CN114168419A - Load prediction method, system, device and computer readable storage medium - Google Patents
Load prediction method, system, device and computer readable storage medium Download PDFInfo
- Publication number
- CN114168419A CN114168419A CN202111498189.8A CN202111498189A CN114168419A CN 114168419 A CN114168419 A CN 114168419A CN 202111498189 A CN202111498189 A CN 202111498189A CN 114168419 A CN114168419 A CN 114168419A
- Authority
- CN
- China
- Prior art keywords
- prediction
- load
- historical
- trend
- online
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/3003—Monitoring arrangements specially adapted to the computing system or computing system component being monitored
- G06F11/3006—Monitoring arrangements specially adapted to the computing system or computing system component being monitored where the computing system is distributed, e.g. networked systems, clusters, multiprocessor systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/30—Monitoring
- G06F11/3051—Monitoring arrangements for monitoring the configuration of the computing system or of the computing system component, e.g. monitoring the presence of processing resources, peripherals, I/O links, software programs
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computing Systems (AREA)
- Physics & Mathematics (AREA)
- Quality & Reliability (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a load prediction method, a system, equipment and a computer readable storage medium, wherein the load prediction method comprises the following steps: acquiring historical tracking load data in a cloud application system, and constructing a historical load sequence; performing offline trend prediction of the load on the historical load sequence to generate a prediction result based on the offline trend prediction; online fluctuation prediction is carried out on the historical load sequence to generate a prediction result based on the online fluctuation prediction; and combining the prediction result based on the offline trend prediction and the prediction result based on the online fluctuation prediction to integrate the comprehensive prediction result. On one hand, the reliability of the model of the cloud load height fluctuation scene is improved, the prediction model realized based on the sliding window can adapt to different load modes, and the model combination strategy is integrated to realize the maximum performance of the advantages of the model; on the other hand, the load prediction tool which can be practically applied to the service resource management system is realized, and executable middleware support is practically provided.
Description
Technical Field
The present invention relates to a load prediction method, system, device and computer readable storage medium, and more particularly, to a load prediction method, system, device and computer readable storage medium.
Background
In the existing load prediction method design facing cloud-native, three defects still exist:
firstly, the cloud load highly fluctuates, and load historical data stored in a time series form is usually chaotic, so that the cloud load highly fluctuates is limited by the characteristics of a chaotic time series, and the performance of a model predicted only based on original load sequence data is usually difficult to continuously improve;
secondly, the size of the sliding window influences the performance of the model, an overlarge sliding window may cause the model to mine irrelevant load information, detail information of load change cannot be concerned, a too small sliding window may cause the model to fail to mine the dependency relationship of a load sequence, and performance jitter of the model is easily caused, most of the existing prediction schemes adopt the sliding window with a fixed size obtained based on experience or experiments to train and predict the model, and the mode is difficult to adapt to highly dynamic load change;
and thirdly, an integration strategy of an integration model is adopted, the existing load prediction model usually adopts a weighted average or simple strategy based on errors, the difference of the performance of a base predictor on a time dimension is not concerned, and the advantages of the integration model are not exerted to the maximum extent.
It is worth noting that the load prediction technology based on time series decomposition focuses on the cloud load height fluctuation problem, and the wavelet decomposition and reconstruction technology is the most popular. The method decomposes a chaotic time sequence by utilizing a scalable translation wavelet to obtain a low-frequency component and a high-frequency component of an original sequence. The low frequency component is a description of the trend information of the original time series, and the high frequency component is a description of the detail information of the original time series. And then respectively establishing prediction models for different components, and integrating the prediction results of the different component models at each prediction moment to obtain a final prediction result. However, it has been proved that the detail information obtained by the time series decomposition technique such as wavelet decomposition has significant variance and noise, so that adding detail components to the prediction model is not obviously helpful, and in extreme cases, the prediction performance of the model is even reduced.
Therefore, how to provide a load prediction method, system, device and computer readable storage medium to solve the defects that the model reliability of the cloud load height fluctuation scene is insufficient, the prediction model realized based on the fixed sliding window cannot adapt to different load modes, the integrated model combination strategy is difficult to maximize the model advantages, and the like in the existing cloud-native-oriented micro-service load prediction technology exists, and the technical problem to be solved by the technical personnel in the field is really urgent.
Disclosure of Invention
In view of the foregoing disadvantages of the prior art, an object of the present invention is to provide a load prediction method, system, device and computer-readable storage medium, for solving the problems that the reliability of a model of a cloud load highly fluctuating scene is insufficient, a prediction model implemented based on a fixed sliding window cannot adapt to different load modes, and an integrated model combination strategy is difficult to maximize the advantages of the model in the existing cloud-native-oriented micro service load prediction technology.
To achieve the above and other related objects, an aspect of the present invention provides a load prediction method, including: acquiring historical tracking load data in a cloud application system, and constructing a historical load sequence; performing offline trend prediction of the load on the historical load sequence to generate a prediction result based on the offline trend prediction; online fluctuation prediction is carried out on the historical load sequence to generate a prediction result based on the online fluctuation prediction; and combining the prediction result based on the offline trend prediction and the prediction result based on the online fluctuation prediction to integrate the comprehensive prediction result.
In an embodiment of the present invention, the historical tracking load data includes various loads of the cloud application system; the various loads of the cloud application system include request loads and/or resource loads.
In an embodiment of the present invention, the step of performing offline trend prediction of load on the historical load sequence to generate a prediction result based on the offline trend prediction includes: extracting low-frequency information of a historical load sequence; and performing trend prediction on the next future cycle based on the historical trend information of the previous cycle to predict a prediction result based on the offline trend prediction.
In an embodiment of the present invention, the online fluctuation prediction of the historical load sequence to generate a prediction result based on the online fluctuation prediction includes: at the initial moment, based on the minimum size of a preset sliding window, expanding the size of the window according to the trend correlation of adjacent windows, wherein the size of the window is expanded to be twice of the original size, repeatedly expanding until the adjacent windows are not related in trend any more, and stopping expanding; judging whether the expanded current window size is larger than the boundary size of a preset sliding window; if yes, the recent load has obvious trend, and the minimum size of a preset sliding window is adopted as the size of the finally selected sliding window; if not, the recent load has no obvious trend, and the next step is carried out: the window size is extended based on the temporal correlation of the historical load sequence.
In an embodiment of the present invention, the step of expanding the window size based on the time correlation of the historical load sequence comprises: based on the maximum size of a preset sliding window, calculating the sequence time correlation of the recent historical load sequence in the window view to obtain a sequence autocorrelation function value; gradually increasing the time lag value until the time lag size of the first time crossing the confidence interval of the sequence autocorrelation function value is found, and taking the time lag size of the first time crossing the confidence interval as the expanded current window size; judging whether the expanded current window size is larger than the boundary size of a preset sliding window; if yes, indicating that the period load has obvious time correlation, and taking the expanded current window size as the finally selected sliding window size; if not, indicating that the recent load does not have obvious time correlation, and directly adopting the maximum size of a preset sliding window as the size of the finally selected sliding window; and constructing a training sample based on the finally selected sliding window size for training to predict the online load fluctuation of the next period.
In an embodiment of the present invention, the step of integrating the comprehensive prediction result by combining the prediction result based on the offline trend prediction and the prediction result based on the online fluctuation prediction includes: constructing error samples of historical M moments based on a prediction result of off-line trend prediction, a prediction result of on-line fluctuation prediction and load real data; the input characteristic of the error sample is a sample matrix comprising 2 rows and M columns; wherein the first behavior is based on a prediction error of an offline trend prediction, and the second behavior is based on a prediction error of an online fluctuation prediction; the label of the error sample is a base predictor type with better performance, namely the error of the off-line trend base predictor is smaller and is 0, and the error of the on-line fluctuation base predictor is smaller and is 1; according to error sample labels of historical M moments, calculating the historical performance stability of the base predictor; comparing the historical performance stability of the base predictor with a stability threshold to judge whether the historical prediction is stable; if so, based on the time locality (if a certain base predictor in the recent past can well predict the load condition in the near future, the base predictor is likely to still give a good prediction result in the near future), a weight distribution algorithm is adopted to endow each sample in the error samples with different weight coefficients so as to train a first multi-class regression model; predicting the weight coefficients of the offline trend base predictor and the online fluctuation base predictor according to the trained first multi-type regression model and the error input characteristic matrix at the current moment; integrating comprehensive prediction results based on the weight coefficients and the load prediction results of the base predictors; if not, training a second multi-class regression model directly according to the error samples of the historical M moments; and predicting the weight coefficients of the offline trend base predictor and the online fluctuation base predictor according to the second multi-type regression model and the error input characteristic matrix at the current moment, and further integrating the comprehensive prediction result.
In an embodiment of the present invention, the labels engraved on the error samples at the historical M moments are label sequences consisting of 1 and/or 0; the step of calculating the historical performance stability of the basis predictor comprises the following steps: calculating the variance corresponding to the label sequence and the reciprocal of the variance; and the reciprocal of the variance corresponding to the label sequence is the historical performance stability of the base predictor.
Another aspect of the present invention provides a load prediction system, including: the data acquisition module is used for acquiring historical tracking load data in the cloud application system and constructing a historical load sequence; the off-line prediction module is used for carrying out off-line trend prediction on the load of the historical load sequence so as to generate a prediction result based on the off-line trend prediction; the online prediction module is used for performing online fluctuation prediction on the historical load sequence to generate a prediction result based on the online fluctuation prediction; and the integration module is used for combining the prediction result based on the off-line trend prediction and the prediction result based on the on-line fluctuation prediction to integrate the comprehensive prediction result.
Yet another aspect of the present invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the load prediction method.
A final aspect of the present invention provides a load prediction apparatus comprising: a processor and a memory; the memory is configured to store a computer program and the processor is configured to execute the computer program stored by the memory to cause the load prediction apparatus to perform the load prediction method.
As described above, the load prediction method, system, device and computer readable storage medium according to the present invention have the following advantages:
according to the load prediction method, the system, the equipment and the computer-readable storage medium, on one hand, the model reliability of the cloud load height fluctuation scene is improved, the prediction model realized based on the sliding window can adapt to different load modes, and the model combination strategy is integrated to realize the maximum performance of the model advantages; on the other hand, the load prediction tool which can be practically applied to the service resource management system is realized, and executable middleware support is practically provided.
Drawings
Fig. 1 is a flowchart illustrating a load prediction method according to an embodiment of the invention.
Fig. 2 is a schematic flow chart of S13 in the load prediction method of the present invention.
FIG. 3 is a schematic diagram illustrating a multi-class regression integration strategy for considering temporal locality of a base predictor proposed by the present invention.
Fig. 4 is a schematic structural diagram of a load prediction system according to an embodiment of the invention.
Description of the element reference numerals
1 load prediction system
51 data acquisition module
52 offline prediction module
53 on-line prediction module
54 integrated module
S11-S14
Detailed Description
The embodiments of the present invention are described below with reference to specific embodiments, and other advantages and effects of the present invention will be easily understood by those skilled in the art from the disclosure of the present specification. The invention is capable of other and different embodiments and of being practiced or of being carried out in various ways, and its several details are capable of modification in various respects, all without departing from the spirit and scope of the present invention. It is to be noted that the features in the following embodiments and examples may be combined with each other without conflict.
It should be noted that the drawings provided in the following embodiments are only for illustrating the basic idea of the present invention, and the components related to the present invention are only shown in the drawings rather than drawn according to the number, shape and size of the components in actual implementation, and the type, quantity and proportion of the components in actual implementation may be changed freely, and the layout of the components may be more complicated.
Example one
The present embodiment provides a load prediction method, including:
acquiring historical tracking load data in a cloud application system, and constructing a historical load sequence;
performing offline trend prediction of the load on the historical load sequence to generate a prediction result based on the offline trend prediction;
online fluctuation prediction is carried out on the historical load sequence to generate a prediction result based on the online fluctuation prediction;
and combining the prediction result based on the offline trend prediction and the prediction result based on the online fluctuation prediction to integrate the comprehensive prediction result.
The following description will focus on the detailed description of the load prediction method provided by the present embodiment. Referring to fig. 1, a flow chart of a load prediction method in an embodiment is shown. As shown in fig. 1, the load prediction method specifically includes the following steps:
and S11, acquiring historical tracking load data in the cloud application system, and constructing a historical load sequence.
In the present embodiment, the history load sequence is constructed with t as a period. The historical tracking load data comprises various loads of the cloud application system; the various loads of the cloud application system include a request load and/or a resource load (CPU, Memory, Disk), and the like.
And S12, performing offline trend prediction of the load on the historical load sequence to generate a prediction result based on the offline trend prediction. In this embodiment, the WAVE-SARIMA model may be used to perform offline trend prediction on the historical load sequence with a period of T (one T period includes a plurality of T periods).
The S12 includes:
s121, extracting low-frequency information of the historical load sequence;
specifically, low-frequency information, i.e., trend information, of the history load sequence is extracted based on the wavelet decomposition technique.
And S122, performing trend prediction on the next future cycle based on the historical trend information of the previous cycle to predict a prediction result based on the offline trend prediction.
Specifically, based on trend information extracted by the wavelet decomposition model of the previous period, the SARIMA model is used for learning trend changes of the load in an off-line mode and performing off-line trend prediction on the trend of the next period in the future, and a prediction result based on the off-line trend prediction is generated.
And S13, performing online fluctuation prediction on the historical load sequence to generate a prediction result based on the online fluctuation prediction.
In the embodiment, the OnS-GBDT model can be utilized to perform online fluctuation prediction on the historical load sequence with the period of t as the period. In the embodiment, the online fluctuation prediction is to train and predict the original load sequence, so that the high-frequency information obtained by wavelet decomposition usually contains too much noise, and the improvement effect on the model capability is small. Therefore, an online prediction model is constructed based on the original load sequence, and since the purpose of the step (2) of offline trend prediction is to control the future load change trend, the online regression prediction model can be effectively assisted. Specifically, an adaptive sliding window algorithm implemented based on trend correlation and time correlation is firstly proposed to enable the online regression model to have reliable prediction performance. Further, an adaptive sliding window algorithm is combined with the GBDT online regression model to realize single-step online prediction of the load. Thus, the method performs online surge prediction based on the original load sequence. In order to enable the model to efficiently adapt to the dynamic changes of the load sequence.
In this embodiment, first, three variables sw are preset for the effective size of the sliding windowmin，swth，swmaxRespectively represents the minimum size of the sliding window, the boundary size of the sliding window and the maximum size of the sliding window, and the relationship among the three is swmin<swth<swmax
Please refer to fig. 2, which shows a flowchart of S13. As shown in fig. 2, the S13 includes:
the window size is extended based on the trend correlation of the historical load sequence.
The method specifically comprises the following steps:
at the initial moment, based on the minimum size sw of the preset sliding windowminExpanding the size of the window according to the trend correlation of the adjacent windows, wherein the size of the window is expanded to be twice of the original size, and repeating the expansion until the adjacent windows are not related to each other in trend any more, and stopping the expansion; at this time, the expanded current window size is recorded as swcur。
Judging whether the expanded current window size is larger than the boundary size sw of the preset sliding windowth(ii) a If yes, the recent load has obvious trend, so that the possibility that the historical load sequence keeps the trend in the future is high, and at the moment, the preset sliding is adoptedThe minimum size of the window is taken as the final selected sliding window size, since the recent window trend is consistent with the recent load trend, using only swminThe load trend can be learned by the samples under the size of the sliding window, so that the prediction accuracy of the model can be ensured under the condition of reducing the training overhead. If not, the expanded current window size swcur<swthThen it means that the recent load has no obvious trend and go to the next step. In this embodiment, if the historical data is not sufficiently analyzed, the model will have difficulty mining the dependencies between useful load data.
The time correlation expansion window size based on the historical load sequence specifically includes:
maximum size sw based on preset sliding windowmaxAnd calculating the sequence time correlation of the recent historical load sequence in the view field of the window to obtain a sequence autocorrelation function value. It is obvious that it is not necessary to start with a time lag of 1, but rather (sw)cur+1) the autocorrelation function values for the sequence are calculated at the beginning of the time lag because trend correlation is more critical to the sequence data relationship than time correlation.
Gradually increasing the time lag value until finding the time lag size of the first time crossing the confidence interval of the sequence autocorrelation function value, and taking the time lag size of the first time crossing the confidence interval as the expanded current window size swcur；
Judging the expanded current window size swcurWhether the boundary size sw is larger than the boundary size sw of the preset sliding windowth(ii) a If so, it indicates that the payload has a significant time correlation, and thus there is a high likelihood that the payload sequence will retain this time correlation in the future. At this time, the expanded current window size swcurAs the finally selected sliding window size, because the trend of the historical load is not significant, it is desirable to be able to sufficiently exploit the time correlation of the load sequence data to ensure the prediction accuracy of the model; if not, indicating that the recent load does not have significant time correlation, then it can be concluded that the recent load is highly dynamic and random. At this point, preset is directly adoptedMaximum size sw of sliding windowmaxAs the final selected sliding window size. Therefore, the historical load sequence information is fully mined by the model, and the excessive jitter of the model prediction effect is prevented.
Finally, after the adaptive sliding window algorithm determines the optimal sliding window size based on the historical load sequence at each moment, the GBDT model constructs a training sample based on the finally selected sliding window size for training so as to predict the online load fluctuation of the next period.
And S14, combining the prediction result based on the off-line trend prediction and the prediction result based on the on-line fluctuation prediction, and integrating the comprehensive prediction result.
Referring to fig. 3, a schematic diagram of the multi-class regression integration strategy for considering temporal locality of the base predictor proposed in S14 is shown. As shown in FIG. 3, the integration strategy of the comprehensive prediction result is to convert the 'evaluation base predictor performance problem' into a 'multi-class regression problem', the multi-class regression problem gives the probability whether the observed value belongs to a group of classes, firstly, a base predictor performance matrix PV is constructed by using the historical prediction errors of the base predictor, PV is the basic element of the training and prediction input data set of the multi-class regression model in the step of evaluating the base predictor, PV is a 2 x m characteristic matrix formed by the prediction errors of the base predictor in the historical prediction, and the first row and the second row are the prediction errors of the WAVE-SARIMA model and the OnS-GBDT model at historical m moments respectively. Therefore, a training data set and a prediction input data set of a training classification model are constructed through PV matrixes at a plurality of historical moments, a multi-class regression model is generated and then used for predicting the probability that each base predictor at the next moment is the best predictor, and then the calculated probability is used as a weighting coefficient of the prediction result of the base predictor to obtain the comprehensive prediction result of the integrated model.
Specifically, the S14 includes the following steps:
constructing error samples of historical M moments based on a prediction result of off-line trend prediction, a prediction result of on-line fluctuation prediction and load real data; the input characteristic of the error sample is a matrix comprising 2 rows and M columns, wherein M is larger than 1; wherein the first behavior is based on a prediction error of an offline trend prediction, and the second behavior is based on a prediction error of an online fluctuation prediction; the label of the error sample is a base predictor type with better performance, namely the error of the off-line trend base predictor is smaller and is 0, and the error of the on-line fluctuation base predictor is smaller and is 1;
according to the error sample labels at the historical M moments, calculating the historical performance stability of the base predictor;
comparing the historical performance stability of the base predictor with a stability threshold to judge whether the historical prediction is stable; if so, based on the time locality (if a certain base predictor in the recent past can well predict the load condition in the near future, the base predictor is likely to still give a good prediction result in the near future), a weight distribution algorithm is adopted to endow each sample in the error samples with different weight coefficients so as to train a first multi-class regression model; predicting the weight coefficients of the offline trend base predictor and the online fluctuation base predictor according to the trained first multi-type regression model and the error input characteristic matrix at the current moment; integrating comprehensive prediction results based on the weight coefficients and prediction results of all the basis predictors; if not, training a second multi-class regression model directly according to the error samples of the historical M moments; and predicting the weight coefficients of the offline trend base predictor and the online fluctuation base predictor according to the second multi-type regression model and the error input characteristic matrix at the current moment, and further integrating the comprehensive prediction result.
In the present embodiment, the fact that the basis predictors exhibit temporal locality means that if a certain basis predictor in the recent past can predict the load situation well in the near future, the basis predictor is likely to still give a good prediction result in the near future. Therefore, if the base predictor exhibits temporal locality within a certain time period, then the prediction accuracy of the integrated model can be enhanced by using an effective integration strategy. In order to measure whether the base predictor performance has time locality in a historical period of time, the prediction method provides a measure of the stability of the base predictor performance for measuring the stability of the base predictor which is the optimal predictor in a past period of time, namely the reciprocal of the corresponding variance of the label sequence of the predictor with the optimal historical performance is represented, the greater the variance value is, the greater the fluctuation of data is, the more the base predictor performance tends to be random, and the weaker the time locality is; the smaller the variance value, the less fluctuating the data, the more stable the base predictor performance and the more significant the temporal locality. And further determining whether the updating considers the time locality of the base predictor performance or not based on the PPD of the calculated base predictor performance stability. According to the time locality theorem, the importance degree of each base predictor performance matrix in time is consistent with the distance of the base predictor performance matrix from a prediction point. In short, the more important the performance matrix is, the more the corresponding weight is, the more distant the performance matrix is, the less important the performance matrix is, the less the corresponding weight is. Therefore, when it is determined that the base predictor performance time locality needs to be considered, each sample in the training sample set PVs can be given different weight, and the method realizes weight distribution of the training samples based on the sigmoid function. The time locality is expressed through the proposed base predictor, so that the multi-class regression integration strategy can have a more important training sample with a perceived attention, and the performance and the generalization capability of the integration model are improved.
The load prediction method improves the model reliability of the cloud load height fluctuation scene on one hand, the prediction model realized based on the sliding window can adapt to different load modes, and the model combination strategy is integrated to realize the maximum performance of the model advantages; on the other hand, the load prediction tool which can be practically applied to the service resource management system is realized, and executable middleware support is practically provided.
The present embodiment also provides a computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements the load prediction method as described in fig. 1.
The present application may be embodied as systems, methods, and/or computer program products, in any combination of technical details. The computer program product may include a computer-readable storage medium having computer-readable program instructions embodied thereon for causing a processor to implement various aspects of the present application.
The computer readable storage medium may be a tangible device that can hold and store the instructions for use by the instruction execution device. The computer readable storage medium may be, for example, but is not limited to, an electronic memory device, a magnetic memory device, an optical memory device, an electromagnetic memory device, a semiconductor memory device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a Static Random Access Memory (SRAM), a portable compact disc read-only memory (CD-ROM), a Digital Versatile Disc (DVD), a memory stick, a floppy disk, a mechanical coding device, such as punch cards or in-groove projection structures having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media as used herein is not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., optical pulses through a fiber optic cable), or electrical signals transmitted through electrical wires.
The computer-readable programs described herein may be downloaded from a computer-readable storage medium to a variety of computing/processing devices, or to an external computer or external storage device via a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. The network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in the respective computing/processing device. The computer program instructions for carrying out operations of the present application may be assembly instructions, Instruction Set Architecture (ISA) instructions, machine related instructions, microcode, firmware instructions, state setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C + + or the like and procedural programming languages, such as the "C" programming language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of 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 the connection may be made to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, the electronic circuitry can execute computer-readable program instructions to implement aspects of the present application by utilizing state information of the computer-readable program instructions to personalize the electronic circuitry, such as a programmable logic circuit, a Field Programmable Gate Array (FPGA), or a Programmable Logic Array (PLA).
Example two
The present embodiment provides a load prediction system, including:
the data acquisition module is used for acquiring historical tracking load data in the cloud application system and constructing a historical load sequence;
the off-line prediction module is used for carrying out off-line trend prediction on the load of the historical load sequence so as to generate a prediction result based on the off-line trend prediction;
the online prediction module is used for performing online fluctuation prediction on the historical load sequence to generate a prediction result based on the online fluctuation prediction;
and the integration module is used for combining the prediction result based on the off-line trend prediction and the prediction result based on the on-line fluctuation prediction to integrate the comprehensive prediction result.
The load prediction system of the present embodiment will be described in detail with reference to the drawings. Please refer to fig. 4, which is a schematic structural diagram of a load prediction system in an embodiment. As shown in fig. 4, the load forecasting system 4 includes a data obtaining module 41, an offline forecasting module 42, an online forecasting module 43, and an integrating module 44.
The data obtaining module 41 is configured to obtain historical tracking load data in the cloud application system, and construct a historical load sequence.
In the present embodiment, the history load sequence is constructed with t as a period. The historical tracking load data comprises various loads of the cloud application system; the various loads of the cloud application system include a request load and/or a resource load (CPU, Memory, Disk), and the like.
The offline prediction module 42 is configured to perform offline trend prediction of load on the historical load sequence to generate a prediction result based on the offline trend prediction. In this embodiment, the offline prediction module 42 may perform offline trend prediction on the historical load sequence with a period of T (a T period includes a plurality of T periods) by using the WAVE-SARIMA model as shown in the upper half of fig. 2.
The offline prediction module 42 extracts low frequency information of the historical load sequence; and performing trend prediction on the next future cycle based on the historical trend information of the previous cycle to predict a prediction result based on the offline trend prediction.
The online prediction module 43 is configured to perform online fluctuation prediction on the historical load sequence to generate a prediction result based on the online fluctuation prediction.
In this embodiment, the online prediction module 43 may perform online fluctuation prediction on the historical load sequence with a period of t by using OnS-GBDT model as shown in the lower half of FIG. 2. In this embodiment, the online fluctuation prediction is based on the original load sequence as the training and prediction object, because the high-frequency information obtained by wavelet decomposition usually contains too much noise, and has little effect on improving the model capability. Therefore, an online prediction model is constructed based on the original load sequence, and since the purpose of the step (2) of offline trend prediction is to control the future load change trend, the online regression prediction model can be effectively assisted. Specifically, an adaptive sliding window algorithm implemented based on trend correlation and time correlation is firstly proposed to enable the online regression model to have reliable prediction performance. Further, an adaptive sliding window algorithm is combined with the GBDT online regression model to realize single-step online prediction of the load. Thus, the method performs online surge prediction based on the original load sequence. In order to enable the model to efficiently adapt to the dynamic changes of the load sequence.
In this embodiment, first, three variables sw are preset to effectively define the window sizemin，swth，swmaxRespectively represents the minimum size of the sliding window, the boundary size of the sliding window and the maximum size of the sliding window, and the relationship among the three is swmin<swth<swmax
The online prediction module 43 expands the window size based on the trend correlation of the historical load sequence; the window size is extended based on the temporal correlation of the historical load sequence.
Specifically, at the initial moment, the online prediction module 43 is based on the minimum size sw of the preset sliding windowminExpanding the size of the window according to the trend correlation of the adjacent windows, wherein the size of the window is expanded to be twice of the original size, and repeating the expansion until the adjacent windows are not related to each other in trend any more, and stopping the expansion; at this time, the expanded current window size is recorded as swcur. Judging whether the expanded current window size is larger than the boundary size sw of the preset sliding windowth(ii) a If yes, the recent load is represented to have obvious trend, so that the possibility that the historical load sequence keeps the trend in the future is high, at the moment, the minimum size of the preset sliding window is adopted as the finally selected sliding window size, because the trend of the recent window is consistent with the recent load trend, and only the sw is usedminThe load trend can be learned by the samples under the size of the sliding window, so that the prediction accuracy of the model can be ensured under the condition of reducing the training overhead. If not, the expanded current window size swcur<swthThen it means that the recent load has no obvious trend and go to the next step. In this embodiment, if the historical data is not sufficiently analyzed, the model will have difficulty mining the dependencies between useful load data.
The online prediction module 43 is based on the maximum size sw of a preset sliding windowmaxCalculating the order of recent historical load sequences within the window field of viewAnd (5) listing time correlation to obtain a sequence autocorrelation function value. It is obvious that it is not necessary to start with a time lag of 1, but rather (sw)cur+1) the autocorrelation function values for the sequence are calculated at the beginning of the time lag because trend correlation is more critical to the sequence data relationship than time correlation. Gradually increasing the time lag value until finding the time lag size of the first time crossing the confidence interval of the sequence autocorrelation function value, and taking the time lag size of the first time crossing the confidence interval as the expanded current window size swcur(ii) a Judging the expanded current window size swcurWhether the boundary size sw is larger than the boundary size sw of the preset sliding windowth(ii) a If so, it indicates that the payload has a significant time correlation, and thus there is a high likelihood that the payload sequence will retain this time correlation in the future. At this time, the expanded current window size swcurAs the finally selected sliding window size, because the trend of the historical load is not significant, it is desirable to be able to sufficiently exploit the time correlation of the load sequence data to ensure the prediction accuracy of the model; if not, indicating that the recent load does not have significant time correlation, then it can be concluded that the recent load is highly dynamic and random. At this time, the maximum size sw of the preset sliding window is directly adoptedmaxAs the final selected sliding window size. Therefore, the historical load sequence information is fully mined by the model, and the excessive jitter of the model prediction effect is prevented.
Finally, after the adaptive sliding window algorithm determines the optimal sliding window size based on the historical load sequence at each moment, the GBDT model constructs a training sample based on the finally selected sliding window size for training so as to predict the online load fluctuation of the next period.
The integration module 44 is configured to combine the prediction result based on the offline trend prediction and the prediction result based on the online fluctuation prediction to integrate the comprehensive prediction result.
In this embodiment, the integration strategy of the integrated prediction result of the integration module 44 is to convert the "evaluation basis predictor performance problem" into a "multi-class regression problem" which gives the probability of whether the observed value belongs to a group of classes, and first, a basis predictor performance matrix PV is constructed by using the history prediction errors of the basis predictor, PV is the basic element of the training and prediction input data set of the multi-class regression model in the evaluation basis predictor step, and is a 2 × m feature matrix composed of the prediction errors of the basis predictor in the history prediction, and the first row and the second row are the prediction errors of the WAVE-SARIMA model and the OnS-GBDT model at history m times, respectively. Therefore, a training data set and a prediction input data set of a training classification model are constructed through PV matrixes at a plurality of historical moments, a multi-class regression model is generated and then used for predicting the probability that each base predictor at the next moment is the best predictor, and then the calculated probability is used as a weighting coefficient of the prediction result of the base predictor to obtain the comprehensive prediction result of the integrated model.
Specifically, the integration module 44 is configured to construct error samples of historical M moments based on a prediction result of offline trend prediction, a prediction result of online fluctuation prediction, and load real data; the input characteristic of the error sample is a sample matrix comprising 2 rows and M columns, wherein M is larger than 1; wherein the first behavior is based on a prediction error of an offline trend prediction, and the second behavior is based on a prediction error of an online fluctuation prediction; the label of the error sample is a base predictor type with better performance, namely the error of the off-line trend base predictor is smaller and is 0, and the error of the on-line fluctuation base predictor is smaller and is 1; according to the error sample labels at the historical M moments, calculating the historical performance stability of the base predictor; comparing the historical performance stability of the base predictor with a stability threshold to judge whether the historical prediction is stable; if so, based on the time locality (if a certain base predictor in the recent past can well predict the load condition in the near future, the base predictor is likely to still give a good prediction result in the near future), a weight distribution algorithm is adopted to endow each sample in the error samples with different weight coefficients so as to train a first multi-class regression model; predicting the weight coefficients of the offline trend base predictor and the online fluctuation base predictor according to the trained first multi-type regression model and the error input characteristic matrix at the current moment; integrating comprehensive prediction results based on the weight coefficients and prediction results of all the basis predictors; if not, training a second multi-class regression model directly according to the error samples of the historical M moments; and predicting the weight coefficients of the offline trend base predictor and the online fluctuation base predictor according to the second multi-type regression model and the error input characteristic matrix at the current moment, and further integrating the comprehensive prediction result.
It should be noted that the division of the modules of the above system is only a logical division, and the actual implementation may be wholly or partially integrated into one physical entity, or may be physically separated. And the modules can be realized in a form that all software is called by the processing element, or in a form that all the modules are realized in a form that all the modules are called by the processing element, or in a form that part of the modules are called by the hardware. For example: the x module can be a separately established processing element, and can also be integrated in a certain chip of the system. In addition, the x-module may be stored in the memory of the system in the form of program codes, and may be called by one of the processing elements of the system to execute the functions of the x-module. Other modules are implemented similarly. All or part of the modules can be integrated together or can be independently realized. The processing element described herein may be an integrated circuit having signal processing capabilities. In implementation, each step of the above method or each module above may be implemented by an integrated logic circuit of hardware in a processor element or an instruction in the form of software. These above modules may be one or more integrated circuits configured to implement the above methods, such as: one or more Application Specific Integrated Circuits (ASICs), one or more microprocessors (DSPs), one or more Field Programmable Gate Arrays (FPGAs), and the like. When a module is implemented in the form of a Processing element scheduler code, the Processing element may be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. These modules may be integrated together and implemented in the form of a System-on-a-chip (SOC).
The load prediction system of this embodiment provides a specific load prediction value at each prediction time based on the load prediction method, and performs model update according to a determined strategy. Meanwhile, the load prediction value given by the load prediction system is in butt joint with the service resource management system, so that a reliable and beneficial resource management decision basis is provided for service resource management operation, and active service resource management is realized.
EXAMPLE III
The present embodiment provides a load prediction apparatus including: a processor, memory, transceiver, communication interface, or/and system bus; the memory and the communication interface are connected with the processor and the transceiver through the system bus and are used for mutually communicating, the memory is used for storing the computer program, the communication interface is used for communicating with other devices, and the processor and the transceiver are used for running the computer program to enable the load prediction device to execute the steps of the load prediction method according to the embodiment I.
The above-mentioned system bus may be a Peripheral Component Interconnect (PCI) bus, an Extended Industry Standard Architecture (EISA) bus, or the like. The system bus may be divided into an address bus, a data bus, a control bus, and the like. For ease of illustration, only one thick line is shown, but this does not mean that there is only one bus or one type of bus. The communication interface is used for realizing communication between the database access device and other equipment (such as a client, a read-write library and a read-only library). The Memory may include a Random Access Memory (RAM), and may further include a non-volatile Memory (non-volatile Memory), such as at least one disk Memory.
The Processor may be a general-purpose Processor, and includes a Central Processing Unit (CPU), a Network Processor (NP), and the like; the Integrated Circuit may also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic device, or discrete hardware components.
The protection scope of the load prediction method according to the present invention is not limited to the execution sequence of the steps illustrated in this embodiment, and all the solutions implemented by the steps addition, subtraction, and step replacement according to the principles of the present invention are included in the protection scope of the present invention.
The present invention also provides a load prediction system, which can implement the load prediction method of the present invention, but the implementation device of the method for accessing a large amount of data of the present invention includes, but is not limited to, the structure of the system for accessing a large amount of data as illustrated in this embodiment, and all the structural modifications and substitutions of the prior art made according to the principles of the present invention are included in the protection scope of the present invention.
In summary, the load prediction method, the system, the device and the computer readable storage medium of the invention improve the model reliability of the cloud load height fluctuation scene, the prediction model realized based on the sliding window can adapt to different load modes, and the model combination strategy is integrated to realize the maximum performance of the model advantages; on the other hand, the load prediction tool which can be practically applied to the service resource management system is realized, and executable middleware support is practically provided. Therefore, the invention effectively overcomes various defects in the prior art and has high industrial utilization value.
The foregoing embodiments are merely illustrative of the principles and utilities of the present invention and are not intended to limit the invention. Any person skilled in the art can modify or change the above-mentioned embodiments without departing from the spirit and scope of the present invention. Accordingly, it is intended that all equivalent modifications or changes which can be made by those skilled in the art without departing from the spirit and technical spirit of the present invention be covered by the claims of the present invention.
Claims (10)
1. A method of load prediction, comprising:
acquiring historical tracking load data in a cloud application system, and constructing a historical load sequence;
performing offline trend prediction of the load on the historical load sequence to generate a prediction result based on the offline trend prediction;
online fluctuation prediction is carried out on the historical load sequence to generate a prediction result based on the online fluctuation prediction;
and combining the prediction result based on the offline trend prediction and the prediction result based on the online fluctuation prediction to integrate the comprehensive prediction result.
2. The load prediction method of claim 1, wherein the historical tracking load data comprises various loads of a cloud application system; the various loads of the cloud application system include request loads and/or resource loads.
3. The load prediction method of claim 1, wherein the step of performing offline trend prediction of the load on the historical load sequence to generate a prediction result based on the offline trend prediction comprises:
extracting low-frequency information of a historical load sequence;
and performing trend prediction on the next future cycle based on the historical trend information of the previous cycle to predict a prediction result based on the offline trend prediction.
4. The load prediction method of claim 3, wherein the step of performing online fluctuation prediction on the historical load sequence to generate a prediction result based on the online fluctuation prediction comprises:
at the initial moment, based on the minimum size of a preset sliding window, expanding the size of the window according to the trend correlation of adjacent windows, wherein the size of the window is expanded to be twice of the original size, repeatedly expanding until the adjacent windows are not related in trend any more, and stopping expanding;
judging whether the expanded current window size is larger than the boundary size of a preset sliding window; if yes, the recent load has obvious trend, and the minimum size of a preset sliding window is adopted as the size of the finally selected sliding window; if not, the recent load has no obvious trend, and the next step is carried out:
the window size is extended based on the temporal correlation of the historical load sequence.
5. The load prediction method of claim 4, wherein the step of extending the window size based on the temporal correlation of the historical load sequence comprises:
based on the maximum size of a preset sliding window, calculating the sequence time correlation of the recent historical load sequence in the window view to obtain a sequence autocorrelation function value;
gradually increasing the time lag value until the time lag size of the first time crossing the confidence interval of the sequence autocorrelation function value is found, and taking the time lag size of the first time crossing the confidence interval as the expanded current window size;
judging whether the expanded current window size is larger than the boundary size of a preset sliding window; if yes, indicating that the recent load has obvious time correlation, and taking the expanded current window size as the finally selected sliding window size; if not, indicating that the recent load does not have obvious time correlation, and directly adopting the maximum size of a preset sliding window as the size of the finally selected sliding window;
and constructing a training sample based on the finally selected sliding window size for training to predict the online load fluctuation of the next period.
6. The load prediction method of claim 4, wherein combining the prediction results based on offline trend prediction and the prediction results based on online fluctuation prediction, the step of integrating the integrated prediction results comprises:
constructing error samples of historical M moments based on a prediction result of off-line trend prediction, a prediction result of on-line fluctuation prediction and load real data; the input characteristic of the error sample is a sample matrix comprising 2 rows and M columns; wherein the first behavior is based on a prediction error of an offline trend prediction, and the second behavior is based on a prediction error of an online fluctuation prediction; the label of the error sample is a base predictor type with better performance, namely the error of the off-line trend base predictor is smaller and is 0, and the error of the on-line fluctuation base predictor is smaller and is 1;
according to error sample labels of historical M moments, calculating the historical performance stability of the base predictor;
comparing the historical performance stability of the base predictor with a stability threshold to judge whether the historical prediction is stable; if so, based on the time locality (if a certain base predictor in the recent past can well predict the load condition in the near future, the base predictor is likely to still give a good prediction result in the near future), a weight distribution algorithm is adopted to assign different weight coefficients to each sample in the error sample set so as to train a first-class regression model; predicting the weight coefficients of the offline trend base predictor and the online fluctuation base predictor according to the trained first multi-type regression model and the error input characteristic matrix at the current moment; integrating comprehensive prediction results based on the weight coefficients and the load prediction results of the base predictors; if not, training a second multi-class regression model directly according to the error samples of the historical M moments; and predicting the weight coefficients of the offline trend base predictor and the online fluctuation base predictor according to the second multi-type regression model and the error input characteristic matrix at the current moment, and further integrating the comprehensive prediction result.
7. The load prediction method according to claim 6, characterized in that the labels at the error samples for the historical M moments are label sequences consisting of 1 and/or 0; the step of calculating the historical performance stability of the basis predictor comprises the following steps:
calculating the variance corresponding to the label sequence and the reciprocal of the variance; and the reciprocal of the variance corresponding to the label sequence is the historical performance stability of the base predictor.
8. A load forecasting system, comprising:
the data acquisition module is used for acquiring historical tracking load data in the cloud application system and constructing a historical load sequence;
the off-line prediction module is used for carrying out off-line trend prediction on the load of the historical load sequence so as to generate a prediction result based on the off-line trend prediction;
the online prediction module is used for performing online fluctuation prediction on the historical load sequence to generate a prediction result based on the online fluctuation prediction;
and the integration module is used for combining the prediction result based on the off-line trend prediction and the prediction result based on the on-line fluctuation prediction to integrate the comprehensive prediction result.
9. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the load prediction method of any one of claims 1 to 7.
10. A load prediction device, comprising: a processor and a memory;
the memory is adapted to store a computer program and the processor is adapted to execute the computer program stored by the memory to cause the load prediction apparatus to perform the load prediction method of any of claims x to y.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111498189.8A CN114168419A (en) | 2021-12-09 | 2021-12-09 | Load prediction method, system, device and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111498189.8A CN114168419A (en) | 2021-12-09 | 2021-12-09 | Load prediction method, system, device and computer readable storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114168419A true CN114168419A (en) | 2022-03-11 |
Family
ID=80484873
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111498189.8A Pending CN114168419A (en) | 2021-12-09 | 2021-12-09 | Load prediction method, system, device and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114168419A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116048910A (en) * | 2022-12-08 | 2023-05-02 | 国网湖北省电力有限公司信息通信公司 | Double-scale prediction method for operation data of data center equipment |
| CN116089021A (en) * | 2023-04-10 | 2023-05-09 | 北京大学 | Deep learning-oriented large-scale load mixed part scheduling method, device and medium |
| CN116204387A (en) * | 2023-04-26 | 2023-06-02 | 之江实验室 | Chip current prediction method and device, medium and electronic equipment |
-
2021
- 2021-12-09 CN CN202111498189.8A patent/CN114168419A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116048910A (en) * | 2022-12-08 | 2023-05-02 | 国网湖北省电力有限公司信息通信公司 | Double-scale prediction method for operation data of data center equipment |
| CN116089021A (en) * | 2023-04-10 | 2023-05-09 | 北京大学 | Deep learning-oriented large-scale load mixed part scheduling method, device and medium |
| CN116204387A (en) * | 2023-04-26 | 2023-06-02 | 之江实验室 | Chip current prediction method and device, medium and electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114168419A (en) | Load prediction method, system, device and computer readable storage medium | |
| US11861474B2 (en) | Dynamic placement of computation sub-graphs | |
| CN109214421B (en) | Model training method and device and computer equipment | |
| US20160239592A1 (en) | Data-driven battery aging model using statistical analysis and artificial intelligence | |
| US11004012B2 (en) | Assessment of machine learning performance with limited test data | |
| CN111523640B (en) | Training method and device for neural network model | |
| CN108108743B (en) | Abnormal user identification method and device for identifying abnormal user | |
| CN111340221B (en) | Neural network structure sampling method and device | |
| US20210064432A1 (en) | Resource needs prediction in virtualized systems: generic proactive and self-adaptive solution | |
| Shariffdeen et al. | Adaptive workload prediction for proactive auto scaling in PaaS systems | |
| CN114298050A (en) | Model training method, entity relation extraction method, device, medium and equipment | |
| CN115222046A (en) | Neural network structure searching method and device, electronic equipment and storage medium | |
| US11715284B2 (en) | Anomaly detection apparatus, anomaly detection method, and program | |
| CN110348581B (en) | User feature optimizing method, device, medium and electronic equipment in user feature group | |
| CN116542673A (en) | Fraud identification method and system applied to machine learning | |
| AU2021251463B2 (en) | Generating performance predictions with uncertainty intervals | |
| CN114970883A (en) | Model quantization method and device, electronic equipment and storage medium | |
| Yan | Novel method to forecast software aging problems | |
| CN113850686A (en) | Insurance application probability determination method and device, storage medium and electronic equipment | |
| CN113329128A (en) | Traffic data prediction method and device, electronic equipment and storage medium | |
| Mohammad | An observation-based runtime configuration framework for the front-end network edge | |
| CN117435885A (en) | Model generation method, evaluation method, system, equipment and medium | |
| CN116861980A (en) | Granularity prediction model training method, resource occupation prediction method and device | |
| CN114237856A (en) | Operation type identification method and device, electronic equipment and storage medium | |
| CN118133168A (en) | Data processing method and device, electronic equipment and computer readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |