CN109905481A - A kind of building of the Qos model based on RTI-DDS and the runnability prediction technique of Qos strategy protocol - Google Patents
A kind of building of the Qos model based on RTI-DDS and the runnability prediction technique of Qos strategy protocol Download PDFInfo
- Publication number
- CN109905481A CN109905481A CN201910175135.4A CN201910175135A CN109905481A CN 109905481 A CN109905481 A CN 109905481A CN 201910175135 A CN201910175135 A CN 201910175135A CN 109905481 A CN109905481 A CN 109905481A
- Authority
- CN
- China
- Prior art keywords
- strategy
- qos
- sample
- matrix
- parameter
- 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
Links
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The building of the Qos model based on RTI-DDS that the invention discloses a kind of and the runnability prediction technique of Qos strategy protocol.The present invention according to RTI-DDS design in entity classification in each node, include each section corresponding Qos strategy as the node of the first dimension of decision tree.In node, contains second dimension, the offering question of Qos strategy is divided into numerical indication and switchgear distribution two parts, is converted to a node using time complexity and space complexity as the linear model of index as second dimension respectively;And the nonlinear model equally based on multi-variable decision tree construction, using this as second node of the second dimension.User can provide the feature and index of transmission data, be derived by suitable model parameter and Qos strategy from decision tree nodes, the complexity of this model modeling is low, and the calculating time is short, and cost is small.
Description
Technical field
The invention belongs to software service technical fields, and in particular to a kind of Qos processing based on RTI-DDS.
Background technique
RTI-DDS is the commercial network middleware DDS of RTI (Real-Time Innovations) company publication
(DataDistribution Service, Data distributing).
Ethernet is the present most common network, it is in the bottom of fieldbus, and active adoption ethernet technology is
The development trend of fieldbus at present;Meanwhile ICP/IP protocol has become the international net being used in conjunction with by the application of internet
Network language.At present in the distributed network system (DNS) constituted using Ethernet, TCP/IP network technology, the real-time of transmission with
The development of Fast Ethernet and switching Ethernet technology and be improved, but bottom networked-induced delay can raising be
Lay a good foundation for distributed real-time systems, and the implementation strategy of communication protocol still greatly affect system real-time and
Synchronism.Then, OMG (Object Management Group, Object Management Organization) is in HLA (High Level
Architecture, high-level architecture) and CORBA (Common Object Request Broker Architecture,
Common Object Request Broker Architecture) etc. the New-generation distributed real time communication middleware Technology rule formulated on the basis of standards
Model --- DDS.
Although the time of DDS exploitation is very long, the communication middleware realized based on DDS mechanism is also relatively more,
Be the correlative study based on DDS mechanism itself it is few, among these includes DDS discovery mechanism, Qos (Quality of
Service) tactful quality services etc., are even more that rare people refers to, and Qos strategy is in communication quality about Qos Policy model
In but play a crucial role, and Qos strategy provided in DDS mechanism is that quantity is more, and simultaneity mechanism is complicated, big portion
Divide designer in use, due to many reasons such as the time cost of research and energy, does not go deep into the Qos machine excessively
System is mostly exploitation convenience using default setting provided by DDS related middleware manufacturer, therefore on communication quality, place
Reason efficiency etc. needs further improvement.
Summary of the invention
Goal of the invention of the invention is: in view of the above problems, providing a kind of Qos model based on RTI-DDS
Construction method.
The construction method of Qos model based on RTI-DDS of the invention, including the following steps:
Based on the Qos strategy that RTI-DDS is provided, construction strategy set, i.e., the Qos strategy provided based on RTI-DDS is obtained
Strategy set;
Feature extraction is carried out to each Qos strategy in strategy set, obtains initial characteristics collection;
By PCA principal component analysis and FA factor analysis is merged to initial characteristics collection table and dimension-reduction treatment, obtains
Final feature set;Wherein each Qos strategy includes at least a feature in the feature set;
Two classification processings of numerical value and configuration are carried out to feature set;
Two-dimensional multi-variable decision tree-model: the tree structure of the entity based on DDS is established, using the entity of DDS as one
Node is tieed up, corresponding tree structure, the corresponding Qos strategy protocol of each tree structure are obtained;And it is every in tree structure
Two classes second dimension node is set in the two-dimensions of a intra-node, and one type is that numerical value class second ties up node (linear Qos strategy
Model), it is another kind of to tie up node (non-linear Qos Policy model) for configuration class second;Wherein, each entity of DDS includes one group
Qos strategy;
Based on preset Qos demand parameter information (system operational parameters indication information), tieed up in node in numerical value class second,
The preset weight matrix of each Qos strategy of respectively each entity (one-dimensional node), wherein each numerical value of each Qos strategy
Category feature respectively corresponds one group of weighted value;It is tieed up in node in configuration class second, each Qos strategy of respectively each entity is preset
About the enabled parameter matrix of configuration mode (switching mode), the enabled parameter is the switching mode of quantization, that is, is respectively every
A kind of one corresponding numerical value of switching mode setting;Wherein weight matrix, enabled parameter matrix are one-dimensional vector, dimension and Qos
The index number that demand parameter information includes is identical;I.e. for numerical value category feature, there is one group of weight and make in each strategy
Energy parameter and each Qos demand parameter are corresponding;
Using numerical value category feature as unit of account, be superimposed all Qos strategy of each entity weight matrix and enabled parameter
The result of product of matrix is superimposed in tree structure as the Prediction Parameters matrix under current value category feature of current entity
All entities Prediction Parameters matrix, then be multiplied, obtained corresponding to each tree structure with current Qos demand parameter parameter
Qos strategy protocol system performance predicted value.
That is, constructing its corresponding two-dimensional multi-variable decision tree-model, base for each Qos strategy protocol to be predicted
In its output system performance predicted value (superposition tree structure in all entities Prediction Parameters matrix, then with currently
Qos demand parameter parameter be multiplied) obtain Qos strategy protocol predictive information to be predicted.
Wherein, the weight matrix and enabled parameter matrix of each strategy is configured by training mode of learning.
Further, the specific set-up mode of the weight matrix of each strategy and enabled parameter matrix are as follows:
It is that each strategy setting in strategy set configures reference value, as Qos strategy R-matrix based on empirical value;
Training learning sample collection is set: using the subset of strategy set as a sample object, while extracting each sample
Sample data of the system operational parameters index of object as each sample object;
And processing is filtered to the trained learning sample collection: calculating the Pearson correlation coefficient of sample, deletes Pierre
Inferior related coefficient is lower than the sample of preset threshold;
To the strategy comprising configuration class in strategy set, preset enabled parameter;Such as pass through training learning sample collection and Qos
Tactful R-matrix obtains its corresponding switching mode in a manner of training study, and corresponding based on preset switching mode
Quantized values obtain corresponding enabled parameter;
Based on the preset enabled parameter of training learning sample collection and Qos strategy R-matrix and part, acquisition strategy collection
The weight matrix and enabled parameter matrix of each strategy in conjunction:
Using each of feature set feature as a computing object o, and construct the computation model about computing object:
Wherein, An、BnAnd XnThe weight matrix, enabled parameter matrix and system operational parameters for respectively indicating n-th of strategy refer to
Matrix, n=1,2 ..., N are marked, parameter N indicates the Qos strategy number of strategy set;
AndWhereinIndicate n-th
J-th of system operational parameters index of a strategy,Corresponding j-th of the system operational parameters of n-th of strategy are respectively indicated to refer to
Target weight and enabled parameter, j=1,2 ..., J, parameter J indicate system operational parameters index number;
Based on the preset enabled parameter in part, corresponding enabled parameter in computation model is initialized;Wherein, for same
Strategy, the enabled parameter under different system operational parameters indexs can be the same;
Parametric solution is carried out to computation model based on Qos strategy R-matrix and training learning sample collection, obtains each plan
Weight matrix A slightlynWith enabled parameter matrix Bn;Wherein, when bringing the trained each sample of learning sample concentration into, by each sample
System operational parameters index corresponding to sheet arrives the sample as the system operational parameters index of strategy included by the sample
The system operational parameters matrix of each strategy involved in this;And by the sample strategy not to be covered weight zero setting.
Further, it is also based on multiple linear regression, using least square method, substitutes into training sample to computation model
Optimize, that is, be based on multiple linear regression, using least square method, weight matrix A that solution is obtainednWith enabled parameter
Matrix BnIt optimizes, further to promote accuracy.
Meanwhile the present invention also provides a kind of runnability prediction technique of Qos strategy protocol, following step is specifically included
It is rapid:
Step 1: the strategy set about Qos strategy is arranged in the Qos demand based on system;
Step 2: carrying out the attribute classification matching treatment of time, space and configuration to each of strategy set strategy, obtain
To the attribute classification matching result of each strategy;
Step 3: being respectively the tactful preset configuration reference value of each of strategy set, obtain Qos strategy R-matrix Q=
{Q1,Q2,......QN, wherein QnIndicate the configuration reference value of n-th of strategy in strategy set, n=1,2 ..., N, parameter N
Indicate the tactful quantity of strategy set;
Step 4: preset system operating parameter indication information, and respectively using time, space as computing object o, building about
The computation model of computing object:
Wherein An、BnAnd XnThe weight matrix, enabled parameter matrix and system operational parameters for respectively indicating n-th of strategy refer to
Mark matrix;AndWhereinIt indicates n-th
J-th of system operational parameters index of strategy,Respectively indicate corresponding j-th of the system operational parameters index of n-th of strategy
Weight and enabled parameter, j=1,2 ..., J, parameter J indicate system operational parameters index number;
Step 4: using the subset of strategy set as a sample object, while extracting the system operation of each sample object
Parameter index matrix, the sample data as each sample object;
The Pearson correlation coefficient of sample is calculated, the sample that Pearson correlation coefficient is lower than preset threshold is deleted, thus
To training sample set;
Step 5: to the strategy comprising configuration generic attribute in strategy set, preset enabled parameter;To obtain configuration attribute
Strategy enabled parameter matrix;Wherein, the enabled parameter for the same strategy, under different system operational parameters indexs
It can be identical;
Parametric solution is carried out to each computation model based on Qos strategy R-matrix and training sample set, obtains each strategy
Weight matrix AnWith enabled parameter matrix Bn;Wherein, when bringing the trained each sample of learning sample concentration into, by each sample
Corresponding system operational parameters index arrives the sample as the system operational parameters index of strategy included by the sample
The system operational parameters matrix of related each strategy;And by the sample strategy not to be covered weight zero setting;
Step 6: for current Qos strategy protocol (a subset of strategy set) to be predicted, the weight obtained based on 5
Matrix AnWith enabled parameter matrix BnIt obtains the weight matrix of each of described Qos strategy protocol to be predicted strategy and enables ginseng
Matrix number, then the system operational parameters index matrix based on the Qos strategy protocol to be predictedIt takes weight matrix, enable
Parameter matrix and matrixProduct as each strategy prediction export, each strategy for the Qos strategy protocol for adding up to be predicted
Prediction output obtain current Qos strategy protocol runnability predictive information to be predicted.
In conclusion by adopting the above-described technical solution, the beneficial effects of the present invention are: in DDS middle unit development
Time and the energy that developer can be saved, improve efficiency, according to RTI-DDS design in entity classification, as decision tree
The node of first dimension in each node, includes the corresponding Qos strategy of each section.In node, contain second dimension, by Qos
The offering question of strategy is divided into numerical indication (linear) and switchgear distribution (non-linear) two parts, is converted to is answered with the time respectively
Miscellaneous degree and space complexity are a node of the linear model of index as second dimension;And it is equally determined based on multivariable
The nonlinear model of plan tree construction, using this as second node of the second dimension.Linear model uses linear regression logarithm
The linear model of index optimizes.User can provide the feature and index of transmission data, derive from decision tree nodes
To suitable model parameter and Qos strategy, the complexity of this model modeling is low, and the calculating time is short, and cost is small.
Detailed description of the invention
Fig. 1 is that the structural diagrams that the first dimension of decision tree divides are intended in specific embodiment.
Fig. 2 is the second dimension decision tree part-structure citing in specific embodiment.
Fig. 3 is the part two dimension decision-tree model schematic diagram of the Publisher in specific embodiment, in Fig. 2.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below with reference to embodiment and attached drawing, to this hair
It is bright to be described in further detail.
In order to save time and the energy of developer in DDS middle unit development, improves efficiency, propose based on decision
The Qos Policy model of tree method is realized by the Qos Policy model and is ensured to the Qos for the DDS middleware developed.
The build process of Qos Policy model of the invention includes: that sample process, the building of Qos Policy model and model are excellent
Change processing, the specific implementation process is as follows:
S1: sample process:
S101:Qos policy characteristics cluster extracts.
In present embodiment, Qos policy characteristics include following six class:
1. time (f1).The strategy of DDS middleware time complexity is influenced with numeric form.
2. space (f2).The strategy of DDS middleware space complexity is influenced with numeric form.
3. enabled (f3).The strategy of Qos strategy use is influenced with switch form.
4. configuring (f4).Influence the strategy of Qos strategy use indirectly in the form that is arranged and can configure.
5. managing (f5).Influence the strategy of DDS middleware indirectly in the form of multi-option and prediction.
6. other (f6).Using user or system preference as the strategy of representative, or influence unrelated with DDS performance of middle piece
It is minimum.I.e. using user or system preference as the strategy of setting reference.
When some Qos strategy obviously contains features above, this feature is assigned to Qos strategy, meanwhile, determine Qos strategy
Under whether there are also multiple substrategies undetermined.
The strategy being now related to is summarized as follows (strategy set of present embodiment): table 1
Wherein, DR is indicated: Data-Reader (data subscription device);DW is indicated: Data_writer (data publication device);DP
It indicates: DomainParticipant (domain member);Rtps is indicated: Real-Time Publish-Subscribe (real-time release-
Subscription agreement), Rtps-ID indicates corresponding ID number.
S102: feature cleaning.
Since characteristic attribute is excessive, catastrophic problem will cause, feature is arranged and degraded at this, to agree with line
Property model and nonlinear model.
In present embodiment, by different cleaning processes three times as follows, the tune to entire Qos policy characteristics table is realized
It is whole:
(1) the first cleaning is carried out using principal component analysis (PCA, Principal Component Analysis).
Define Qos policy characteristics: feature={ f1,f2,f3,f4,f5,f6, and to it includes 6 characteristic values, carry out
It analyzes one by one, rejects or correlation unrelated for overall model and system performance minimum (i.e. correlation is lower than preset threshold)
Feature.The characteristic value for rejecting feature=" other " as a result, is left characteristic value: time, space, enabled, configuration, management, complete
At the dimensionality reduction of feature.
(2) the second cleaning is carried out using factorial analysis (FA, Factor Analysis).
Here, feature feature={ f1,f2,f3,f4,f5It is signal observing matrix, due to having under each feature cluster
Connection and influence between Qos strategy, therefore internal dependence is carried out to it and is combed, obtain: " time " and " space " is special
Sign cluster be it is in close relations, most of the time performance is to sacrifice space and be stored as cost, and vice versa;" management " and " configuration "
Cluster has common performance more, and inner link complexity is obscure;" enabled " feature can independent observation.Therefore second the present invention is based on FA is clear
It washes and specifically includes following cleanings twice:
Therefore carry out first based on FA first time cleaning: extract ingredient feature=" enabled ", this be it is original, can
That directly observes shows in variable, " enabled " feature cluster is removed, the preset parameter value being reduced in model, it may be assumed that when this parameter is 0
When, it is invalid to enable;When parameter is not 0, represents default and enable the strategy.
Then, carry out second based on FA cleaning: " time " and " space " cluster can directly representative models and middleware
Performance, " management " and " configuration " cluster relationship are obscure, partial function overlapping, therefore two features are merged, later " to match
Set " amalgamation result of representative " management " and " configuration " cluster.
It after the cleaning for completing feature, obtains feature={ " time ", " space ", " configuration " }, " time " and " space "
Cluster is put into linear model set, and " configuration " cluster is included into nonlinear model;
Based on obtained feature wash result feature={ " time ", " space ", " configuration " }, it is clear that feature is carried out to table 1
After washing, obtained strategy set is as shown in table 2:
Table 2
S102: sample collection.
Based on empirical value, each strategy setting in the strategy set enumerated by table 2 configures reference value, as Qos plan
Slightly R-matrix, i.e. experience optimal solution.
In order to realize sample collection, the present invention is primarily based on the Qos demand of system, and Provisioning Policy set is (as table 1 wraps
The strategy included), it is then based on characteristic descriptor set and closes { " time ", " space ", " configuration " }, be each of strategy set strategy
(time, space and/or configuration) is described with corresponding feature;It is again respectively the corresponding office of each strategy setting in strategy set
Allocation optimum reference value in portion's obtains Qos strategy R-matrix Q={ Q1,Q2,......QN, wherein Qn(n=1,2 ..., N) table
Show the local optimum reference value of Different Strategies, N indicates the tactful quantity of strategy set.
Pass through the various combination mode to the strategy in strategy set, available a variety of strategy combination (i.e. strategy sets
Subset), for realizing the Qos tactical management to corresponding object, be referred to as the completed policy combination of matching Qos demand.This
In invention, it regard Different Strategies combination as a sample object, while extracting the system operational parameters conduct of each sample object
The sample data (alternatively referred to as sample index) of each sample object.Wherein system operational parameters are alternatively referred to as Qos demand parameter
Parameter.
In present embodiment, the system operational parameters are specifically included as shown in table 3:
Table 3
Serial number | Title |
1 | Bandwidth (handling capacity) |
2 | Subscribe to obtain the ratio between with the data packet of publication |
3 | Subscribe to the obtained time interval with the data packet of publication |
In present embodiment, it is contemplated that can there is a situation where that exceptional sample and data possible accuracy be not high, pass through
It constantly extracts sample and carries out abnormality processing, until leaving 100000 samples, be used for model according to common ratio, 70000
It practises, 10000 for testing, and 20000 for being fitted recurrence and model verifying.
S103: sample data processing.
Exceptional sample processing, is based on collaborative filtering (CFR, Collaborative Filtering
Recommendation it) improves, using the collaborative filtering based on large sample mean value, the Pearson came for extracting sample is related
Coefficient, Pearson correlation coefficient calculation formula are as follows:
Wherein, X indicates the index matrix of each sample, i.e. X={ x1,x2,x3, x1,x2,x3It respectively corresponds shown in table 3
Three sample index;Y indicate all samples calculated each index Mean Matrix, i.e. Y={ Y1,Y2,Y3};Cov indicates association
Variance, σX,σYThe respectively standard deviation of X, Y, μX、μYThe respectively desired value of X, Y.
Such as indicate sample specificator with m, then the index matrix of m-th of sample can indicate are as follows:
ThenWherein, M indicates total sample number.
Then, screening sample processing is carried out according to the Pearson correlation coefficient of each sample calculated, that is, extracted
Pearson correlation coefficient is greater than sample of the sample of preset threshold as model learning and training.
In present embodiment, the Pearson correlation coefficient being calculated is divided into five grades, specifically such as 4 institute of table
Show:
Table 4
Range | Correlation |
0.8~1.0 | Extremely strong correlation |
0.6~0.8 | Strong correlation |
0.4~0.6 | Moderate correlation |
0.2~0.4 | Weak correlation |
0~0.2 | Extremely it is weak it is related, without related or negatively correlated |
Then by algorithm filter, extract degree of correlation in the medium above sample, and for different brackets (strong correlation,
Extremely strong correlation) corresponding sample is respectively set, such as 1000 samples are respectively set, wherein 500 for learning, and 500 for instructing
Practice.
Step S2: the building of model.
Step S201: the nonlinear model frame based on switchgear distribution is established.
It establishes the model framework based on two-dimensional multivariable decision tree: being designed a model according to DDS, with the class of entity in DDS
Type divides as shown in figure Fig. 1 (tree structure of the entity of i.e. existing DDS) as node, the first dimension of decision tree, wherein
DPfactory indicates DomainParticipant-Factory (domain member factory), and Publisher indicates publisher,
Subscriber indicates that subscriber, Topic indicate theme.Wherein, the Qos strategy of each one-dimensional node is as shown in table 5:
Table 5
In each node of first dimension, all contain two child nodes, respectively correspond two seed nodes: numeric type and matching
Set type, that is, linearity and non-linearity model.Wherein, linear model includes two nodes of " time " and " space " feature cluster;
And in nonlinear model, certain nodes include multiple from strategy.By taking Publiser as an example, the second dimension decision tree part-structure is lifted
Such as shown in Fig. 2;, according to qos policy classifying content, refine out part two dimension decision-tree model such as Fig. 3 institute of Publisher
Show, wherein dotted portion represents switch relevant to present node, and bold portion is derivative node.In Fig. 3, Pulish_mode_
Switch is indicated: release model switch;Availability_switch is indicated: utilizability switch;Durability_
Switch is indicated: persistence switch;Deadline_switch is indicated: deadline switch;History_switch is indicated: being gone through
Records of the Historian record switch;Kind_switch is indicated: type of switch;Depth_switch is indicated: depth switch.
Step S202: the linear model based on numerical value is established:
For the node of " time " and " space " feature cluster, with time complexity o (Time) and space complexity o
(space) lvalue, its corresponding performance indicator are used as;In the Qos strategy containing " time " and " space " feature, each is had
The strategy of numeric type will all participate in model, it influences time, spatial model with weight size, and enabling parameter is (to inductive switch
The quantized value of mode, such as two switches, quantifiable codomain are { 0,1 }, wherein 0 indicates to close, 1 indicates to open), then tie
Sample index X is closed, using the linear of all policies in all policies set and as r value, it is established that linear model is as follows: its
In, " time " linear model of cluster is as follows:
Wherein, It respectively indicates under time model, the corresponding jth of n-th of strategy
The weight of a sample index and enabled parameter, xt1,xt2,xt3It indicates under time model, corresponding three sample index of strategy (gulp down
The amount of spitting, subscription obtain with the ratio between the data packet of publication and subscribe to the obtained time interval with the data packet of publication), due to right
For the strategy included by each sample, corresponding to sample index matrix X it is identical, therefore while calculating can not have to distinguish it is different
Strategy.
Similarly, the linear model of " space " cluster can be obtained:
Wherein, It respectively indicates under spatial model, the corresponding jth of n-th of strategy
The weight of a sample index and enabled parameter, xs1,xs2,xs3Under representation space model, corresponding three sample index of strategy.
It is translated into matrix form, defines matrix: Wherein, subscript λ ∈ { t, s }, n
=1,2 ..., N, ()TRepresenting matrix transposition;Then " time ", " space " cluster linear model can convert the square being as follows
Formation formula:
O (Time)=Kt(Xt)T;
O (Space)=Ks(Xs)T;
Step S203: the study of the nonlinear model based on configuration.
By learning sample Sample={ s1,s2......s70000And experience optimal solution Q be put into based on the non-thread of configuration
Property decision tree inside, obtain one group of switch solution R={ r of nonlinear model1,r2,......rm, wherein ri(i=1,2,
It 3......m) is the amount of enumerating to inductive switch, m indicates the quantity that switch solution can be obtained in strategy set, and different enumerates
Value represent different switches dial method (mode), by the switch solution can determine " time ", " space " cluster linear model in
It is partially enabled parameterRemaining study for then passing through line drag obtains.
The verifying of model function: after bringing switch solution into, pass through pre-prepd test sample (such as Sample'=
{s1',s2'......s10000') functional verification is carried out, in order to the availability of detection model.
Step S204: the study of the linear model based on numerical value.
Based on pre- previously ready learning sample, in present embodiment, it is related to 70000 learning samples altogether, and
Definition study sample set are as follows: Sample={ s1,s2......s70000, in conjunction with experience optimal solution Q and switch solution R=
{r1,r2,......rm, respectively to linear model (time, space segment) carry out parameter learning (learning sample and it is corresponding most
Excellent solution is put into inside linear dimensions), respectively obtain the parameter dematrix in space, timeWith
For any one sample, for example, it include strategy set in the 1st, 3 and 5 strategy, then bring into " when
Between ", the linear model of " space " cluster when, in the study processing of model, directly enable that be not belonging to strategy corresponding to the sample
The tactful weight of collection is 0.
Step S3: the optimization of model
Step S301: the numerical model optimization based on linear regression:
For the linear model of obtained o (Time) and o (Space), what is obtained by model is a predicted value
(system operational parameters about a group policy subset), are denoted as F (X), it and original o (Time) and o (Space) are inclined
Difference is based on linear regression principle, substitutes into ready training sample Sample "={ s1”,s2”......s20000" using minimum
Square law optimizes its model parameter, and the numerical matrix of o (Time) Yu o (Space) are indicated with Y ', this process is converted
It is as follows for mathematic(al) representation:
Wherein, Y '={ y '1,y′2,…y′N, indicate model (time, space) corresponding numerical matrix, X ' expression strategy
Index matrix (the i.e. corresponding index matrix X={ x of sample belonging to each of strategy set strategy of set1,x2,x3Composition
Matrix), i.e. X '={ x '1,x′2,…x′N, wherein Indicate three fingers of n-th of strategy
Mark information: handling capacity, subscription obtain with the ratio between the data packet of publication and subscribe between the obtained time with the data packet of publication
Every;K={ k1,k2,......kN, indicate tactful coefficient matrix (product of weight and enabled parameter).
Solve parameter matrix K={ k1,k2,......kN, so that obtaining Optimal Parameters when E (o (Y), X) this value minimum
K'={ k'1,k'2,......k'N}。
The solution procedure of this multiple linear regression is as follows:
By E (o (Y), X) to x 'nDerivation is carried out, is obtained:
Enabling this expression formula is zero, obtains knOptimal solution closed solutions:
To each x 'nAfter solution, Optimal Parameters K'={ k' is obtained1,k'2,......k'n}。
So that it is determined that in strategy set each strategy weight and enabled information, and save.
When needing to predict the system performance of certain group Qos strategy subset (Qos strategy protocol), then based on acquired
The weight of relative strategy and enabled information are based on the linear model of o (Time) and o (Space), available current Qos strategy
Time and spatial information of the scheme when system is run can run for the prediction of certain Qos strategy protocol through the invention
Performance.To provide reference information to the selection of Qos strategy protocol for system Qos personnel.
The above description is merely a specific embodiment, any feature disclosed in this specification, except non-specifically
Narration, can be replaced by other alternative features that are equivalent or have similar purpose;Disclosed all features or all sides
Method or in the process the step of, other than mutually exclusive feature and/or step, can be combined in any way.
Claims (8)
1. the construction method of the Qos model based on RTI-DDS, characterized in that it comprises the following steps:
Based on the Qos strategy that RTI-DDS is provided, construction strategy set;
Feature extraction is carried out to each Qos strategy in strategy set, obtains initial characteristics collection;
By PCA principal component analysis and FA factor analysis is merged to initial characteristics collection table and dimension-reduction treatment, obtains final
Feature set;Wherein each Qos strategy includes at least a feature in the feature set;
Two classification processings of numerical value and configuration are carried out to feature set;
Two-dimensional multi-variable decision tree-model: the tree structure of the entity based on DDS is established, using the entity of DDS as one-dimensional section
Point obtains corresponding tree structure, the corresponding Qos strategy protocol of each tree structure;And each section in tree structure
Two classes second dimension node is set in the two-dimensions inside point, and one type is that numerical value class second ties up node, another kind of for configuration class
Second dimension node;Wherein, each entity of DDS includes one group of Qos strategy;
Based on preset Qos demand parameter information, tieed up in node in numerical value class second, each Qos plan of respectively each entity
Slightly preset weight matrix, wherein each numerical value category feature of each Qos strategy respectively corresponds one group of weighted value;In configuration class the
In two-dimentional node, the preset enabled parameter matrix about configuration mode of each Qos strategy of respectively each entity is described enabled
Parameter is the switching mode of quantization;Wherein weight matrix, enabled parameter matrix are one-dimensional vector, and dimension and Qos demand parameter are believed
The index number that breath includes is identical;
Using numerical value category feature as unit of account, it is superimposed the weight matrix and enabled parameter matrix of all Qos strategy of each entity
Result of product, as the Prediction Parameters matrix under current value category feature of current entity, the institute that is superimposed in tree structure
There is the Prediction Parameters matrix of entity, then be multiplied with current Qos demand parameter parameter, obtains corresponding to each tree structure
The system performance predicted value of Qos strategy protocol.
2. the method as described in claim 1, which is characterized in that the weight matrix of each strategy and enabled parameter matrix it is specific
Set-up mode are as follows:
It is that each strategy setting in strategy set configures reference value, as Qos strategy R-matrix based on empirical value;
Training learning sample collection is set: using the subset of strategy set as a sample object, while extracting each sample object
Sample data of the system operational parameters index as each sample object;
And processing is filtered to the trained learning sample collection: calculating the Pearson correlation coefficient of sample, deletes Pearson came phase
Relationship number is lower than the sample of preset threshold;
To the strategy comprising configuration class in strategy set, preset enabled parameter;
Based on the preset enabled parameter of training learning sample collection and Qos strategy R-matrix and part, in acquisition strategy set
The weight matrix and enabled parameter matrix of each strategy:
Using each of feature set feature as a computing object o, and construct the computation model about computing object:
Wherein, An、BnAnd XnRespectively indicate the weight matrix, enabled parameter matrix and system operational parameters index square of n-th of strategy
Battle array, n=1,2 ..., N, parameter N indicate the Qos strategy number of strategy set;
AndWhereinIndicate n-th of plan
J-th of system operational parameters index slightly,Respectively indicate corresponding j-th of the system operational parameters index of n-th of strategy
Weight and enabled parameter, j=1,2 ..., J, parameter J indicate system operational parameters index number;
Based on the preset enabled parameter in part, corresponding enabled parameter in computation model is initialized;
Parametric solution is carried out to computation model based on Qos strategy R-matrix and training learning sample collection, obtains each strategy
Weight matrix AnWith enabled parameter matrix Bn;Wherein, when bringing the trained each sample of learning sample concentration into, by each sample institute
Corresponding system operational parameters index arrives the sample institute as the system operational parameters index of strategy included by the sample
Each of it is related to the system operational parameters matrix of strategy;And by the sample strategy not to be covered weight zero setting.
3. method according to claim 2, which is characterized in that be based on multiple linear regression, using least square method, substitute into instruction
Practice sample to the weight matrix A of computation modelnWith enabled parameter matrix BnOptimize processing.
4. the method as described in claim 1, which is characterized in that initial characteristics collection specifically includes:
Time: the strategy of DDS middleware time complexity is influenced with numeric form;
Space: the strategy of DDS middleware space complexity is influenced with numeric form;
It is enabled: the strategy of Qos strategy use is influenced with switch form;
Configuration: the strategy of Qos strategy use is influenced indirectly in the form that is arranged and can configure;
Management: the strategy of DDS middleware is influenced indirectly in the form of multi-option and prediction;
Other: are using user or system preference as the strategy of setting reference.
5. method as claimed in claim 4, which is characterized in that initial characteristics collection merges and dimension-reduction treatment, obtains final
Feature set are as follows: { " time ", " space ", " configuration " }, wherein being configured to the amalgamation result for enabling, configuring, managing three features.
6. a kind of runnability prediction technique of Qos strategy protocol, specifically includes the following steps:
Step 1: the strategy set about Qos strategy is arranged in the Qos demand based on system;
Step 2: carrying out the attribute classification matching treatment of time, space and configuration to each of strategy set strategy, obtain every
The attribute classification matching result of a strategy;
Step 3: being respectively the tactful preset configuration reference value of each of strategy set, obtain Qos strategy R-matrix Q={ Q1,
Q2,……QN, wherein QnIndicate the configuration reference value of n-th of strategy in strategy set, n=1,2 ..., N, parameter N indicates plan
The tactful quantity slightly gathered;
Step 4: preset system operating parameter indication information, and respectively using time, space as computing object o, it constructs about calculating
The computation model of object:
Wherein An、BnAnd XnRespectively indicate the weight matrix, enabled parameter matrix and system operational parameters index square of n-th of strategy
Battle array;AndWhereinIndicate n-th of strategy
J-th of system operational parameters index,Respectively indicate the power of corresponding j-th of the system operational parameters index of n-th of strategy
Weight and enabled parameter, j=1,2 ..., J, parameter J indicate system operational parameters index number;
Step 4: using the subset of strategy set as a sample object, while extracting the system operational parameters of each sample object
Index matrix, the sample data as each sample object;
The Pearson correlation coefficient of sample is calculated, the sample that Pearson correlation coefficient is lower than preset threshold is deleted, to be instructed
Practice sample set;
Step 5: to the strategy comprising configuration generic attribute in strategy set, preset enabled parameter;To obtain the plan of configuration attribute
Enabled parameter matrix slightly;
Parametric solution is carried out to each computation model based on Qos strategy R-matrix and training sample set, obtains the power of each strategy
Weight matrix AnWith enabled parameter matrix Bn;Wherein, when bringing the trained each sample of learning sample concentration into, each sample institute is right
The system operational parameters index answered arrives involved by the sample as the system operational parameters index of strategy included by the sample
And each of strategy system operational parameters matrix;And by the sample strategy not to be covered weight zero setting;
Step 6: for current Qos strategy protocol to be predicted, the weight matrix A obtained based on 5nWith enabled parameter matrix Bn?
To the weight matrix and enabled parameter matrix of each of the Qos strategy protocol to be predicted strategy, then based on described to be predicted
Qos strategy protocol system operational parameters index matrixTake weight matrix, enabled parameter matrix and matrixProduct make
It is exported for the prediction of each strategy, the prediction of each of Qos strategy protocol for adding up to be predicted strategy exports to obtain current to be predicted
Qos strategy protocol runnability predictive information.
7. method as claimed in claim 6, which is characterized in that system operational parameters indication information includes 3 system operation ginsengs
Number indexs: bandwidth, subscription obtain with the ratio between the data packet of publication and subscribe to the obtained time interval with the data packet of publication.
8. method according to claim 6 or 7, which is characterized in that in step 4, Pearson correlation coefficient is lower than to 0.6 sample
This deletion, to obtain training sample set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910175135.4A CN109905481B (en) | 2019-03-08 | 2019-03-08 | Qos model construction and Qos strategy scheme operation performance prediction method based on RTI-DDS |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910175135.4A CN109905481B (en) | 2019-03-08 | 2019-03-08 | Qos model construction and Qos strategy scheme operation performance prediction method based on RTI-DDS |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109905481A true CN109905481A (en) | 2019-06-18 |
CN109905481B CN109905481B (en) | 2020-09-25 |
Family
ID=66946840
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910175135.4A Active CN109905481B (en) | 2019-03-08 | 2019-03-08 | Qos model construction and Qos strategy scheme operation performance prediction method based on RTI-DDS |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109905481B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112532422A (en) * | 2020-10-28 | 2021-03-19 | 南京航空航天大学 | HLA/DDS bridge based on mixed time management system |
CN113094614A (en) * | 2019-12-23 | 2021-07-09 | 中国移动通信集团湖南有限公司 | Data distribution method, system and device |
CN113472590A (en) * | 2021-07-14 | 2021-10-01 | 南京邮电大学 | QoS configuration method based on machine learning in DDS |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103166975A (en) * | 2013-04-03 | 2013-06-19 | 上海航天测控通信研究所 | Data distribution service (DDS) communication system |
US20130346580A1 (en) * | 2012-02-12 | 2013-12-26 | Electronics And Telecommunications Research Institute | Apparatus and method for generating qos profile for supporting data distribution service in cyber-physical system |
US20150370546A1 (en) * | 2014-06-24 | 2015-12-24 | Electronics And Telecommunications Research Institute | Method and apparatus for generating data distribution service application |
US20160337272A1 (en) * | 2015-05-12 | 2016-11-17 | Jeda Networks, Inc. | Methods, systems and apparatus for the virtualization of high performance networks |
CN107070539A (en) * | 2017-05-03 | 2017-08-18 | 成都国恒空间技术工程有限公司 | Seasat Radio Data System and method |
CN107404512A (en) * | 2016-05-19 | 2017-11-28 | 华为技术有限公司 | Resource subscription method, resource subscription device and resource subscription Xi System |
CN109412890A (en) * | 2018-10-17 | 2019-03-01 | 哈尔滨工业大学 | Combined Trials platform middleware node state detection method based on DDS |
-
2019
- 2019-03-08 CN CN201910175135.4A patent/CN109905481B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20130346580A1 (en) * | 2012-02-12 | 2013-12-26 | Electronics And Telecommunications Research Institute | Apparatus and method for generating qos profile for supporting data distribution service in cyber-physical system |
CN103166975A (en) * | 2013-04-03 | 2013-06-19 | 上海航天测控通信研究所 | Data distribution service (DDS) communication system |
US20150370546A1 (en) * | 2014-06-24 | 2015-12-24 | Electronics And Telecommunications Research Institute | Method and apparatus for generating data distribution service application |
US20160337272A1 (en) * | 2015-05-12 | 2016-11-17 | Jeda Networks, Inc. | Methods, systems and apparatus for the virtualization of high performance networks |
CN107404512A (en) * | 2016-05-19 | 2017-11-28 | 华为技术有限公司 | Resource subscription method, resource subscription device and resource subscription Xi System |
CN107070539A (en) * | 2017-05-03 | 2017-08-18 | 成都国恒空间技术工程有限公司 | Seasat Radio Data System and method |
CN109412890A (en) * | 2018-10-17 | 2019-03-01 | 哈尔滨工业大学 | Combined Trials platform middleware node state detection method based on DDS |
Non-Patent Citations (4)
Title |
---|
MOHAMED MAZOUZI等: "Challenges and solutions in configuring, rapid developing and deploying of a QoS-enabled component middleware", 《2008 3RD INTERNATIONAL DESIGN AND TEST WORKSHOP》 * |
YIBIN TU等: "Research on Improvement of KD-RTI Communication Mechanism Base on DDS", 《2016 3RD INTERNATIONAL CONFERENCE ON INFORMATION SCIENCE AND CONTROL ENGINEERING (ICISCE)》 * |
冷枫麟: "基于发布/订阅的分布式数据分发技术的研究与实现", 《万方学位论文数据库》 * |
梁林波: "基于RTI-DDS的无人机地面站通信系统研究与设计", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113094614A (en) * | 2019-12-23 | 2021-07-09 | 中国移动通信集团湖南有限公司 | Data distribution method, system and device |
CN112532422A (en) * | 2020-10-28 | 2021-03-19 | 南京航空航天大学 | HLA/DDS bridge based on mixed time management system |
CN112532422B (en) * | 2020-10-28 | 2021-12-07 | 南京航空航天大学 | HLA/DDS bridge based on mixed time management system |
CN113472590A (en) * | 2021-07-14 | 2021-10-01 | 南京邮电大学 | QoS configuration method based on machine learning in DDS |
Also Published As
Publication number | Publication date |
---|---|
CN109905481B (en) | 2020-09-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Bodin et al. | Improving network approaches to the study of complex social–ecological interdependencies | |
CN109905481A (en) | A kind of building of the Qos model based on RTI-DDS and the runnability prediction technique of Qos strategy protocol | |
Latora et al. | Social cohesion, structural holes, and a tale of two measures | |
CN108874959A (en) | A kind of user's dynamic interest model method for building up based on big data technology | |
Liao et al. | Accurate sub-swarms particle swarm optimization algorithm for service composition | |
Du | The causal relationship between land urbanization quality and economic growth: Evidence from capital cities in China | |
DE102018110138A1 (en) | Workflow engine framework | |
Fanjiang et al. | Search based approach to forecasting QoS attributes of web services using genetic programming | |
Krishnaswamy et al. | Data-driven learning to predict wan network traffic | |
Xu | A practical procedure for group decision making under incomplete multiplicative linguistic preference relations | |
Spangenberg et al. | A Big Data architecture for intra-surgical remaining time predictions | |
Dornhöfer et al. | A data-driven smart city transformation model utilizing the green knowledge management cube | |
Van der Heijden et al. | Network relationships and standard adoption: Diffusion effects in transnational regulatory networks | |
KR20210042709A (en) | Method and server for stock movement prediction using corporate relation data | |
Tan et al. | A method towards Web service combination for cross-organisational business process using QoS and cluster | |
Mercuur et al. | The use of values for modeling social agents | |
Seidl et al. | Niklas Luhmann’s autopoietic theory of organisations: Contributions, limitations, and future prospects | |
Vellucci et al. | Microscopic modeling and analysis of collective decision making: equality bias leads suboptimal solutions | |
Schulz et al. | A network approach to consumption | |
Cunha et al. | Integrated Reporting and Group Decision-Making for Sustainable Development: A Bibliometric Review | |
Khanra et al. | Orchestration of an e-government network: Capturing the dynamics of e-government service delivery through theoretical analysis and mathematical forecasting | |
Ye et al. | The pvad algorithm to learn partial-value variable associations with application to modelling for engineering retention | |
Duggan | A simulator for continuous agent-based modelling | |
CN117422206B (en) | Method, equipment and storage medium for improving engineering problem decision and scheduling efficiency | |
Zheng et al. | The Electrical Load Forecasting Base on an Optimal Selection Method of Multiple Models in DSM. |
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 |