CN106357437B - A kind of Web Service QoS prediction technique based on multivariate time series - Google Patents
A kind of Web Service QoS prediction technique based on multivariate time series Download PDFInfo
- Publication number
- CN106357437B CN106357437B CN201610800115.8A CN201610800115A CN106357437B CN 106357437 B CN106357437 B CN 106357437B CN 201610800115 A CN201610800115 A CN 201610800115A CN 106357437 B CN106357437 B CN 106357437B
- Authority
- CN
- China
- Prior art keywords
- qos
- data
- value
- sample
- time series
- 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.)
- Active
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/147—Network analysis or design for predicting network behaviour
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L43/00—Arrangements for monitoring or testing data switching networks
- H04L43/50—Testing arrangements
- H04L43/55—Testing of service level quality, e.g. simulating service usage
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L67/00—Network arrangements or protocols for supporting network services or applications
- H04L67/01—Protocols
- H04L67/02—Protocols based on web technology, e.g. hypertext transfer protocol [HTTP]
Abstract
The Web Service QoS prediction technique based on multivariate time series that the invention discloses a kind of.This method mainly collects the QoS advertising data information of QoS attribute historical data and recent service provider publication, carries out the direct multi-step prediction for combining improved RBF neural network model to realize QoS attribute after data prediction.The multiple attributes for considering QoS, are given a forecast based on multivariate time series;There are complicated non-linear relations between QoS attribute, to describe this relationship, the history QoS attribute data being collected into are done phase space reconfiguration, approximation restores the non-linear relation system of multiple QoS attributes;The QoS data message reflection of service provider the issued in the recent period variation and development trend of the following QoS, QoS attribute data of this partial information after phase space reconfiguration combined, and constitutes QoS integrated data set;The dynamic multi-step prediction of QoS attribute is done in RBF neural network model after training.
Description
Technical field
The present invention relates to a kind of Web Service QoS prediction techniques, more particularly to the Web based on multivariate time series
Service QoS prediction technique, belongs to information technology field.
Background technique
With the fast development and application of Web service technology, occur many intimate Web services on network,
During selection meets the Web service of user demand, non-functional requirement is often ignored by people.In recent years, as NOT function
Service quality (QoS) beginning of energy sexual factor is gradually taken seriously, and more and more researchers start to be dedicated to Web service
QoS Predicting Technique.
The QoS attribute of Web service includes and performance-relevant response time (Response time), handling capacity
(Throughput), reliability (Reliability), availability (Availability), scalability
And price relevant to non-performance (Price), safety (Security), prestige (Credit) etc. (Expandability),
Attribute.Existing prediction model is divided into the Web service QoS prediction technique based on artificial intelligence, base mainly for single Web service
In the QoS prediction technique of similarity, QoS prediction technique based on the equation of structure and it is based on time series QoS prediction technique.Wherein
Prediction technique based on time series mainly considers the single attribute or two attributes of QoS, selects ARIMA model or neural network
Model is predicted, such as China considers two response time, availability attributes, is predicted using ARIMA model realization;Zia ur
Rehman etc. uses the response time attribute of ARIMA model prediction QoS;Zhang etc. uses RBF neural short-term forecast QoS
Attribute value;Liu etc. proposes the combination forecasting of a kind of ARIMA, gray model and neural network.But between the attribute of QoS
In the presence of complicated relation of interdependence, consider that the forecasting accuracy of a factor is poor merely, traditional ARIMA model is a kind of
Short-term Forecasting Model, the requirement to data is relatively high, needs to carry out multi step strategy to data.
Therefore the method for the present invention considers the correlation between QoS attribute, and multiple QoS attributes are formed data source.Performance phase
Qos value can all be influenced by closing attribute and non-performance association attributes, and the correlation between attribute can not use accurate formulation schema table
Show, therefore using the method for phase space reconfiguration, do not need to determine the relationship between each attribute, also without the concern for neighbouring time point
Between data correlation, QoS attribute historical data is mapped in a dynamical system, it is non-that approximation restores original multidimensional
Linear system.In addition with the promotion of technology, the overall development trend of Web service QoS has with the strategy change of service provider
The characteristic for having dynamic changeable directly does multistep using historical data and directly predicts to reflect current QoS attribute value, influences to predict
Precision.The trend of short-term QoS advertising representative current QoS attribute value, therefore by the QoS ad data used time of short-term service provider
Between sequence indicate, be added to prediction data concentrate carry out multi-step prediction.
RBF neural network model can further indicate that the complex relationship between input variable and predictive variable, with black box
Model comparision accurately describes this relationship, and the attribute value of prediction is relatively accurate.Traditional neural network weight training is one
The complex process to iterate, therefore this method utilizes Hesse matrices and the training of Levenberg-Marquardt optimization algorithm
Weight improves operation efficiency, reduces time overhead and space expense.For the changeable feature of QoS attribute value dynamic, in future
During prediction, collecting sample dynamic updates the parameter value of neural network, realizes dynamic prediction, improves precision of prediction.
Summary of the invention
Goal of the invention: aiming at the problems existing in the prior art with deficiency, in order to improve Web Service QoS prediction
Efficiency and accuracy consider the complex relationship between the changeable and multiple QoS attribute of dynamic of QoS attribute value, the present invention provides
A kind of Web Service QoS prediction technique based on multivariate time series.
Technical solution: a kind of Web Service QoS prediction technique based on multivariate time series includes the following steps:
Step 1: collecting the QoS attribute data of determining single Web service;
The pretreatment of step 2:QoS attribute data;
Step 3: establishing multivariate time series RBF (radial base) neural network prediction model;
Step 4: step-up error threshold value δ, Training RBF Neural Network determine the parameter value of neural network;
Step 5: utilizing RBF neural, realize QoS attribute value dynamic multi-step prediction.
The step 1, collecting data mainly includes two aspects, and (1) is collected from Service Proxy (service broker)
QoS attribute value is as historical data.Channel of the service broker as QoS attribute history data collection, as a kind of method call
The Client Agent service of property, makes application program send and receive message using channel, as needed creation, opening, calling
It services and closes when not needed.Application program can reuse service broker and be connected to identical service repeatedly, without extra
Time overhead and resource overhead.It is embedded in function code according to the different definition of each QoS attribute, is continuously obtained from service broker
Take the QoS attribute data of same Web service.(2) the QoS Information that the short-term service publisher of acquisition provides
Advertisement (QoS ad data).Skill is found by the Web service based on UDDI (Unify legislation discovery integration specification)
Art obtains the advertising information of specified Web service.The overall development trend of QoS is changed with the strategy of service provider has dynamic
Changeable characteristic, directly predicts cannot to be well reflected using historical data multistep current QoS attribute value, and short-term QoS advertisement
The trend of QoS attribute value in the following short period is described, ad data is utilized in prediction, to current QoS attribute value
Trend plays the role of very big, and can be improved precision of prediction.Therefore the QoS ad data of short-term service provider is used into the time
Sequence indicates, is added to prediction data concentration.
Process of data preprocessing mainly constructs QoS relation on attributes system, and generates the input data of prediction model, described
Step 2 is further are as follows:
Step 21: indicating QoS attribute historical data and the short-term advertisement data of service provider with multivariate time series.
The multivariate time series of QoS attribute historical data indicate are as follows:
Q={ Q1, Q2...,Qi,...,QN, i=1,2 ..., N
Qi={ qI, 1,qI, 2..., qi,j,...,qi,M, j=1,2 ..., M
The multivariate time series of QoS attribute short-term advertisement data indicate are as follows:
A={ A1, A2...,Ai,...AN, i=1,2 ..., N
Ai={ aI, 1,aI, 2’...,ai,j,...,ai,M, j=1,2 ..., M
Wherein, N is number of samples;M is QoS attribute number;QiIndicate i-th of QoS attribute historical data;qi,jIndicate i-th
J-th of QoS attribute value of a historical data;AiIndicate i-th of ad data;ai,jIndicate j-th of QoS of i-th of ad data
Attribute value;
Step 22: noise smoothing processing being carried out to QoS attribute historical data using Wavelet Transform Threshold Denoising Algorithm, reduction is made an uproar
Influence of the sound to time series forecasting.(1) assume that the data with white Gaussian noise are yi=Xi+αi(wherein i=1,2 ..., N,
αiFor white noise, XiFor the data after denoising);(2) calculate orthogonal wavelet transformation, select N number of sample QoS historical data as
Wavelet decomposition number of plies j is arranged, by y in discrete waveletiFormulaWith
(wherein, k=1,2 ..., N, yk=c0,k, n=1,2 ..., N, cj,kFor the jth layer scale coefficient of k-th of sample, dj,kIt is
The jth layer wavelet coefficient of k sample, hn-2kAnd gn-2kConstant type the unit vector h and g that the length of composition is 2N are a pair of orthogonal
Mirror filter group QMF, h-1G=0, j are constant type Decomposition order, and N is that discrete sampling is counted) wavelet decomposition to jth layer, obtains
To corresponding coefficient of wavelet decomposition;(3) formula is used(wherein, t indicates threshold value constant, Y
Indicate the wavelet coefficient of input,Wavelet coefficient after indicating the threshold process of output) wavelet coefficient after decomposition is done at threshold value
Reason;(4) wavelet inverse transformation is carried out, by the wavelet coefficient formula after threshold processReconstruct, obtains estimated value
Step 23:QoS attribute historical data and ad data do variable change of scale respectively, and range scale control is existed
[0,1].Because effect is best when the range of the input variable of neural network is between [0,1] or [- 1,1];
The multivariate time series of step 24:QoS attribute historical data carry out phase space reconfiguration, using average displacement method meter
Calculate Embedded dimensions m and delay time T.Multivariate time series after reconstruct indicate are as follows:
Q '={ Q1', Q2’...,Qi’,...,QN', i=1,2 ..., N
Wherein mjIndicate the dimension of j-th of attribute insertion;τjIndicate the delay time of j-th of attribute;Qi' indicate QiReconstruct
Time series afterwards.The thought of average displacement method is to give data Q by introducing average displacement (AD)iOne hypothesis Embedded dimensions
M finds out delay time;
Step 25: treated historical data and ad data are synthesized into QoS integrated data set.QoS integrated data set table
It is shown as: X={ X1, X2...,Xi,...,XN, i=1,2 ..., N, Xi=[Qi',Ai]T, i=1,2 ..., N, while by data
Collection is set as input XiT is exported with targetiPairs of sample.
RBF neural prediction is a kind of method of time series forecasting, by a series of layer by predictive variable and mesh
Mark variable connects, and constructs complicated non-linear relation with the mode of similar black box, the step 3 is further are as follows:
Step 31: establishing the RBF neural of three-decker, including input layer, hidden layer and output layer.Wherein first layer
Input layer is made of sension unit, by input vector XiIt is introduced into neural network;Second layer hidden layer maps vector from low-dimensional
To higher-dimension, the excitation function of hidden layer is Gaussian function;The last layer output layer is set as single output, by the output of hidden layer into
Row linear weighted combination arrives final output value, is denoted as Yi.The Gaussian function formula of hidden layer are as follows:
R=| | Xi-Ck||2, i=1,2 ..., N k=1,2 ..., L
Wherein, φ (r) is the output function of hidden layer;CkFor the center of RBF neural hidden layer;σ is extension constant;
L is the number of hidden layer central node.
The output layer formula of RBF neural are as follows:
Enable W=[w1,w2,...,wL]T
Wherein, YiTo export layer functions;wkFor the weight between k-th of hidden node and output node;CmaxIn choosing
Maximum distance between the heart;W is the weight matrix of hidden layer and output layer;
Step 32: determining the hidden layer central node C of RBF neuralk(k=1,2 ..., L), the number of hidden node
For the number of one group of integrated data set sample input, central node CkValue correspond to the value of each sample;
Step 33: the initial weight matrix W of random given RBF neural hidden layer and output layer;
In order to find suitable neural network parameter value, the step 4 is further are as follows:
Step 41: the training sample in data set sample being divided into several groups, every group of number of samples is db, step-up error
Threshold value δ;
Step 42: indicating hidden layer output function with matrix Γ, the matrix for exporting layer functions is expressed as Γ W=Y, Y=
[Y1,Y2,...,YN]T, enable
Step 43: one group of db sample data of input;
Step 44: the number of hidden layer node is the number of one group of integrated data set sample of input, i.e. L=db, center
Node CkValue correspond to the value of each sample;
Step 45: converting optimization problem for training weight W using Levenberg-Marquardt (LM) algorithm.Accidentally
Difference functionThe formula for calculating W is W=W+ Δ W, Δ W=(JTJ+μI)-1·JTS, wherein error
Matrix S=[s1,s2,...,si,...,sN]T, si=Yi-Ti, regularization coefficient μ is the constant greater than 0.The Hesse matrices H of f (W)
Are as follows:
The Jacobian matrix J of f (W) are as follows:
Simplify Δ W to calculate: (1)It enables
, enableIn W training process, every acquisition one
Sample calculates a εkAnd βk, enable N=db as the sample of one group of trained W;
Step 46: as f (W) < δ, going to step 43, otherwise input next group of sample data;
Step 47: when the value of f (W) reduces,When the value of f (W) increases, the μ of μ=2 is enabled, step 44 is gone to.
The characteristics of for Web Service QoS data dynamic change based on multivariate time series, the step 5,
During future anticipation, the parameter value of RBF neural needs dynamic to update, every to collect the corresponding ε of a sample calculatingkAnd βk,
When sample number reaches db, updates the value of W and μ is adjusted according to step 47.In sample data the QoS historical data of phase space reconfiguration and
The synthesis of short-term QoS advertising time sequence data, the prediction step that setting QoS is predicted is η (η > 1), with the data X of t momenttIn advance
Survey the QoS attribute value Y at t+ η momentt+η, realize multi-step prediction.
Detailed description of the invention
Fig. 1 is the method overall step figure of the embodiment of the present invention;
Fig. 2 is RBF neural network structure figure;
Fig. 3 is the flow chart of present invention method.
Specific embodiment
Combined with specific embodiments below, the present invention is furture elucidated, it should be understood that these embodiments are merely to illustrate the present invention
Rather than limit the scope of the invention, after the present invention has been read, those skilled in the art are to various equivalences of the invention
The modification of form falls within the application range as defined in the appended claims.
As shown in Figure 1, being the Web Service QoS prediction technique overall step of the invention based on multivariate time series
Figure mainly includes 5 steps:
Step 1: collecting the QoS attribute data of determining single Web service;
The pretreatment of step 2:QoS attribute data;
Step 3: establishing RBF (radial base) neural network prediction model based on multivariate time series;
Step 4: step-up error threshold value δ, Training RBF Neural Network determine the parameter value of neural network;
Step 5: utilizing RBF neural, realize QoS attribute value dynamic multi-step prediction;
As shown in Fig. 2, input vector is inputted from first layer, multiple minds for the RBF neural network structure figure that the present invention uses
It indicates to input multiple QoS attributes through first node, indicates input vector using multivariate time series;Hidden layer node and input layer section
It counts identical;Output node only one, indicate output QoS attribute number be 1;
As shown in figure 3, being the Web Service QoS prediction technique process proposed by the present invention based on multivariate time series
Figure, the specific steps are as follows:
Step 101: choosing the underlying attribute of QoS, including the main attribute of performance of QoS and non-performance attribute, Response to selection
The representative attribute such as time, handling capacity, success rate, cost;
Wherein collecting QoS attribute value as the specific steps of historical data from service broker includes:
The tool for generating service broker, service broker's work are provided in step 102:Web service aid packet (Toolkits)
Tool is a part in WSDL tool packet, describes file (WSDL) according to Web service and calls proxygen order, generates and collect
The service broker quotient of QoS attribute historical data;
Step 103: according to the definition of QoS attribute, artificial addition instruction or modification instruction complete additional function (when response
Between add code Dose times, handling capacity add code statistical unit time tranfer data bulk), clothes are reruned after compiling
Business process;
Step 104: the quantity of collecting sample as needed, the time that setting acquisition historical data circulation executes, text text
Shelves save the data that operation code is collected into;
The QoS ad data that short-term service publisher provides, specific steps are acquired using UDDI Relevant Service Discovery Technologies are as follows:
Step 105: searching the QoSInformation entity of Web service supplier;
Step 106:QoSInformation entity includes one or more QoS attribute information, each uses 5 tuples
<attributeName, attributeType, attributeValue, attributeUnit, constraints>description, really
Surely the QoS attribute needed, and return to the description information inquired;
Step 107: QoS advertising information is saved in text document.
QoS attribute data pre-processes specific processing step are as follows:
Step 108: indicating QoS attribute historical data and the short-term advertisement data of service provider with multivariate time series.
The multivariate time series of QoS attribute historical data indicate are as follows:
Q={ Q1,Q2...,Qi,...,QN, i=1,2 ..., N
Qi={ qI, 1,qI, 2..., qi,j,...,qi,M, j=1,2 ..., M
The multivariate time series of QoS attribute short-term advertisement data indicate are as follows:
A={ A1,A2...,Ai,...AN, i=1,2 ..., N
Ai={ aI, 1,aI, 2..., ai,j,...,ai,M, j=1,2 ..., M
Ai={ ai, 1, ai, 2 ..., ai, j ..., ai, M }, j=1,2 ..., M
Wherein, N is number of samples, and the time interval of adjacent sample is identical;M is QoS attribute number;QiIndicate i-th of QoS
Attribute historical data;qi,jIndicate j-th of QoS attribute value of i-th of historical data;AiIndicate i-th of ad data;ai,jIt indicates
J-th of QoS attribute value of i-th of ad data;
Step 109: noise smoothing processing being carried out to QoS attribute historical data using Wavelet Transform Threshold Denoising Algorithm: (1) false
If having the data y of white Gaussian noisei=Xi+αi(wherein i=1,2 ..., N, αiFor white noise, XiFor the data after denoising);
(2) orthogonal wavelet transformation is calculated, selects N number of sample QoS historical data as discrete wavelet, the wavelet decomposition number of plies j, y is setiFortune
Use formulaWith(wherein, k=1,2 ..., N, yk=c0,k, n=1,
2 ..., N, cj,kFor the jth layer scale coefficient of k-th of sample, dj,kFor the jth layer wavelet coefficient of k-th of sample, hn-2kAnd gn-2k
Constant type the unit vector h and g that the length of composition is 2N are a pair of orthogonal mirror filter group QMF, h-1G=0, j are constant
Type Decomposition order, N be discrete sampling points) wavelet decomposition to jth layer, obtain correspond to coefficient of wavelet decomposition;(3) formula is used(wherein, t indicates that threshold value constant, Y indicate the wavelet coefficient of input,Indicate output
Threshold process after wavelet coefficient) to each scale coefficient cj,kWavelet coefficient after decomposition does threshold process;(4) it carries out small
Wave inverse transformation, by the wavelet coefficient formula after threshold processReconstruct, obtains
To estimated value
Step 110:QoS attribute historical data and ad data do variable change of scale respectively, and range scale control is existed
[0,1].QoS attribute historical data change of scale formula isQoS short-term advertisement number
It is according to change of scale formulaWherein q.,jIndicate j-th of category in historical data sample
The all values of property, a.,jIndicate all values of j-th of attribute in ad data sample, (a.,j)maxIndicate a.,jMaximum value,
(a.,j)minIndicate a.,jMinimum value, (q.,j)maxIndicate q.,jMaximum value, (q.,j)minIndicate q.,jMinimum value;
Step 111:QoS attribute historical data phase space reconfiguration calculates Embedded dimensions m and delay using average displacement method
Time τ.Multivariate time series after reconstruct indicate are as follows:
Q '={ Q1', Q2’...,Qi’,...,QN', i=1,2 ..., N
Wherein mjIndicate the dimension of j-th of attribute insertion;τjIndicate the delay time of j-th of attribute;Qi' indicate QiReconstruct
Time series afterwards.
The thought of average displacement method is to give data Q by introducing average displacement (AD)iOne hypothesis Embedded dimensions m, finds out
Delay time.The average displacement formula of j-th of attribute of Q are as follows:With
τjThe increase of value, as < Sm(τj) > growth slope when being reduced to the 40% of initial value for the first time, corresponding point is required delay
Time;
Step 112: historical data that treated and ad data form QoS integrated data set.QoS integrated data set representations
Are as follows: X={ X1, X2...,Xi,...,XN, i=1,2 ..., N, Xi=[Qi',Ai]T, i=1,2 ..., N, while by data set
It is set as input XiT is exported with targetiPairs of sample.
The specific construction step of RBF neural network structure are as follows:
Step 113: establishing the RBF neural of three-decker, including input layer, hidden layer and output layer.Wherein first
Layer input layer is made of perception neuron, by input vector XiIntroduce neural network;Second layer hidden layer reflects vector from low-dimensional
It is mapped to higher-dimension, realizes higher-dimension curve matching, the excitation function of hidden layer is Gaussian function;The last layer output layer is set as single defeated
Out, the output of hidden layer is subjected to linear weighted combination as final output value, is denoted as Yi.The Gaussian function formula of hidden layer
Are as follows:
R=| | Xi-Ck||2, i=1,2 ..., N k=1,2 ..., L
Wherein, φ (r) is hidden layer output function;CkFor RBF neural hidden layer center;σ is extension constant;L is
Hidden layer central node number.
RBF neural output layer formula are as follows:
Enable W=[w1,w2,...,wL]T
Wherein, YiTo export layer functions;wkFor the weight between k-th of hidden node and output node;CmaxIn choosing
Maximum distance between the heart;W is the weight matrix of hidden layer and output layer;
Step 114: determining the hidden layer central node C of RBF neuralk(k=1,2 ..., L), hidden node number are
The number of one group of integrated data set sample of input, central node CkValue correspond to the value of each sample;
Step 115: the initial weight matrix W of random given RBF neural hidden layer and output layer.
The specific steps of training neural network and dynamic multi-step prediction process are as follows:
Step 116: the training sample in data set sample being divided into several groups, every group of number of samples is db, and setting misses
Poor threshold value δ;
Step 117: matrix Γ indicates that hidden layer output function, the matrix for exporting layer functions are expressed as Γ W=Y, Y=
[Y1,Y2,...,YN]T,
Step 118: one group of db sample data of input;
Step 119: hidden node number is the number of one group of input integrated data set sample, i.e. L=db, central node Ck
Value correspond to the value of each sample;
Step 120: converting optimization problem for training weight W using Levenberg-Marquardt (LM) algorithm.When
Error functionWhen being minimized, corresponding W is last solution, and the iterative formula for calculating W is W=W+
Δ W, Δ W=(JTJ+μI)-1·JTS, wherein error matrix S=[s1,s2,...,si,...,sN]T, si=Yi-Ti, regularization system
Number μ is the constant greater than 0.The Hesse matrices H of f (W) are as follows:
The Jacobian matrix J of f (W) are as follows:
Simplify Δ W to calculate: (1)It enables
, enable
W training process is not necessarily to the matrix Γ and Γ of one-time calculation complexityT, every one sample of acquisition is only needed to calculate a εk
And βk, Δ W is calculated when N=db, as one group of W more new samples;
Step 121: as f (W) < δ, saves prediction result and go to step 118, otherwise input next group of sample data,
Continue next step;
Step 122: when the value of f (W) reduces,When the value of f (W) increases, the μ of μ=2 is enabled;Go to step 119;
Step 123: during the following multi-step prediction, the value of W being updated according to step 120 and μ is adjusted according to step 122;If
The prediction step for setting QoS prediction is η (η > 1), with the data X of t momenttPredict the QoS attribute value Y at t+ η momentt+η。
Claims (5)
1. a kind of Web Service QoS prediction technique based on multivariate time series, which comprises the steps of:
Step 1: collecting the QoS attribute data of determining single Web service;
The pretreatment of step 2:QoS attribute data;Include:
Step 21: indicating QoS attribute historical data and the short-term advertisement data of service provider with multivariate time series;
Step 22: noise smoothing processing being carried out to QoS attribute historical data using Wavelet Transform Threshold Denoising Algorithm, reduces noise pair
The influence of time series forecasting;
Step 23:QoS attribute historical data and ad data do variable change of scale respectively, and range scale is controlled in [0,1];
Step 24:QoS attribute historical data phase space reconfiguration calculates Embedded dimensions m and delay time using average displacement method
τ;
Step 25: treated historical data and ad data are synthesized into QoS integrated data set;
Step 3: establishing multivariate time series RBF (radial base) neural network prediction model;
Step 4: setting neural metwork training error threshold δ, Training RBF Neural Network determine the parameter value of neural network;It will count
It is divided into several groups according to the training sample in collection sample, using LM algorithm optimization training weight in one group of sample data, works as error
When functional value is not less than error threshold δ, next group of sample data is inputted;
Step 5: utilizing RBF neural, realize QoS attribute value dynamic multi-step prediction.
2. the Web Service QoS prediction technique based on multivariate time series as described in claim 1, which is characterized in that
The step 1, collecting data mainly includes two aspects, and (1) collects QoS attribute value as historical data from service broker;Root
It is embedded in function code according to the different definition of each QoS attribute, the QoS attribute of same Web service is continuously acquired from service broker
Data;(2) the QoS ad data that the short-term service publisher of acquisition provides;By the QoS ad data used time of short-term service provider
Between sequence indicate, be added to prediction data concentration.
3. the Web Service QoS prediction technique based on multivariate time series as described in claim 1, which is characterized in that
The step 3 includes:
Step 31: establishing the RBF neural of three-decker, including input layer, hidden layer and output layer;Wherein first layer inputs
Layer is made of sension unit, by input vector XiIt is introduced into neural network;Vector is mapped to height from low-dimensional by second layer hidden layer
Dimension, the excitation function of hidden layer are Gaussian function;The last layer output layer is set as single output, and the output of hidden layer is carried out line
Property weighted array obtains final output value, is denoted as Yi;
Step 32: determining the hidden layer central node C of RBF neuralk, the number of hidden layer node is one group of integrated data set
The number of sample input, central node CkValue correspond to the value of each sample;
Step 33: the initial weight matrix W of random given RBF neural hidden layer and output layer.
4. the Web Service QoS prediction technique based on multivariate time series as described in claim 1, which is characterized in that
In order to find suitable neural network parameter value, the step 4 includes:
Step 41: the training sample in data set sample being divided into several groups, every group of number of samples is db, step-up error threshold value
δ;
Step 42: indicating hidden layer output function and output layer functions with matrix Γ;
Step 43: one group of db sample data of input;
Step 44: the number of hidden layer node is the number of one group of integrated data set sample of input, i.e. L=db, central node Ck
Value correspond to the value of each sample;
Step 45: converting optimization problem for training weight W using Levenberg-Marquardt (LM) algorithm, solve and miss
Vector when difference minimum, if error function
Step 46: as f (W) < δ, going to step 43, otherwise input next group of sample data;
Step 47: resetting the regularization coefficient of right value update for the value situation of change of f (W), then go to step 44.
5. the Web Service QoS prediction technique based on multivariate time series as claimed in claim 4, which is characterized in that
The step 5 when the sample number of collection reaches db, updates the value of weight W and is adjusted according to step 47 during future anticipation
Regularization coefficient μ;QoS historical data has restored the non-thread of multivariate time series by phase space reconfiguration well in sample data
Property system, short-term QoS advertisement auxiliary prediction model realizes the prediction of bigger step-length, the prediction step that setting QoS is predicted be η (η >
1), with the QoS attribute value at the data prediction t+ η moment of t moment.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610800115.8A CN106357437B (en) | 2016-08-31 | 2016-08-31 | A kind of Web Service QoS prediction technique based on multivariate time series |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610800115.8A CN106357437B (en) | 2016-08-31 | 2016-08-31 | A kind of Web Service QoS prediction technique based on multivariate time series |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106357437A CN106357437A (en) | 2017-01-25 |
CN106357437B true CN106357437B (en) | 2019-07-12 |
Family
ID=57858782
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610800115.8A Active CN106357437B (en) | 2016-08-31 | 2016-08-31 | A kind of Web Service QoS prediction technique based on multivariate time series |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106357437B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107038503A (en) * | 2017-04-18 | 2017-08-11 | 广东工业大学 | A kind of Demand Forecast method and system of shared equipment |
CN107612704B (en) * | 2017-07-28 | 2019-10-01 | 北京邮电大学 | The sort method and device of Internet of Things service |
CN113420421B (en) * | 2021-05-28 | 2023-07-25 | 西安邮电大学 | QoS prediction method based on time sequence regularized tensor decomposition in mobile edge calculation |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101834753A (en) * | 2010-03-30 | 2010-09-15 | 华中科技大学 | Network flow prediction method based on covariant orthogonality and neural network |
CN104270281A (en) * | 2014-09-03 | 2015-01-07 | 河海大学 | Web Service QoS (Quality of Service) combined forecasting method |
CN105046045A (en) * | 2015-06-05 | 2015-11-11 | 河海大学 | Bayesian combination-based Web service QoS prediction method |
CN105071961A (en) * | 2015-07-30 | 2015-11-18 | 中国人民解放军信息工程大学 | Method and device for predicting service quality of Web service |
-
2016
- 2016-08-31 CN CN201610800115.8A patent/CN106357437B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101834753A (en) * | 2010-03-30 | 2010-09-15 | 华中科技大学 | Network flow prediction method based on covariant orthogonality and neural network |
CN101834753B (en) * | 2010-03-30 | 2012-12-19 | 华中科技大学 | Network flow prediction method based on covariant orthogonality and neural network |
CN104270281A (en) * | 2014-09-03 | 2015-01-07 | 河海大学 | Web Service QoS (Quality of Service) combined forecasting method |
CN105046045A (en) * | 2015-06-05 | 2015-11-11 | 河海大学 | Bayesian combination-based Web service QoS prediction method |
CN105071961A (en) * | 2015-07-30 | 2015-11-18 | 中国人民解放军信息工程大学 | Method and device for predicting service quality of Web service |
Non-Patent Citations (4)
Title |
---|
一种基于Web Service的分级QoS的研究与实现;徐明伟;《计算机研究与发展》;20050430;第42卷(第4期);669-675 |
基于RBF神经网络的Web服务QoS短期预测;张金宏;《辽宁工程技术大学学报》;20101031;第29卷(第5期);52-56 |
基于径向基神经网络的Web Service QoS属性值组合预测方法;刘宗磊;《计算机与现代化》;20151231(第12期);918-921 |
基于时间序列分析的Web Service QoS预测方法;华哲邦;《计算机科学与探索》;20130331;第7卷(第3期);218-226 |
Also Published As
Publication number | Publication date |
---|---|
CN106357437A (en) | 2017-01-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Talagala et al. | Meta-learning how to forecast time series | |
CN109345302B (en) | Machine learning model training method and device, storage medium and computer equipment | |
Liu et al. | Short-term wind speed forecasting based on spectral clustering and optimised echo state networks | |
Zhu et al. | Streamflow estimation by support vector machine coupled with different methods of time series decomposition in the upper reaches of Yangtze River, China | |
CN112418482B (en) | Cloud computing energy consumption prediction method based on time series clustering | |
Gul et al. | Mid-term electricity load prediction using CNN and Bi-LSTM | |
CN106357437B (en) | A kind of Web Service QoS prediction technique based on multivariate time series | |
CN109711887A (en) | Generation method, device, electronic equipment and the computer media of store recommendation list | |
CN107704970A (en) | A kind of Demand-side load forecasting method based on Spark | |
CN110533112A (en) | Internet of vehicles big data cross-domain analysis and fusion method | |
Yin et al. | Research on stock trend prediction method based on optimized random forest | |
CN109389424B (en) | Flow distribution method and device, electronic equipment and storage medium | |
Wang et al. | Multivariate time series prediction based on optimized temporal convolutional networks with stacked auto-encoders | |
Li et al. | A novel deep learning approach for tourism volume forecasting with tourist search data | |
Wang et al. | Semi-empirical service composition: A clustering based approach | |
CN116542395A (en) | Low-carbon building monitoring system and method | |
CN111507505A (en) | Method for constructing reservoir daily input prediction model | |
CN101226521A (en) | Machine learning method for ambiguity data object estimation modeling | |
US11961115B2 (en) | Machine learning with data synthesization | |
Godahewa et al. | An accurate and fully-automated ensemble model for weekly time series forecasting | |
CN102103714A (en) | Real-time processing platform for predicting service data and predicting method | |
Liao et al. | Runoff Forecast Model Based on an EEMD-ANN and Meteorological Factors Using a Multicore Parallel Algorithm | |
Xu et al. | Federated learning for interpretable short-term residential load forecasting in edge computing network | |
CN116151449A (en) | Short-term prediction method and device for power load | |
Li et al. | A new dual weights optimization incremental learning algorithm for time series forecasting |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | 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 |