data of the performance index has a periodic waveform characteristic by combining the fast Fourier transform and the autocorrelation function.

[0110] Please refer to Figure 6A , Figure 6A which is a schematic flowchart of the trend recognition algorithm provided by the present invention. As Figure 6A shown, based on the pre-set trend recognition algorithm, the trend waveform feature recognition of the timing data of each performance index at least includes:

[0111] 601. Perform a first-order polynomial fitting on the timing data of each performance index to obtain a straight line corresponding to the timing data of each performance index.

[0112] 602. Determine the sum of the maximum distances by which the timing data of each performance index deviates from the straight line on both sides of the determined corresponding straight line.

[0113] 603. If the sum of the maximum distances by which the determined timing data of the performance index deviates from the corresponding straight line is less than the pre-set distance threshold, it is determined that the timing data of the corresponding performance index has a trend waveform feature.

[0114] In the embodiments of the present invention, first, through the polynomial fitting algorithm, a first-order polynomial fitting can be performed on the timing data of each performance index to obtain a straight line d corresponding to the timing data of each performance index, as Figure 6B shown. Then, the sum of the maximum distances by which the timing data of each performance index deviates from the straight line d on both sides can be calculated, that is, the sum of the maximum distances D = d1 + d2 in the direction of the normal vector of the straight line d. Then, the sum of the maximum distances D in the direction of the normal vector of the calculated straight line d can be compared with the pre-set distance threshold. When the sum of the maximum distances D in the direction of the normal vector of the calculated straight line d is less than the pre-set distance threshold, it can be considered whether the timing data of the corresponding performance index has a trend waveform feature. Among them, the value of the pre-set distance threshold can be set according to experience, and the embodiments of the present invention do not limit this. For example, the pre-set distance threshold can be taken as 1.

[0115] Please refer to Figure 7A , Figure 7A which is a schematic flowchart of the step-type recognition algorithm provided by the present invention. As Figure 7AAs shown, based on a preset stepped recognition algorithm, the stepped waveform feature recognition of the time series data of each performance metric at least includes:

[0116] 701. Sample the time series data of each performance metric based on a preset time window.

[0117] 702. If the number of data types in the sampled data is less than a preset third quantity threshold, determine that the time series data of the corresponding performance metric has a stepped waveform feature.

[0118] In the embodiments of the present invention, first, the time series data of each performance metric can be sampled according to a preset time window, as Figure 7B shown. Then, the number of data types in the sampled data can be counted. When the number of data types in the sampled data is less than the preset third quantity threshold, it can be considered that the time series data of this performance metric has a stepped waveform feature. Among them, the value of the preset third quantity threshold can be set according to experience, and the embodiments of the present invention do not limit this. For example, the preset third quantity threshold can be 3.

[0119] Please refer to Figure 8 , Figure 8 which is a schematic flowchart of an application scenario of the waveform recognition method for the time series data of performance metrics provided by the present invention. As Figure 8 shown, first, the original operation data can be regularly collected from the hardware components and software components in the enterprise IT information system through the Agent technology, and the collected operation data can be stored in the data warehouse, processed and aggregated to obtain the time series data of the performance metrics of the IT information system. Then, the obtained time series data of the performance metrics can be distributed in parallel to the four major analysis centers of sudden rise and fall type analysis, periodic type analysis, trend type analysis, and stepped type analysis. The waveform recognition algorithms are run in each center to recognize various waveform features of the time series data of the performance metrics. Among them, the sudden rise and fall type analysis center includes two sudden rise and fall type recognition algorithms, namely peak analysis or gradation analysis and 3 - standard - deviation detection; the periodic type analysis center also includes two periodic type recognition algorithms, namely periodic analysis and Fourier transform combined with ACF test; the trend type analysis center and the stepped type analysis center each include one trend type recognition algorithm and one stepped type recognition algorithm. After obtaining the recognition results of the sudden rise and fall type waveform features, periodic type waveform features, trend type waveform features, and stepped type waveform features of the time series data of each performance metric through the four major analysis centers, the recognition results of each waveform feature of the time series data of each performance metric can be aggregated according to Table 1 and numerical conversion can be performed.

[0120] Table 1

[0121]

[0122] Among them, data1, data2, data3... represent performance indicator 1, performance indicator 2, performance indicator 3..., burst-1 represents the peak analysis algorithm, burst-2 represents the 3 times standard deviation detection, period-1 represents the period analysis algorithm, period-2 represents the Fourier transform combined with the ACF test, smooth represents a trend type recognition algorithm, discrete represents a step type recognition algorithm, 1 represents meeting this feature, and 0 represents not meeting this feature.

[0123] After that, the waveform feature of each performance indicator obtained based on Table 1 can be input into the classification model to obtain the prediction result of the waveform form of each performance indicator output by the classification model. The waveform forms of each performance indicator predicted by the classification model are shown in Table 2.

[0124] Table 2

[0125] Waveform category data1 Sudden rise and fall type data2 Periodic type data3 Trend type …… ……

[0126] Next, the waveform recognition device for performance indicator time series data provided by the present invention will be described. The waveform recognition device for performance indicator time series data described below can be correspondingly referred to the waveform recognition method for performance indicator time series data described above.

[0127] Please refer to Figure 9 , Figure 9 which is the schematic diagram of the composition structure of the waveform recognition device for performance indicator time series data provided by the present invention. Figure 9 The waveform recognition device for performance indicator time series data shown can be set in the server, and can be used to execute Figure 1 the waveform recognition method for performance indicator time series data. For example, the server can be a physical server including an independent host, a virtual server hosted by a host cluster, a cloud server, etc. The embodiments of the present invention do not limit this. As Figure 10 shown, the waveform recognition device for performance indicator time series data at least includes:

[0128] The acquisition module 910 is used to acquire the operation data of the target system within the target time period and determine the time series data of at least one performance indicator.

[0129] The recognition module 920 is used to perform waveform feature recognition on the determined time series data of each performance indicator based on the preset waveform recognition algorithm; among them, the waveform recognition algorithm includes the sudden rise and fall type recognition algorithm, the periodic type recognition algorithm, the trend type recognition algorithm, and the step type recognition algorithm.

[0130] A prediction module 930, configured to input the recognized waveform features into a classification model to obtain the prediction results of the waveform patterns of each performance metric output by the classification model within a target time period; wherein, the classification model is trained based on the waveform features of the time series data of the performance metric samples and their labeled waveform patterns.

[0131] Optionally, the recognition module 920 includes:

[0132] A sudden rise and fall feature recognition unit, configured to perform sudden rise and fall waveform feature recognition on the time series data of each performance metric based on a preset sudden rise and fall recognition algorithm.

[0133] A periodic feature recognition unit, configured to perform periodic waveform feature recognition on the time series data of each performance metric based on a preset periodic recognition algorithm.

[0134] A trend feature recognition unit, configured to perform trend waveform feature recognition on the time series data of each performance metric based on a preset trend recognition algorithm.

[0135] A stepped feature recognition unit, configured to perform stepped waveform feature recognition on the time series data of each performance metric based on a preset stepped recognition algorithm.

[0136] A feature aggregation unit, configured to aggregate the features obtained by performing sudden rise and fall waveform feature recognition, periodic waveform feature recognition, trend waveform feature recognition, and stepped waveform feature recognition on the time series data of each performance metric to obtain the waveform features of the time series data of each performance metric.

[0137] Optionally, the sudden rise and fall feature recognition unit includes:

[0138] A peak analysis sub-unit, configured to determine the extreme values in the time series data of each performance metric and determine the average value of the time series data of a target number of data points near each extreme value;

[0139] Based on the determined extreme values and the average values of the time series data near them, determine the peaks in the time series data of each performance metric and determine the proportion of each peak in the target time period;

[0140] If the determined peak proportion is within a preset peak proportion threshold range, determine that the time series data of the corresponding performance metric has sudden rise and fall waveform features. And / or,

[0141] A three-standard deviation detection sub-unit, configured to determine the average value of the time series data of each performance metric and determine the standard deviation of the time series data of each performance metric;

[0142] Based on three times the standard deviation of the determined standard deviation and the corresponding average value, determine the range of the time series data of each performance metric;

[0143] If there is data in the time series data of the performance index that is not within the range of the time series data of the corresponding determined performance index, it is determined that the time series data of the corresponding performance index has a sudden rise and fall waveform feature.

[0144] Optionally, the periodic feature recognition unit includes:

[0145] A sequence analysis subunit, configured to determine the peaks in the time series data of each performance index based on the magnitude relationship between any value in the time series data of each performance index and its previous and subsequent values;

[0146] If the number of peaks in the time series data of the performance index is greater than a preset first quantity threshold, the increasing subsequence, decreasing subsequence and / or constant subsequence in the time series data of the corresponding performance index are determined;

[0147] If the difference of the determined subsequences of the same type is a fixed value, based on the number of times the data appears in the subsequence being greater than a preset number threshold, it is determined that the time series data of the corresponding performance index has a periodic waveform feature. And / or,

[0148] A Fourier transform and ACF test subunit, configured to construct the waveform of the time series data of the corresponding performance index based on the time series data of each performance index, and perform a fast Fourier transform on the constructed waveform of the time series data of each performance index to obtain the amplitude spectrum of the time series data of each performance index;

[0149] If the number of inflection points in the amplitude spectrum of the time series data of the performance index is less than a preset second quantity threshold, it is determined that the time series data of the corresponding performance index has a periodic waveform feature;

[0150] Based on a preset time window, determine the autocorrelation coefficient of the waveform of the time series data of the performance index with a periodic waveform feature in different time windows, and obtain the period of the time series data of the performance index with a periodic waveform feature.

[0151] Optionally, a trend feature recognition unit, configured to perform a first-order polynomial fitting on the time series data of each performance index to obtain a straight line corresponding to the time series data of each performance index;

[0152] Determine the sum of the maximum distances by which the time series data of each performance index deviates from the determined corresponding straight line on both sides of the straight line;

[0153] If the sum of the maximum distances by which the determined time series data of the performance index deviates from the corresponding straight line is less than a preset distance threshold, it is determined that the time series data of the corresponding performance index has a trend waveform feature.

[0154] Optionally, a step-type feature recognition unit is configured to sample the time-series data of each performance metric based on a preset time window;

[0155] If the number of data types in the sampled data is less than a preset third quantity threshold, it is determined that the time-series data of the corresponding performance metric has a step-type waveform feature.

[0156] Optionally, a feature aggregation unit is configured to aggregate the features obtained by performing sudden-rise-and-fall type waveform feature recognition, periodic waveform feature recognition, trend-type waveform feature recognition, and step-type waveform feature recognition on the time-series data of each performance metric, and convert them into numerical features to obtain a waveform feature vector of the time-series data of each performance metric.

[0157] A prediction module 930 is configured to input the waveform feature vector of the time-series data of each performance metric into a classification model to obtain a prediction result of the waveform form of each performance metric within a target time period output by the classification model.

[0158] Figure 10 An example of a schematic physical structure diagram of an electronic device is shown as Figure 10 shown. The electronic device may include: a processor 1010, a communication interface 1020, a memory 1030, and a communication bus 1040. Among them, the processor 1010, the communication interface 1020, and the memory 1030 complete mutual communication through the communication bus 1040. The processor 1010 may call logical instructions in the memory 1030 to execute a waveform recognition method for time-series data of performance metrics. The method includes:

[0159] Collect the operation data of the target system within the target time period, and determine the time-series data of at least one performance metric;

[0160] Based on a preset waveform recognition algorithm, perform waveform feature recognition on the time-series data of each determined performance metric; wherein, the waveform recognition algorithm includes a sudden-rise-and-fall type recognition algorithm, a periodic recognition algorithm, a trend-type recognition algorithm, and a step-type recognition algorithm;

[0161] Input the recognized waveform features into a classification model to obtain a prediction result of the waveform form of each performance metric within the target time period output by the classification model; wherein, the classification model is trained based on the waveform features of the time-series data of performance metric samples and their labeled waveform forms.

[0162] In addition, when the logical instructions in the above-mentioned memory 1030 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0163] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute the waveform recognition method for performance index time series data provided in the above-mentioned various embodiments. The method includes:

[0164] Collect the operation data of the target system within the target time period and determine the time series data of at least one performance index;

[0165] Based on a preset waveform recognition algorithm, perform waveform feature recognition on the time series data of each determined performance index; wherein, the waveform recognition algorithm includes a sudden rise and fall type recognition algorithm, a periodic type recognition algorithm, a trend type recognition algorithm, and a stepped type recognition algorithm;

[0166] Input the recognized waveform features into a classification model to obtain the prediction results of the waveform forms of each of the performance indexes within the target time period output by the classification model; wherein, the classification model is trained based on the waveform features of the time series data of performance index samples and their labeled waveform forms.

[0167] On yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is configured to execute the waveform recognition method for performance index time series data provided in the above-mentioned various embodiments. The method includes:

[0168] Collect the operation data of the target system within the target time period and determine the time series data of at least one performance index;

[0169] Based on a preset waveform recognition algorithm, perform waveform feature recognition on the time-series data of each determined performance metric; wherein, the waveform recognition algorithm includes a sudden rise and fall type recognition algorithm, a periodic type recognition algorithm, a trend type recognition algorithm, and a stepped type recognition algorithm;

[0170] Input the recognized waveform features into a classification model to obtain the prediction results of the waveform forms of each of the performance metrics within the target time period output by the classification model; wherein, the classification model is trained based on the waveform features of the time-series data of performance metric samples and their labeled waveform forms.

[0171] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0172] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution, in essence, or the part that contributes to the prior art can be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0173] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A waveform recognition method for time series data of performance metrics, characterized in that, it includes: Collect the operation data of the target system within the target time period, and determine the time series data of at least one performance metric; The performance metrics include the CPU occupancy rate of the host, the number of processes, and the memory usage rate; Based on a preset waveform recognition algorithm, perform waveform feature recognition on the time series data of each determined performance metric; wherein, the waveform recognition algorithm includes a sudden rise and fall type recognition algorithm, a periodic type recognition algorithm, a trend type recognition algorithm, and a stepped type recognition algorithm; Input the recognized waveform features into a classification model to obtain the prediction result of the waveform form of each performance metric within the target time period output by the classification model; wherein, the classification model is trained based on the waveform features of the time series data of performance metric samples and their labeled waveform forms; The performing waveform feature recognition on the time series data of each determined performance metric based on a preset waveform recognition algorithm includes: Based on a preset sudden rise and fall type recognition algorithm, perform sudden rise and fall type waveform feature recognition on the time series data of each performance metric; Based on a preset periodic type recognition algorithm, perform periodic type waveform feature recognition on the time series data of each performance metric; Based on a preset trend type recognition algorithm, perform trend type waveform feature recognition on the time series data of each performance metric; Based on a preset stepped type recognition algorithm, perform stepped type waveform feature recognition on the time series data of each performance metric; Converge the features obtained by performing sudden rise and fall type waveform feature recognition, periodic type waveform feature recognition, trend type waveform feature recognition, and stepped type waveform feature recognition on the time series data of each performance metric to obtain the waveform features of the time series data of each performance metric; The performing sudden rise and fall type waveform feature recognition on the time series data of each performance metric based on a preset sudden rise and fall type recognition algorithm includes: Determine the extreme values in the time series data of each performance metric, and determine the average value of the time series data of a target number near each extreme value; Based on the determined extreme values and the average value of the time series data nearby, determine the peak value in the time series data of each performance metric, and determine the proportion of each peak value in the target time period; If the determined peak proportion is within a preset peak proportion threshold range, it is determined that the time series data of the corresponding performance metric has sudden rise and fall type waveform features; and / or, Determine the average value of the time series data of each performance metric, and determine the standard deviation of the time series data of each performance metric; Based on three times the standard deviation of the determined standard deviation and the corresponding average value, determine the range of the time series data of each performance metric; If there is data in the time series data of the performance metric that is not within the determined range of the time series data of the corresponding performance metric, it is determined that the time series data of the corresponding performance metric has sudden rise and fall type waveform features; The performing periodic type waveform feature recognition on the time series data of each performance metric based on a preset periodic type recognition algorithm includes: Based on the magnitude relationship between each data and its preceding and succeeding data in the time-series data of each of the performance metrics, determine the peaks in the time-series data of each of the performance metrics; If the number of peaks in the time-series data of the performance metric is greater than a preset first quantity threshold, determine the increasing subsequence, decreasing subsequence, and / or constant subsequence in the time-series data of the corresponding performance metric; If the difference of the determined subsequences of the same type is a fixed value, based on the number of times the data appears in the subsequence being greater than a preset number threshold, determine that the time-series data of the corresponding performance metric has a periodic waveform characteristic; and / or, Construct the waveform of the time-series data of the corresponding performance metric based on the time-series data of each of the performance metrics, and perform a fast Fourier transform on the constructed waveform of the time-series data of each performance metric to obtain the amplitude spectrum of the time-series data of each of the performance metrics; If the number of inflection points in the amplitude spectrum of the time-series data of the performance metric is less than a preset second quantity threshold, determine that the time-series data of the corresponding performance metric has a periodic waveform characteristic; Based on a preset time window, determine the autocorrelation coefficient of the waveform of the time-series data of the performance metric having the periodic waveform characteristic in different time windows, and obtain the period of the time-series data of the performance metric having the periodic waveform characteristic; The trend-type waveform feature recognition of the time-series data of each of the performance metrics based on a preset trend-type recognition algorithm includes: Perform a first-order polynomial fitting on the time-series data of each of the performance metrics to obtain a straight line corresponding to the time-series data of each of the performance metrics; Determine the sum of the maximum distances by which the time-series data of each of the performance metrics deviates from the determined corresponding straight line on both sides of the straight line; If the sum of the maximum distances by which the determined time-series data of the performance metric deviates from the corresponding straight line is less than a preset distance threshold, determine that the time-series data of the corresponding performance metric has a trend-type waveform characteristic; The step-type waveform feature recognition of the time-series data of each of the performance metrics based on a preset step-type recognition algorithm includes: Based on a preset time window, sample the time-series data of each of the performance metrics; If the number of types of data in the sampled data is less than a preset third quantity threshold, determine that the time-series data of the corresponding performance metric has a step-type waveform characteristic.

2. The waveform recognition method for time-series data of performance metrics according to claim 1, wherein, The convergence of the features obtained by performing sudden rise and fall type waveform feature recognition, periodic waveform feature recognition, trend type waveform feature recognition, and step type waveform feature recognition on the time-series data of each of the performance metrics to obtain the waveform feature of the time-series data of each of the performance metrics includes: Converge the features obtained by performing sudden rise and fall type waveform feature recognition, periodic waveform feature recognition, trend type waveform feature recognition, and step type waveform feature recognition on the time-series data of each of the performance metrics, and convert them into numerical features to obtain the waveform feature vector of the time-series data of each of the performance metrics; Inputting the recognized waveform features into a classification model to obtain the prediction results of the waveform patterns of each of the performance metrics output by the classification model during the target time period, including: Inputting the waveform feature vectors of the time series data of each of the performance metrics into the classification model to obtain the prediction results of the waveform patterns of each of the performance metrics output by the classification model during the target time period.

3. A waveform recognition device for time series data of performance metrics Characterized in that it includes: An acquisition module, configured to acquire the operation data of the target system during the target time period and determine the time series data of at least one performance metric; The performance metrics include the CPU occupancy rate of the host, the number of processes, and the usage rate of the memory; A recognition module, configured to perform waveform feature recognition on the determined time series data of each performance metric based on a preset waveform recognition algorithm; wherein, the waveform recognition algorithm includes a sudden rise and fall type recognition algorithm, a periodic type recognition algorithm, a trend type recognition algorithm, and a stepped type recognition algorithm; A prediction module, configured to input the recognized waveform features into a classification model to obtain the prediction results of the waveform patterns of each of the performance metrics output by the classification model during the target time period; wherein, the classification model is trained based on the waveform features of the time series data of performance metric samples and their labeled waveform patterns; The recognition module includes: A sudden rise and fall type feature recognition unit, configured to perform sudden rise and fall type waveform feature recognition on the time series data of each performance metric based on a preset sudden rise and fall type recognition algorithm; A periodic type feature recognition unit, configured to perform periodic type waveform feature recognition on the time series data of each performance metric based on a preset periodic type recognition algorithm; A trend type feature recognition unit, configured to perform trend type waveform feature recognition on the time series data of each performance metric based on a preset trend type recognition algorithm; A stepped type feature recognition unit, configured to perform stepped type waveform feature recognition on the time series data of each performance metric based on a preset stepped type recognition algorithm; A feature aggregation unit, configured to aggregate the features obtained by performing sudden rise and fall type waveform feature recognition, periodic type waveform feature recognition, trend type waveform feature recognition, and stepped type waveform feature recognition on the time series data of each performance metric to obtain the waveform features of the time series data of each performance metric; The sudden rise and fall type feature recognition unit includes: a peak analysis subunit, configured to determine the extreme values in the time series data of each performance metric, and determine the average value of the time series data of the target quantity near each extreme value; based on the determined extreme values and the average value of the time series data nearby, determine the peaks in the time series data of each performance metric, and determine the proportion of each peak in the target time period; if the determined peak proportion is within a pre-set peak proportion threshold range, determine that the time series data of the corresponding performance metric has a sudden rise and fall type waveform feature; and / or, a three-standard deviation detection subunit, configured to determine the average value of the time series data of each performance metric, and determine the standard deviation of the time series data of each performance metric; based on three times the standard deviation of the determined standard deviation and the corresponding average value, determine the range of the time series data of each performance metric; if there is data in the time series data of the performance metric that is not within the determined range of the time series data of the corresponding performance metric, determine that the time series data of the corresponding performance metric has a sudden rise and fall type waveform feature; The periodic type feature recognition unit includes: a sequence analysis subunit, configured to determine the peaks in the time series data of each performance metric based on the magnitude relationship between any value in the time series data of each performance metric and its previous and subsequent values; if the number of peaks in the time series data of the performance metric is greater than a pre-set first quantity threshold, determine the increasing subsequence, decreasing subsequence and / or constant subsequence in the time series data of the corresponding performance metric; if the difference between the determined subsequences of the same type is a fixed value, based on the number of times the data appears in the subsequence being greater than a pre-set number threshold, determine that the time series data of the corresponding performance metric has a periodic type waveform feature; and / or, a Fourier transform and ACF test subunit, configured to construct the waveform of the time series data of the corresponding performance metric based on the time series data of each performance metric, and perform a fast Fourier transform on the constructed waveform of the time series data of each performance metric to obtain the amplitude spectrum of the time series data of each performance metric; if the number of inflection points in the amplitude spectrum of the time series data of the performance metric is less than a pre-set second quantity threshold, determine that the time series data of the corresponding performance metric has a periodic type waveform feature; based on a pre-set time window, determine the autocorrelation coefficient of the waveform of the time series data of the performance metric with a periodic type waveform feature in different time windows, and obtain the period of the time series data of the performance metric with a periodic type waveform feature; The trend type feature recognition unit is configured to perform a first-order polynomial fitting on the time series data of each performance metric to obtain a straight line corresponding to the time series data of each performance metric; determine the sum of the maximum distances by which the time series data of each performance metric deviates from the straight line on both sides of the determined corresponding straight line; if the sum of the maximum distances by which the determined time series data of the performance metric deviates from the corresponding straight line is less than a pre-set distance threshold, determine that the time series data of the corresponding performance metric has a trend type waveform feature; A stepped feature recognition unit is configured to sample the time series data of each performance metric based on a preset time window; if the number of data types in the sampled data is less than a preset third quantity threshold, it is determined that the time series data of the corresponding performance metric has a stepped waveform feature.

4. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, when the processor executes the program, the steps of the waveform recognition method for the time series data of the performance metric according to any one of claims 1 to 2 are implemented.

5. A non-transitory computer-readable storage medium, on which a computer program is stored, wherein, when the computer program is executed by a processor, the steps of the waveform recognition method for the time series data of the performance metric according to any one of claims 1 to 2 are implemented.

Images