[go: up one dir, main page]

CN114064203A - Cloud virtual machine load prediction method based on multi-scale analysis and deep network model - Google Patents

Cloud virtual machine load prediction method based on multi-scale analysis and deep network model Download PDF

Info

Publication number
CN114064203A
CN114064203A CN202111265373.8A CN202111265373A CN114064203A CN 114064203 A CN114064203 A CN 114064203A CN 202111265373 A CN202111265373 A CN 202111265373A CN 114064203 A CN114064203 A CN 114064203A
Authority
CN
China
Prior art keywords
data
virtual machine
cloud virtual
time
output
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.)
Granted
Application number
CN202111265373.8A
Other languages
Chinese (zh)
Other versions
CN114064203B (en
Inventor
孟海宁
张嘉薇
李维
朱磊
杨哲
童新宇
黑新宏
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian University of Technology
Original Assignee
Xian University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xian University of Technology filed Critical Xian University of Technology
Priority to CN202111265373.8A priority Critical patent/CN114064203B/en
Publication of CN114064203A publication Critical patent/CN114064203A/en
Application granted granted Critical
Publication of CN114064203B publication Critical patent/CN114064203B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44—Arrangements for executing specific programs
    • G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
    • G06F9/45533—Hypervisors; Virtual machine monitors
    • G06F9/45558—Hypervisor-specific management and integration aspects
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
    • G06F17/148—Wavelet transforms
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5061—Partitioning or combining of resources
    • G06F9/5072—Grid computing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5061—Partitioning or combining of resources
    • G06F9/5077—Logical partitioning of resources; Management or configuration of virtualized resources
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44—Arrangements for executing specific programs
    • G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
    • G06F9/45533—Hypervisors; Virtual machine monitors
    • G06F9/45558—Hypervisor-specific management and integration aspects
    • G06F2009/4557—Distribution of virtual machine instances; Migration and load balancing
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44—Arrangements for executing specific programs
    • G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
    • G06F9/45533—Hypervisors; Virtual machine monitors
    • G06F9/45558—Hypervisor-specific management and integration aspects
    • G06F2009/45583—Memory management, e.g. access or allocation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44—Arrangements for executing specific programs
    • G06F9/455—Emulation; Interpretation; Software simulation, e.g. virtualisation or emulation of application or operating system execution engines
    • G06F9/45533—Hypervisors; Virtual machine monitors
    • G06F9/45558—Hypervisor-specific management and integration aspects
    • G06F2009/45595—Network integration; Enabling network access in virtual machine instances
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Computing Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明公开了基于多尺度分析和深度网络模型的云虚拟机负载预测方法，收集云虚拟机的资源和性能数据；获取云虚拟机资源和性能序列数据；对原始数据序列进行小波分解；利用阈值去噪；得到去噪后数据序列；对去噪后数据序列进行一阶差分，用一阶差分数据序列构造一个时间步矩阵；将时间步矩阵分为训练集和测试集；对训练集和测试集数据进行归一化处理；采取训练集数据对DLSTM预测模型进行训练，利用训练好的DLSTM预测模型对测试集数据进行预测，并对DLSTM预测模型进行性能评价；本发明解决了传统预测方法对长时间运行、数据量大的云虚拟机负载预测准确度不高的问题。

Figure 202111265373

The invention discloses a cloud virtual machine load prediction method based on multi-scale analysis and a deep network model, which collects the resource and performance data of the cloud virtual machine; obtains the cloud virtual machine resource and performance sequence data; performs wavelet decomposition on the original data sequence; Denoising; obtain the denoised data sequence; perform first-order difference on the denoised data sequence, and construct a time step matrix with the first-order difference data sequence; divide the time step matrix into training set and test set; The set data is normalized; the training set data is used to train the DLSTM prediction model, the trained DLSTM prediction model is used to predict the test set data, and the performance of the DLSTM prediction model is evaluated; the invention solves the problem of traditional prediction methods. The problem of low accuracy of load prediction of cloud virtual machines that run for a long time and have a large amount of data.

Figure 202111265373

Description

Cloud virtual machine load prediction method based on multi-scale analysis and deep network model
Technical Field
The invention belongs to the technical field of time series prediction, and particularly relates to a cloud virtual machine load prediction method based on multi-scale analysis and a deep network model.
Background
Cloud computing has been rapidly developed in recent years, and has attracted attention of professionals in various fields. Experts try to apply cloud computing to various fields and test the performance of cloud computing for processing various tasks. The wide application of the cloud computing technology not only brings new technology and high performance, but also brings high load to the system. Virtualization is one of important support technologies of cloud computing, and has high expansibility, high flexibility and high cost performance, so that people can obtain corresponding services according to own requirements, the expenditure can be saved, and the resource utilization rate can be improved. But as with other servers, the load conditions can greatly affect the performance of the virtual machine. In order to further optimize the application of the cloud virtual machine, it becomes important to explore the influence of the load condition of the cloud virtual machine on the virtual machine.
With the continuous development of cloud computing technology and cloud systems, more and more experts begin to research the load change rule of the cloud system and explore how to provide services for people by using a cloud virtual machine in a better load state.
The existing load prediction methods include LSTM and BP neural networks and the like. The LSTM can satisfy the demand of memorizing data information for a long period of time to achieve a prediction effect, but its performance in time series prediction is not very good. In the BP neural network method, the selection of parameters is difficult, and it is difficult to select the most suitable parameters to achieve the best prediction effect.
Disclosure of Invention
The invention provides a cloud virtual machine load prediction method based on multi-scale analysis and a deep network model, and solves the problem that the traditional prediction method is not accurate enough in prediction of cloud virtual machine load data which runs for a long time and has large data volume.
The technical scheme adopted by the invention is that the cloud virtual machine load prediction method based on the multi-scale analysis and the deep network model is implemented according to the following steps:
step 1, collecting data indexes of load conditions of a cloud virtual machine;
step 2, acquiring time sequence data of cloud virtual machine resources and performance parameters;
step 3, performing wavelet transformation on the sequence data obtained in the step 2, and denoising the original data according to a set threshold;
step 4, preprocessing the cloud virtual machine load data subjected to denoising in the step 3;
step 5, dividing the cloud virtual machine load data preprocessed in the step 4 into a training set and a testing set;
step 6, carrying out normalization processing on the training set and the test set of the cloud virtual machine load data in the step 5;
step 7, constructing a DLSTM prediction model of a cloud virtual machine load data time sequence;
step 8, training the prediction DLSTM model in the step 7 by using the normalized training set data in the step 6;
and 9, predicting the data of the test set after normalization in the step 6 by using the DLSTM prediction model after training in the step 7, and evaluating the performance of the DLSTM prediction model.
The invention is also characterized in that:
the time sequence data of the cloud virtual machine performance in the step 2 is CPU response time;
the specific contents of the wavelet transform denoising method in the step 3 are as follows: performing wavelet transformation by using noisy data, setting a threshold lambda, denoising original data, and reconstructing through inverse wavelet transformation to obtain denoised data;
wherein the wavelet in step 3 is db8 in the Daubechies (dbN) wavelet family; setting a threshold lambda, and then adopting a fixed threshold estimation method to remove dryness, wherein the fixed threshold is as follows:
Figure BDA0003326809930000031
wherein the pretreatment process in the step 4 specifically comprises the following steps: firstly, carrying out first-order difference on denoised data; recording the sequence after de-noising as X ═ X1，x2，...，xn) (n is the length of the entire time series), and the data series after the difference is Y ═ Y (Y is the length of the entire time series)1，y2，...，yn-1) (ii) a Using the post-value minus the pre-value in the sequence, i.e.:
yi＝xi+1-xi (2)
obtaining a first-order difference data sequence Y by using a formula (2), thereby eliminating the time dependence of the time sequence;
secondly, converting the first-order difference data sequence into a time step matrix, wherein each unit in the matrix comprises a data segment with the length of a time step for prediction; the time step used in the scheme is 2, and the construction process is as follows: converting the original sequence into a matrix P of n x 11(ii) a Inserting a 0 before the original sequence, and converting into a matrix P of n x 12(ii) a Will matrix P1And P2Merging into an n x 2 matrix P'; i.e. P1＝[y1，y2，...，yn]T，P2＝[0，y1，y2，...，yn-1]T，
P′＝[P2 P1] (3)；
Wherein the normalization process in the step 5 specifically comprises the following steps: use of
Figure BDA0003326809930000032
Denotes xiNormalized value, | x | YmaxIs the maximum value in the absolute value of the data after differenceNormalizing the data in the matrix P' to [ -1, 1 [ ]]An interval;
the structure of DLSTM in step 6 is specifically as follows: DLSTM is stacked from multiple LSTMs, each of which maintains a traditional structure; the input of the DLSTM model input layer at the moment of t is recorded as xtThe output of the output layer is ht(ii) a In order to prevent overfitting of the model, a dropout layer is arranged, so that the activation value of a neuron stops working at a certain probability p during forward propagation, and p is set to be 0.3 in the scheme;
connecting an activation layer behind the hidden layer so that the matrix operation result has nonlinearity; the activation function for the forgetting gate and the output gate in LSTM is a Sigmoid function, i.e.
Figure BDA0003326809930000041
The function outputs a value of 0 or 1, where outputting 0 indicates discarding the current information and outputting 1 indicates retaining the current information;
the input gate activation function being a tanh function, i.e.
Figure BDA0003326809930000042
For calculating candidate value vector information;
the input of the i-th layer LSTM of the DLSTM prediction model at the time t is xtAnd h is output at the time t-1t-1T-1 Module State Ct-1And i-1 layer LSTM hidden states
Figure BDA0003326809930000043
The components are combined together; output h at time ttAnd state CtTransmitting to the t +1 moment; i-th layer LSTM output at time t
Figure BDA0003326809930000044
Transferring to the next layer of LSTM for auxiliary prediction until the last layer of LSTM obtains an output value; hidden state of first layer LSTM at time t
Figure BDA0003326809930000045
Transmitting the LSTM of the second layer as input, and repeating the steps until the last LSTM outputs a result;
wherein the LSTM has a forgetting gate, an input gate and an output gate; the forgetting gate calculation method comprises the following steps:
Figure BDA0003326809930000046
the input gate calculation method comprises the following steps:
Figure BDA0003326809930000047
the candidate value vector calculation method comprises the following steps:
Figure BDA0003326809930000048
the state output is:
Figure BDA0003326809930000049
the output gate calculation method comprises the following steps:
Figure BDA00033268099300000410
the output is:
ht＝ot*tanh(Ct) (9)；
in the expressions (4) to (9), W is weight information, htIs the output at time t, xtFor the input at the time t, the input is,
Figure BDA0003326809930000051
hidden state of i-th layer LSTM at time t, and b is bias term
Wherein, the output data in the step 8 and the step 9 needs to be subjected to normalization removal and differencing removal to obtain a predicted value; the evaluation prediction model adopts a root mean square error RMSE and a minimized root mean square error RMSPE; the formulas are respectively as follows:
Figure BDA0003326809930000052
Figure BDA0003326809930000053
where N is the length of the data, yiTo predict value, xiThe cloud virtual machine load is raw data.
The invention has the beneficial effects that:
the cloud virtual machine load prediction method based on the multi-scale analysis and the deep network model solves the problem that the traditional prediction method is low in accuracy in cloud virtual machine load prediction for long-time operation and large data volume. A wavelet transformation method is provided for carrying out multi-scale decomposition on the data, and the data are decomposed into a high-frequency subsequence and a low-frequency subsequence; denoising the wavelet sequence of each scale by using a proper threshold value; through wavelet inverse transformation reconstruction, denoised data are obtained so as to obtain a better prediction effect; compared with the traditional LSTM method, the DLSTM method has the advantages that each layer of LSTM in the DLSTM runs on different time scales, and the result is transmitted to the next layer of LSTM, so that the DLSTM can effectively utilize each layer of LSTM, and more complex time sequence data can be learned. The prediction accuracy of DLSTM is therefore higher when a large amount of data is predicted.
Drawings
FIG. 1 is a DLSTM structure diagram of a cloud virtual machine load prediction method based on multi-scale analysis and a deep network model;
FIG. 2 is a general framework diagram of the cloud virtual machine load prediction method based on multi-scale analysis and a deep network model according to the present invention;
FIG. 3 is a wavelet denoising frame diagram of the cloud virtual machine load prediction method based on multi-scale analysis and a depth network model;
FIG. 4 is a time sequence diagram of raw data of the cloud virtual machine load prediction method based on multi-scale analysis and a deep network model according to the present invention;
FIG. 5 is a time sequence diagram of denoised data of the cloud virtual machine load prediction method based on multi-scale analysis and a deep network model;
FIG. 6 is a prediction result diagram of the cloud virtual machine load prediction method based on multi-scale analysis and a deep network model according to the present invention;
fig. 7 is a prediction comparison diagram of the cloud virtual machine load prediction method based on multi-scale analysis and a deep network model.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention provides a cloud virtual machine load prediction method based on multi-scale analysis and a Deep network model, the overall framework is shown in fig. 2, the method specifically comprises a method for predicting cloud virtual machine load by a Wavelet Transform (WT) and Deep Long Short Term Memory (DLSTM) neural network model, and the method is implemented by the following steps:
step 1, collecting data indexes of load conditions of a cloud virtual machine;
step 2, acquiring time series data of cloud virtual machine resources and performance parameters, as shown in fig. 4;
step 3, as shown in fig. 3, performing wavelet denoising on the sequence data obtained in step 2; performing wavelet transformation by using the noisy data; setting a threshold lambda, denoising the cloud virtual machine load original data, and reconstructing through inverse wavelet transform to obtain denoised data, as shown in fig. 5. Wherein the fixed threshold is:
Figure BDA0003326809930000061
step 4, preprocessing the sequence data after denoising in the step 3; firstly, carrying out first-order difference on data, and specifically comprising the following steps: recording the sequence after de-noising as X ═ X1，x2，...，xn) (n is the length of the time series), and the differentiated data series is as follows: y ═ Y1，y2，...，yn-1) The later value in the sequence is subtracted from the former value to obtain a first-order difference data sequence Y, so that the time dependence of the time sequence is eliminatedSex; namely:
yi＝xi+1-xi (2)
secondly, converting the first-order difference data sequence into a time step matrix, wherein each unit in the matrix comprises a data segment with the length of a time step for prediction; the time step used in the invention is 2, and the conversion process is as follows: converting the original sequence into a matrix P of n x 11(ii) a Inserting a 0 before the original sequence, and converting into a matrix P of n x 12(ii) a Will matrix P1And P2Merging into an n x 2 matrix P'; i.e. P1＝[y1，y2，...，yn]T，P2＝[0，y1，y2，...，yn-1]TThe merging method comprises the following steps:
P′＝[P2 P1] (3)；
step 5, dividing the cloud virtual machine load data preprocessed in the step 4 into a training set and a testing set;
step 6, carrying out normalization processing on the training set and the test set of the cloud virtual machine load data in the step 5; use of
Figure BDA0003326809930000071
(
Figure BDA0003326809930000072
Denotes xiNormalized value, | x | YmaxMaximum of absolute values of the data after differentiation) to [ -1, 1]An interval;
step 7, constructing a DLSTM prediction model of a cloud virtual machine load data time sequence; as shown in fig. 1, DLSTM is made up of a stack of LSTMs, each of which maintains a conventional structure. The input of the DLSTM model input layer at the moment of t is recorded as xtThe output of the output layer is ht(ii) a In order to prevent overfitting of the model, a dropout layer is arranged, so that the activation value of a neuron stops working at a certain probability p during forward propagation, and p is set to be 0.3 in the method; the DLSTM hidden layer is used for carrying out multi-level abstraction on input features; for each node's input, the hidden layer has a different connection weightThe neurons of the output layer adjust the weights of the neurons of the hidden layer, so that the output result tends to real data;
the active layer is connected after the hidden layer, so that the matrix operation result has nonlinearity, and the active function used by the forgetting gate and the output gate in the LSTM is a Sigmoid function, namely
Figure BDA0003326809930000081
The function outputs a value of 0 or 1, where output 0 indicates discarding the current information and output 1 indicates retaining the current information. The input gate activation function being a tanh function, i.e.
Figure BDA0003326809930000082
For calculating candidate value vector information;
the input of the i-th layer LSTM of the DLSTM prediction model at the time t is xtAnd h is output at the time t-1t-1T-1 Module State Ct-1And i-1 layer LSTM hidden states
Figure BDA0003326809930000083
The components are combined together; output h at time ttAnd state CtTransmitting to the t +1 moment; i-th layer LSTM output at time t
Figure BDA0003326809930000084
Pass to the next layer of LSTM to assist prediction until the last layer of LSTM takes an output value. Hidden state of first layer LSTM at time t
Figure BDA0003326809930000085
Transmitting the LSTM of the second layer as input, and repeating the steps until the last LSTM outputs a result;
each layer of LSTM has a forgetting gate, an input gate and an output gate;
the forgetting gate calculation method comprises the following steps:
Figure BDA0003326809930000086
the input gate calculation method comprises the following steps:
Figure BDA0003326809930000087
the candidate value vector calculation method comprises the following steps:
Figure BDA0003326809930000088
the state output is:
Figure BDA0003326809930000089
the output gate calculation method comprises the following steps:
Figure BDA00033268099300000810
the output is:
ht＝ot*tanh(Ct) (9)；
where W is weight information, htIs the output at time t, xtFor the input at the time t, the input is,
Figure BDA00033268099300000811
hidden state of i-th layer LSTM at time t, and b is bias term
Step 8, training the prediction DLSTM model in the step 7 by using the normalized training set data in the step 6;
step 9, predicting the data of the test set after normalization in the step 6 by using the DLSTM prediction model after training in the step 7, and evaluating the performance of the DLSTM prediction model, as shown in FIG. 6;
examples
The embodiment adopts the load condition of the cloud virtual machine as an example; comparing the prediction result with the LSTM based on the multi-scale analysis and the DLSTM model, and the result is shown in FIG. 7; taking a mean square root error (RMSE) and a minimized Root Mean Square Prediction Error (RMSPE) as evaluation indexes, wherein the formulas are respectively shown as (10) to (11);
Figure BDA0003326809930000091
Figure BDA0003326809930000092
where N is the length of the data, yiTo predict value, xiThe method comprises the steps of obtaining original data of cloud virtual machine load;
the method comprises the following specific steps:
step 1, collecting data indexes of load conditions of a cloud virtual machine;
step 2, acquiring time sequence data of cloud virtual machine resources and performance parameters;
step 3.1, performing wavelet transformation on the sequence data obtained in the step 2;
step 3.2, setting a threshold value, and denoising the cloud virtual machine load sequence data;
step 3.3, obtaining de-noised data through wavelet inverse transformation reconstruction, wherein the fixed threshold value is
Figure BDA0003326809930000093
Step 4.1, performing first-order difference on the denoised data in the step 3, and specifically comprising the following steps: recording the sequence after de-noising as X ═ X1，x2，...，xn) (n is the length of the entire time series), and the data series after the difference is Y ═ Y (Y is the length of the entire time series)1，y2，...，yn-1) Using the value after the sequence minus the value before, i.e. yi＝xi+1-xiObtaining a first-order difference data sequence Y, thereby eliminating the time dependence of the time sequence;
step 4.2, converting the first-order difference data sequence into a time step matrix, wherein each unit in the matrix comprises a data segment with the length of a time step for prediction; the time step used in the invention is 2, and the construction process is as follows: converting the original sequence into a matrix of n x 1P1(ii) a Inserting a 0 before the original sequence, and converting into a matrix P of n x 12I.e. P1＝[y1，y2，...，yn]T，P2＝[0，y1，y2，...，yn-1]T；
Step 4.3, the matrix P1And P2Are combined into an n x 2 matrix P ', i.e. P' ═ P2 P1]；
Step 5, dividing the cloud virtual machine load data preprocessed in the step 4 into a training set and a testing set;
step 6, carrying out normalization processing on the training set and the test set of the cloud virtual machine load data in the step 5; the normalization processing process specifically comprises the following steps: use of
Figure BDA0003326809930000101
(
Figure BDA0003326809930000102
Denotes xiNormalized value, | x | YmaxFor the maximum of the absolute values of the differenced data) to normalize the data in matrix P' to [ -1, 1]An interval;
step 7.1, constructing a DLSTM prediction model of a cloud virtual machine load data time sequence; DLSTM is stacked from multiple LSTMs, each of which maintains a conventional structure. The input of the DLSTM model input layer at the moment of t is recorded as xtThe output of the output layer is ht(ii) a In order to prevent overfitting of the model, a dropout layer is arranged, so that the activation value of a neuron stops working at a certain probability p during forward propagation, and in the method, p is set to be 0.3, so that the generalization of the model is stronger and the model does not depend on certain local characteristics; the DLSTM hidden layer is used for carrying out multi-level abstraction on input features; for the input of each node, the hidden layer has different connection weights, and the neurons of the output layer adjust the weights of the neurons of the hidden layer, so that the output result tends to real data.
The active layer is connected after the hidden layer so that the matrix operation result has non-linearity. The active function used by the forgetting gate and the output gate in the LSTM is a Sigmoid function, and the function outputs a value of 0 or 1, wherein the output of 0 indicates discarding the current information, and the output of 1 indicates keeping the current information. The input gate activation function is a tanh function and is used for calculating candidate value vector information;
step 8, training the prediction DLSTM model in the step 7 by using the normalized training set data in the step 6;
and 9, predicting the data of the test set after normalization in the step 6 by using the DLSTM prediction model after training in the step 7, and evaluating the performance of the DLSTM prediction model.

Claims (9)

1. The cloud virtual machine load prediction method based on the multi-scale analysis and the deep network model is characterized by comprising the following steps:
step 1, collecting data indexes of load conditions of a cloud virtual machine;
step 2, acquiring time sequence data of cloud virtual machine resources and performance parameters;
step 3, performing wavelet transformation on the sequence data obtained in the step 2, and denoising the original data according to a set threshold;
step 4, preprocessing the cloud virtual machine load data subjected to denoising in the step 3;
step 5, dividing the cloud virtual machine load data preprocessed in the step 4 into a training set and a testing set;
step 6, carrying out normalization processing on the training set and the test set of the cloud virtual machine load data in the step 5;
step 7, constructing a DLSTM prediction model of a cloud virtual machine load data time sequence;
step 8, training the prediction DLSTM model in the step 7 by using the normalized training set data in the step 6;
and 9, predicting the data of the test set after normalization in the step 6 by using the DLSTM prediction model after training in the step 7, and evaluating the performance of the DLSTM prediction model.
2. The method for predicting the load of the cloud virtual machine based on the multi-scale analysis and the deep network model according to claim 1, wherein the time series data of the cloud virtual machine performance in the step 2 is CPU response time.
3. The cloud virtual machine load prediction method based on the multi-scale analysis and the deep network model as claimed in claim 1, wherein the specific content of the wavelet transformation denoising method in the step 3 is as follows: and performing wavelet transformation by using the noisy data, setting a threshold lambda, denoising the original data, and reconstructing through inverse wavelet transformation to obtain denoised data.
4. The method for predicting the load of the cloud virtual machine based on the multi-scale analysis and depth network model according to claim 3, wherein the wavelet in the step 3 is db8 in the Daubechies (dbN) wavelet family; setting a threshold lambda, and then adopting a fixed threshold estimation method to remove dryness, wherein the fixed threshold is as follows:
Figure FDA0003326809920000021
5. the cloud virtual machine load prediction method based on the multi-scale analysis and the deep network model according to claim 1, wherein the preprocessing process in the step 4 specifically comprises: firstly, carrying out first-order difference on denoised data; recording the sequence after de-noising as X ═ X1，x2，...，xn) (n is the length of the entire time series), and the data series after the difference is Y ═ Y (Y is the length of the entire time series)1，y2，...，yn-1) (ii) a Using the post-value minus the pre-value in the sequence, i.e.:
yi＝xi+1-xi (2)
obtaining a first-order difference data sequence Y by using a formula (2), thereby eliminating the time dependence of the time sequence;
secondly, converting the first-order difference data sequence into a time step matrix, wherein each unit in the matrix comprises a data segment with the length of a time step for prediction; the time step used in the scheme is 2, and the scheme is constructedThe process is as follows: converting the original sequence into a matrix P of n x 11(ii) a Inserting a 0 before the original sequence, and converting into a matrix P of n x 12(ii) a Will matrix P1And P2Merging into an n x 2 matrix P'; i.e. P1＝[y1，y2，...，yn]T，P2＝[0，y1，y2，...，yn-1]T，
P′＝[P2P1] (3)。
6. The cloud virtual machine load prediction method based on the multi-scale analysis and the deep network model according to claim 1 or 5, wherein the normalization processing in the step 5 specifically comprises: use of
Figure FDA0003326809920000022
Figure FDA0003326809920000023
Denotes xiNormalized value, | x | YmaxNormalizing the data in the matrix P' to [ -1, 1 ] for the maximum of the absolute values of the differentiated data]An interval.
7. The cloud virtual machine load prediction method based on the multi-scale analysis and the deep network model according to claim 1, wherein the construction of the DLSTM in the step 6 is specifically as follows: DLSTM is stacked from multiple LSTMs, each of which maintains a traditional structure; the input of the DLSTM model input layer at the moment of t is recorded as xtThe output of the output layer is ht(ii) a In order to prevent overfitting of the model, a dropout layer is arranged, so that the activation value of a neuron stops working at a certain probability p during forward propagation, and p is set to be 0.3 in the scheme;
connecting an activation layer behind the hidden layer so that the matrix operation result has nonlinearity; the activation function for the forgetting gate and the output gate in LSTM is a Sigmoid function, i.e.
Figure FDA0003326809920000031
The function outputs a value of 0 or 1, where outputting 0 indicates discarding the current information and outputting 1 indicates retaining the current information;
the input gate activation function being a tanh function, i.e.
Figure FDA0003326809920000032
For calculating candidate value vector information;
the input of the i-th layer LSTM of the DLSTM prediction model at the time t is xtAnd h is output at the time t-1t-1T-1 Module State Ct-1And i-1 layer LSTM hidden states
Figure FDA0003326809920000033
The components are combined together; output h at time ttAnd state CtTransmitting to the t +1 moment; i-th layer LSTM output at time t
Figure FDA0003326809920000034
Transferring to the next layer of LSTM for auxiliary prediction until the last layer of LSTM obtains an output value; hidden state of first layer LSTM at time t
Figure FDA0003326809920000035
And transmitting the LSTM of the second layer as an input, and the like until the last LSTM outputs a result.
8. The cloud virtual machine load prediction method based on multi-scale analysis and deep network model of claim 7, wherein the LSTM has a forgetting gate, an input gate and an output gate; the forgetting gate calculation method comprises the following steps:
Figure FDA0003326809920000036
the input gate calculation method comprises the following steps:
Figure FDA0003326809920000041
the candidate value vector calculation method comprises the following steps:
Figure FDA0003326809920000042
the state output is:
Figure FDA0003326809920000043
the output gate calculation method comprises the following steps:
Figure FDA0003326809920000044
the output is:
ht＝ot*tanh(Ct) (9)。
in the expressions (4) to (9), W is weight information, htIs the output at time t, xtFor the input at the time t, the input is,
Figure FDA0003326809920000045
for the hidden state of the ith layer LSTM at time t, b is the bias term.
9. The method for predicting the load of the cloud virtual machine based on the multi-scale analysis and the deep network model according to claim 1, wherein the output data in the steps 8 and 9 need to be subjected to denormalization and dedifferentiation to obtain a predicted value; the evaluation prediction model adopts a root mean square error RMSE and a minimized root mean square error RMSPE; the formulas are respectively as follows:
Figure FDA0003326809920000046
Figure FDA0003326809920000047
where N is the length of the data, yiTo predict value, xiThe cloud virtual machine load is raw data.
CN202111265373.8A 2021-10-28 2021-10-28 Cloud virtual machine load prediction method based on multi-scale analysis and depth network model Active CN114064203B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111265373.8A CN114064203B (en) 2021-10-28 2021-10-28 Cloud virtual machine load prediction method based on multi-scale analysis and depth network model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111265373.8A CN114064203B (en) 2021-10-28 2021-10-28 Cloud virtual machine load prediction method based on multi-scale analysis and depth network model

Publications (2)

Publication Number Publication Date
CN114064203A true CN114064203A (en) 2022-02-18
CN114064203B CN114064203B (en) 2024-07-23

Family

ID=80235901

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111265373.8A Active CN114064203B (en) 2021-10-28 2021-10-28 Cloud virtual machine load prediction method based on multi-scale analysis and depth network model

Country Status (1)

Country Link
CN (1) CN114064203B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114897045A (en) * 2022-03-30 2022-08-12 重庆大学 Material removal state prediction method based on multi-scale attention mechanism
CN119718861A (en) * 2024-12-06 2025-03-28 合肥工业大学 Cloud system load prediction method based on multi-model optimization
CN120162138A (en) * 2025-01-16 2025-06-17 浙江工商大学 Cloud platform resource load prediction method based on adaptive multi-scale patch network
CN120415917A (en) * 2025-07-03 2025-08-01 深圳宸元网信科技有限公司 A communication dynamic analysis method for virtual network security

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109542585A (en) * 2018-11-14 2019-03-29 山东大学 A kind of Virtual Machine Worker load predicting method for supporting irregular time interval
CN111209083A (en) * 2020-01-08 2020-05-29 中国联合网络通信集团有限公司 Method, device and storage medium for container scheduling
US20200193268A1 (en) * 2018-12-14 2020-06-18 Advanced Micro Devices, Inc. Multi-instance recurrent neural network prediction
US20200380351A1 (en) * 2019-05-28 2020-12-03 Sap Se Automated Scaling Of Resources Based On Long Short-Term Memory Recurrent Neural Networks And Attention Mechanisms
CN112488396A (en) * 2020-12-01 2021-03-12 国网福建省电力有限公司 Wavelet transform-based electric power load prediction method of Holt-Winters and LSTM combined model
CN113033780A (en) * 2021-03-24 2021-06-25 西北大学 Cloud platform resource prediction method based on double-layer attention mechanism
CN113220450A (en) * 2021-04-29 2021-08-06 南京邮电大学 Load prediction method, resource scheduling method and device for cloud-side multi-data center
CN113222145A (en) * 2021-06-04 2021-08-06 西安邮电大学 MODWT-EMD-based time sequence hybrid prediction method
CN114510872A (en) * 2022-01-10 2022-05-17 西安理工大学 Cloud server aging prediction method based on self-attention mechanism DLSTM

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109542585A (en) * 2018-11-14 2019-03-29 山东大学 A kind of Virtual Machine Worker load predicting method for supporting irregular time interval
US20200193268A1 (en) * 2018-12-14 2020-06-18 Advanced Micro Devices, Inc. Multi-instance recurrent neural network prediction
US20200380351A1 (en) * 2019-05-28 2020-12-03 Sap Se Automated Scaling Of Resources Based On Long Short-Term Memory Recurrent Neural Networks And Attention Mechanisms
CN111209083A (en) * 2020-01-08 2020-05-29 中国联合网络通信集团有限公司 Method, device and storage medium for container scheduling
CN112488396A (en) * 2020-12-01 2021-03-12 国网福建省电力有限公司 Wavelet transform-based electric power load prediction method of Holt-Winters and LSTM combined model
CN113033780A (en) * 2021-03-24 2021-06-25 西北大学 Cloud platform resource prediction method based on double-layer attention mechanism
CN113220450A (en) * 2021-04-29 2021-08-06 南京邮电大学 Load prediction method, resource scheduling method and device for cloud-side multi-data center
CN113222145A (en) * 2021-06-04 2021-08-06 西安邮电大学 MODWT-EMD-based time sequence hybrid prediction method
CN114510872A (en) * 2022-01-10 2022-05-17 西安理工大学 Cloud server aging prediction method based on self-attention mechanism DLSTM

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
HAINING MENG: ""Track Irregularity Prediction Based on DWT-DLSTM Model"", 《2022 INTERNATIONAL CONFERENCE ON NETWORKING AND NETWORK APPLICATIONS (NANA)》, 30 December 2022 (2022-12-30), pages 1 - 5, XP034260738, DOI: 10.1109/NaNA56854.2022.00095 *
MINGYANG SUN: ""Using Bayesian Deep Learning to Capture Uncertainty for Residential Net Load Forecasting"", 《IEEE TRANSACTIONS ON POWER SYSTEMS》, vol. 35, no. 1, 21 June 2019 (2019-06-21), pages 188 - 201, XP011765898, DOI: 10.1109/TPWRS.2019.2924294 *
吉莹: ""基于神经网络组合模型的多因素电力负荷预测研究"", 《中国优秀硕士学位论文全文数据库 工程科技II辑》, no. 2020, 15 September 2020 (2020-09-15), pages 042 - 70 *

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114897045A (en) * 2022-03-30 2022-08-12 重庆大学 Material removal state prediction method based on multi-scale attention mechanism
CN119718861A (en) * 2024-12-06 2025-03-28 合肥工业大学 Cloud system load prediction method based on multi-model optimization
CN120162138A (en) * 2025-01-16 2025-06-17 浙江工商大学 Cloud platform resource load prediction method based on adaptive multi-scale patch network
CN120415917A (en) * 2025-07-03 2025-08-01 深圳宸元网信科技有限公司 A communication dynamic analysis method for virtual network security
CN120415917B (en) * 2025-07-03 2025-09-19 深圳宸元网信科技有限公司 Communication dynamic analysis method for virtual network security

Also Published As

Publication number Publication date
CN114064203B (en) 2024-07-23

Similar Documents

Publication Publication Date Title
CN114064203A (en) Cloud virtual machine load prediction method based on multi-scale analysis and deep network model
CN108197648B (en) Hydroelectric generating set fault diagnosis method and system based on LSTM deep learning model
Abderrahmane et al. Design space exploration of hardware spiking neurons for embedded artificial intelligence
CN117389824A (en) A cloud server load prediction method based on signal decomposition and hybrid model
WO2020176295A1 (en) Artificial neural network compression via iterative hybrid reinforcement learning approach
CN115168443B (en) Anomaly detection method and system based on GCN-LSTM and attention mechanism
CN109583565A (en) Forecasting Flood method based on the long memory network in short-term of attention model
CN115204035B (en) Method, device and storage medium for predicting operation parameters of generator set based on multi-scale time sequence data fusion model
CN110443419A (en) Medium-term and long-term Runoff Forecast method based on ICEEMDAN and extreme learning machine
CN115759461A (en) Internet of things-oriented multivariate time sequence prediction method and system
CN107403188A (en) A kind of quality evaluation method and device
CN116976512A (en) UAV cluster trajectory prediction method based on graph neural network
CN114490065A (en) Load prediction method, device and equipment
CN109741733B (en) Speech Phoneme Recognition Method Based on Consistent Routing Network
CN116665483A (en) A Novel Method for Predicting Remaining Parking Spaces
CN116052254A (en) Visual Continuous Emotion Recognition Method Based on Extended Kalman Filter Neural Network
CN111064617A (en) Network flow prediction method and device based on empirical mode decomposition clustering
CN118917412A (en) Perception model self-recommendation method and device based on super-parameter intelligent optimization
CN118364238A (en) Ship motion attitude prediction method based on CNN-GRU-Attention
Pandhiani et al. Time series forecasting by using hybrid models for monthly streamflow data
CN116362408A (en) Dissolved Oxygen Prediction Method and System Based on Data Fluctuation Decomposition and Time Series Mining
Xing et al. Extreme-Long-short term memory for Time-series Prediction
CN116566840A (en) A Container Cloud Resource Prediction Method Based on Empirical Wavelet Transform and Combined Model
CN110909453A (en) EEMD-based power transmission line icing grade prediction method
CN116205135A (en) SO based on data decomposition and neural network 2 Emission prediction method and system

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
GR01 Patent grant
GR01 Patent grant