CN109343951A - Mobile computing resource allocation methods, computer readable storage medium and terminal - Google Patents
Mobile computing resource allocation methods, computer readable storage medium and terminal Download PDFInfo
- Publication number
- CN109343951A CN109343951A CN201810933422.2A CN201810933422A CN109343951A CN 109343951 A CN109343951 A CN 109343951A CN 201810933422 A CN201810933422 A CN 201810933422A CN 109343951 A CN109343951 A CN 109343951A
- Authority
- CN
- China
- Prior art keywords
- sample
- normalized
- concentrated
- training dataset
- computing resource
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR 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]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Abstract
A kind of mobile computing resource allocation methods, computer readable storage medium and terminal, which comprises to the computation requests that send of received mobile terminal analyze, obtain corresponding calculating task data to be predicted；Obtained calculating task data to be predicted are inputted into trained computing resource prediction model in advance, obtain the classification results of the calculating task data to be predicted；Obtained classification results are made decisions, and using the corresponding calculating task of computation requests described in the mobile terminal corresponding with the court verdict or mobile edge calculations server process.Above-mentioned scheme can calculate for shifter edge and distribute reasonable computing resource, to improve the utilization rate of resource.
Description
Technical field
It, can more particularly to a kind of mobile computing resource allocation methods, computer the present invention relates to internet of things field
Read storage medium and terminal.
Background technique
With the rapid development and extensive use of cloud and artificial intelligence technology, more and more novel intelligent Internet of Things
Equipment is come into being.In order to meet the service quality and user experience of Intelligent internet of things application, mobile edge calculations are introduced
(Mobile Edge Computing, MEC), using the available resources moved in edge cloud network, to improve the use body of user
It tests.
Mobile edge calculations, can be regarded as the extension cloud computing model from core net to edge access network, be a kind of
Height virtualization technology, by the way that the factors such as transmission delay, energy consumption, fund and operation cost are added, to promote the mobility of network
Support, real-time, interactive and scalability.QoS of customer not only can be improved in mobile edge calculations, can also reduce network biography
Defeated data volume.
However, how to carry out the reasonable of computing resource when a large amount of calculating tasks are transferred to mobile edge and are calculated
Distribution, to improve the utilization rate of resource, becomes urgent problem to be solved.
Summary of the invention
Present invention solves the technical problem that being how to calculate the reasonable computing resource of distribution for shifter edge, to improve resource
Utilization rate.
In order to solve the above technical problems, the embodiment of the invention provides a kind of mobile computing resource allocation methods, the side
Method includes:
To the computation requests that send of received mobile terminal analyze, obtain corresponding calculating task number to be predicted
According to；
Obtained calculating task data to be predicted are inputted into trained computing resource prediction model in advance, are obtained described
The classification results of calculating task data to be predicted；
Obtained classification results are made decisions, and using the mobile terminal corresponding with the court verdict or
The corresponding calculating task of computation requests described in mobile edge calculations server process.
Optionally, computing resource prediction model training by the way of following obtains:
Training dataset and predictive data set are constructed respectively using history computational resource allocation data；
Constructed training dataset and predictive data set are normalized respectively；
The functional expression of best hyperplane is constructed, so that the training data that the best hyperplane will obtain after normalized
The sample that collection and prediction data are concentrated is divided into two classes；
The sample concentrated based on the training dataset obtained after normalized and prediction data is to the best hyperplane
Distance minimum value, the constraint condition of the best hyperplane is set；
The sample and the constraint item concentrated based on the training dataset obtained after the normalized and prediction data
Part determines the best configuration parameter of the best hyperplane；
The functional expression that the best configuration parameter is substituted into the best hyperplane obtains the computing resource prediction mould
Type.
Optionally, the sample concentrated based on the training dataset obtained after the normalized and prediction data and
The constraint condition determines the best configuration parameter of the best hyperplane, comprising:
The sample that the training dataset and prediction data that obtain after the normalized are concentrated substitutes into the constraint item
Part obtains the best configuration parameter of the best hyperplane.
Optionally, the constraint condition are as follows:
yi[(w·xi)+b] >=1, i=1,2 ..., l；
Wherein, yiIndicate the i-th sample that the training dataset obtained after the normalized and prediction data are concentrated
Feature tag, xiIndicate the training dataset obtained after the normalized and the i-th sample that prediction data is concentrated, wxi+b
Indicate that the best hyperplane, w indicate the Slope Parameters of the best hyperplane, b indicates that the plane of the best hyperplane is normal
Number parameter, l indicate the training dataset obtained after normalized and the total sample number that prediction data is concentrated.
Optionally, the sample concentrated based on the training dataset obtained after the normalized and prediction data and
The constraint condition determines the best configuration parameter of the best hyperplane, comprising:
It is respectively set for the sample that the training dataset and prediction data that obtain after the normalized are concentrated corresponding
Relaxation vector；
The sample and the best hyperplane concentrated based on the training dataset obtained after normalized and prediction data
The distance between maximize and minimize the error, construct about the best hyperplane Slope Parameters and the normalized
The objective function of the relaxation vector for the sample that the training dataset and prediction data obtained afterwards is concentrated；
The dual function of the objective function, and the instruction that will be obtained after normalized are sought using method of Lagrange multipliers
The sample for practicing data set and prediction data concentration substitutes into the dual function respectively, seeks the maximum value of the dual function, makees
For the minimum value of the objective function；
By the striked corresponding Slope Parameters value of minimum value, the best configuration number of the slope as the best hyperplane
Value；
By the best configuration parameter of the slope of the best hyperplane and the training dataset that will be obtained after normalized
The constraint condition is substituted into the sample that prediction data is concentrated, obtains the best configuration number of the plane constant of the best hyperplane
Value.
Optionally, the training number obtained after the Slope Parameters Yu the normalized of the constructed best hyperplane
According to the objective function of collection and the relaxation vector of the sample of prediction data concentration are as follows:
Wherein, C indicates preset nonnegative constant, ξiIndicate the training dataset obtained after the normalized and prediction
The relaxation vector of the i-th sample in data set.
Optionally, the dual function of the objective function are as follows:
And:
λi>=0, λi≥0；
Wherein, L (λ, w, b) indicates that the dual function of the objective function, λ indicate Lagrange multiplier, λiIt indicates
The Lagrange multiplier for i-th of sample that the training dataset and prediction data obtained after the normalized is concentrated, λiTable
Show the Lagrange for j-th of sample that the training dataset for indicating to obtain after the normalized and prediction data are concentrated
Multiplier.
Optionally, obtained classification results are made decisions using following formula:
Wherein, f (x) indicates that, to the court verdict for the unknown vector x for belonging to class, sgn () indicates return function,
The embodiment of the invention also provides a kind of computer readable storage mediums, are stored thereon with computer instruction, described
The step of computer instruction executes mobile computing resource allocation methods described in any of the above embodiments when running.
The embodiment of the invention also provides a kind of terminal, including memory and processor, energy is stored on the memory
Enough computer instructions run on the processor, the processor execute any of the above-described when running the computer instruction
The step of described mobile computing resource allocation methods.
Compared with prior art, the technical solution of the embodiment of the present invention has the advantages that
Above-mentioned scheme, it is pre- by the way that obtained calculating task data to be predicted are inputted trained computing resource in advance
Model is surveyed, obtains the classification results of the calculating task data to be predicted, and make decisions to obtained classification results, thus
It can be asked using being calculated described in the mobile terminal corresponding with the court verdict or mobile edge calculations server process
Corresponding calculating task is sought, computing resource can rationally be divided between the mobile terminal and mobile Edge Server
Match, compared with the calculating task of all mobile terminals is submitted to mobile edge calculations server process, so as to mention
The treatment effeciency of high mobile computing task, improves the utilization rate of resource.
Detailed description of the invention
Fig. 1 is a kind of flow diagram of mobile computing resource allocation methods of the embodiment of the present invention；
Fig. 2 shows the flow diagrams of the training method of one of embodiment of the present invention computing resource prediction model；
Fig. 3 is a kind of flow diagram of autonomous distributor of mobile computing resource of the embodiment of the present invention.
Specific embodiment
Technical solution in the embodiment of the present invention passes through obtained calculating task data input training in advance to be predicted
Good computing resource prediction model obtains the classification results of the calculating task data to be predicted, and ties to obtained classification
Fruit makes decisions, so as to using the mobile terminal corresponding with the court verdict or mobile edge calculations server
Handle the corresponding calculating task of the computation requests, can by computing resource the mobile terminal and mobile Edge Server it
Between carry out reasonable distribution, submit to mobile edge calculations server process phase with by the calculating task of all mobile terminals
Than improving the utilization rate of resource so as to improve the treatment effeciency of mobile computing task.
It is understandable to enable above-mentioned purpose of the invention, feature and beneficial effect to become apparent, with reference to the accompanying drawing to this
The specific embodiment of invention is described in detail.
Fig. 1 is a kind of flow diagram of mobile computing resource allocation methods of the embodiment of the present invention.Referring to Fig. 1, this hair
A kind of mobile computing resource allocation methods of bright embodiment, can specifically include following step:
Step S101: to the computation requests that send of received mobile terminal analyze, obtain corresponding meter to be predicted
Calculate task data.
In specific implementation, when needing to handle mobile computing task, user can submit corresponding meter by client layer
Calculate request.It wherein, include corresponding calculating task and the processing speed and consumption of the calculating task in the computation requests that user is submitted
Calculating task data including the demands such as energy.Therefore, when receiving the computation requests of user's submission, by being submitted to user
Computation requests parsed, can determine corresponding calculating task data.
Step S102: by obtained calculating task data input to be predicted, trained computing resource predicts mould in advance
Type obtains the classification results of the calculating task data to be predicted.
In specific implementation, the computing resource prediction model is to be carried out using the mobile computing resource allocation data of history
Training obtains, and the calculating task that user submits can be divided into two different classes, obtain corresponding two classification results.Its
In, the training process of the computing resource prediction model refers to the detailed description in Fig. 2.
Step S103: making decisions obtained classification results, and uses the shifting corresponding with the court verdict
The corresponding calculating task of computation requests described in dynamic terminal or mobile edge calculations server process.
It in specific implementation, will be acquired when obtaining two classification results of the calculating task data that the user submits
Two classification results input preset discrimination model, corresponding court verdict can be obtained.In this year invention one is implemented, adopt
Obtained classification results are made decisions with following formula:
Wherein, f (x) indicates that, to the court verdict for the unknown vector x for belonging to class, sgn () indicates return function,
Wherein, when the result being calculated by formula (1) is 1, the calculating task is handled using mobile terminal；Instead
It, when the result being calculated by formula (1) is -1, then using calculating task described in mobile edge calculations server.
The training method of the computing resource prediction model in the embodiment of the present invention will be described in detail below.
Fig. 2 shows the flow diagrams of the training method of one of embodiment of the present invention computing resource prediction model.
As shown in Fig. 2, the training method of one of embodiment of the present invention computing resource prediction model, is suitable for using history computing resource
Distribution data training obtains the computing resource prediction model classified to the calculating task of mobile terminal, specifically can be using such as
Under operation realize:
Step S201: training dataset and predictive data set are constructed respectively using history computational resource allocation data.
In specific implementation, the training data concentration includes obtaining the sample of the computing resource prediction model for training
Notebook data, it includes for obtaining the sample number that the computing resource prediction model is tested to training that the prediction data, which is concentrated,
According to.
Step S202: constructed training dataset and predictive data set are normalized respectively.
It in specific implementation, can be first to constructed instruction in order to improve the classification accuracy to calculating task data
The sample for practicing data set and prediction data concentration is normalized.In an embodiment of the present invention, using following formula
The sample concentrated to constructed training dataset and prediction data is normalized:
xnew=(xpre-xmin)/(xmax-xmin) (2)
Wherein, xminAnd xmaxThe minimal eigenvalue and maximum feature that the respectively described training dataset and prediction data are concentrated
Value, xpreIt is the sample before normalized, xnewIt is the sample after normalized.
Step S203: constructing the functional expression of best hyperplane, so that the best hyperplane will obtain after normalized
Training dataset and prediction data concentrate sample be divided into two classes.
In specific implementation, the sample that training dataset and prediction data are concentrated can be divided into two by the best hyperplane
Class, functional expression namely the best hyperplane can indicate are as follows:
Wx+b=0 (3)
Wherein, w indicates the Slope Parameters of the best hyperplane, indicates that the dot product of vector, b indicate described best super flat
The plane constant parameter in face.
Step S204: based on the sample that the training dataset obtained after normalized and prediction data are concentrated arrive described in most
The constraint condition of the best hyperplane is arranged in the minimum value of the distance of good hyperplane.
In specific implementation, it is normalized in the sample to the training dataset and prediction data concentration
Afterwards, corresponding feature tag is respectively set in the sample that can be concentrated to training dataset after normalized and prediction data, i.e.,
For the l sample in predeterminable area, each sample constitutes corresponding vector, i.e. (x with corresponding label1, y1), (x2,
y2) ..., (xl, yl), i ∈ (0, l).Wherein, xiI-th of sample that training dataset and prediction data after indicating normalization are concentrated
This, yiSample x after indicating normalizationiFeature tag, identify the class classified belonging to computing resource belonging to corresponding sample
Not.
When finding the best configuration parameter of the optimal hyperplane, with the training dataset and prediction number after normalization
Distance according to two class samples to the best hyperplane of concentration is maximum, and for the training dataset and prediction number after normalization
According to all sample x of concentrationi, so that:
So, the minimum range of sample and this best hyperplane can indicate are as follows:
|w·xi+ b |/| w |=1/ | w | (5)
Therefore, according to formula (4) and (5) it can be concluded that the constraint condition that the best hyperplane should meet are as follows:
yi[(w·xi)+b] >=1, i=1,2 ..., l (6)
Step S205: the sample concentrated based on the training dataset obtained after the normalized and prediction data and institute
Constraint condition is stated, determines the best configuration parameter of the best hyperplane.
In an embodiment of the present invention, the training dataset and predictive data set by will be obtained after the normalized
In sample substitute into the constraint condition, can find to obtain the best configuration parameter of the best hyperplane namely the tune of w, b
Excellent numerical value.
In an alternative embodiment of the invention, in order to further increase the classification performance of the best hyperplane, return to be described
Corresponding loose vector is respectively set in the sample that the training dataset and prediction data obtained after one change processing is concentrated, it may be assumed that
ξi>=0, i=1,2 ..., l (7)
Wherein, the sample x concentrated based on the training dataset obtained after normalized and prediction dataiRelaxation vector
ξiIt indicates.
Then, the training data obtained after Slope Parameters and the normalized about the best hyperplane is constructed
The objective function of the relaxation vector for the sample that collection and prediction data are concentrated, it may be assumed that
Wherein, C indicates preset nonnegative constant, ξiIndicate the training dataset obtained after the normalized and prediction
The relaxation vector of the i-th sample in data set.
In above-mentioned formula (8), above formulaSo that the training dataset obtained after normalized and prediction number
Should be big as far as possible according to the distance of sample to the best hyperplane of concentration, to improve generalization ability,Then make most preferably super
The error in classification of plane is small as far as possible.It in other words, is the problem of determining the best configuration parameter of the best hyperplane, to become
Under the premise of formula (6) and (7), the problems of value of formula (8) is minimized.
Then, the minimum value of formula (8) is sought.It in an embodiment of the present invention, will by using method of Lagrange multipliers
Formula (8) is converted to its dual function:
And:
λi>=0, λj≥0 (10)
Wherein, L (λ, w, b) indicates that the dual function of the objective function, λ indicate Lagrange multiplier, λiIt indicates
The Lagrange multiplier for i-th of sample that the training dataset and prediction data obtained after the normalized is concentrated, λiTable
Show the Lagrange for j-th of sample that the training dataset for indicating to obtain after the normalized and prediction data are concentrated
Multiplier.
Later, it is described right the sample that the training dataset obtained after normalized and prediction data are concentrated to be substituted into respectively
Even function seeks the maximum value of the dual function in formula (9) to get the minimum value of the objective function arrived in formula (8).
Finally, by the striked corresponding Slope Parameters value of minimum value, slope as the best hyperplane it is best
With setting value, and the sample concentrated with the training dataset and prediction data that will be obtained after normalized substitutes into the constraint item
Part, so that two class samples are the sum of to best hyperplane minimum rangeMaximum, so that the plane for obtaining the best hyperplane is normal
Several best configuration numerical value.
The best configuration parameter: being substituted into the functional expression of the best hyperplane by step S206, obtains the calculating money
Source prediction model.
In specific implementation, by the best of the Slope Parameters w of the obtained best hyperplane and plane constant parameter b
With setting value, the functional expression of the best hyperplane is substituted into, can be obtained the training dataset and test data after normalization
The sample of concentration is correctly divided into the best hyperplane of two classes.In other words, having obtained can be by calculating task data to be predicted
The calculating task data locally calculated using mobile terminal are classified as still to be calculated using mobile edge calculations server
Calculating task data classifier.
The above-mentioned mobile computing resource allocation methods in the embodiment of the present invention are described, below will be to above-mentioned side
The corresponding device of method is introduced.
Fig. 3 shows the structure of the autonomous distributor of one of embodiment of the present invention mobile computing resource.Referring to Fig. 3,
One of embodiment of the present invention mobile computing resource allocation device 30 may include analytical unit 301,302 and of taxon
Decision package 303, in which:
The analytical unit 301, suitable for the computation requests that send of received mobile terminal analyze, corresponded to
Calculating task data to be predicted；
The taxon 302 is suitable for obtained calculating task data to be predicted inputting trained calculating in advance
Resources model obtains the classification results of the calculating task data to be predicted；
The decision package 303 suitable for making decisions to obtained classification results, and uses and the court verdict pair
The corresponding calculating task of computation requests described in the mobile terminal answered or mobile edge calculations server process.In the present invention
In one embodiment, the decision package 303, suitable for being made decisions using following formula to obtained classification results:Wherein, f (x) indicates that, to the court verdict for the unknown vector x for belonging to class, sgn () is indicated
Return function, and: whenNumerical value .f (x) at the top of support vector machines value be 1；WhenNumerical value .f (x) at the lower section of support vector machines value be -1.
In an embodiment of the present invention, the autonomous distributor 30 of mobile computing resource can also include model training unit
304, in which:
The model training unit 304, suitable for using history computational resource allocation data construct respectively training dataset and
Predictive data set；Constructed training dataset and predictive data set are normalized respectively；Construct best hyperplane
Functional expression so that the sample that the best hyperplane concentrates the training dataset and prediction data that obtain after normalized
It is divided into two classes；The sample concentrated based on the training dataset obtained after normalized and prediction data is to the best hyperplane
Distance minimum value, the constraint condition of the best hyperplane is set；Based on the training number obtained after the normalized
The sample concentrated according to collection and prediction data and the constraint condition, determine the best configuration parameter of the best hyperplane；By institute
The functional expression that best configuration parameter substitutes into the best hyperplane is stated, the computing resource prediction model is obtained.
In an embodiment of the present invention, the model training unit 304, suitable for the instruction that will be obtained after the normalized
The sample for practicing data set and prediction data concentration substitutes into the constraint condition, obtains the best configuration ginseng of the best hyperplane
Number.Wherein, the constraint condition are as follows: yi[(w·xi)+b] >=1, i=1,2 ..., l；Wherein, yiIndicate the normalized
The feature tag for the i-th sample that the training dataset and prediction data obtained afterwards is concentrated, xiIt is obtained after indicating the normalized
The i-th sample that the training dataset and prediction data arrived is concentrated, wxi+ b indicates that the best hyperplane, w indicate described best
The Slope Parameters of hyperplane, b indicate that the plane constant parameter of the best hyperplane, l indicate the instruction obtained after normalized
Practice the total sample number of data set and prediction data concentration.
In still another embodiment of the process, the model training unit 304 is suitable for obtaining after the normalized
Corresponding loose vector is respectively set in the sample that training dataset and prediction data are concentrated；Based on the instruction obtained after normalized
The distance between the sample that white silk data set and prediction data are concentrated and the best hyperplane are maximized and are minimized the error, and are constructed
It is concentrated about the training dataset and prediction data obtained after the Slope Parameters and the normalized of the best hyperplane
Sample relaxation vector objective function；The dual function of the objective function is sought using method of Lagrange multipliers, and will
The sample that the training dataset and prediction data obtained after normalized is concentrated substitutes into the dual function respectively, seeks described
The maximum value of dual function, the minimum value as the objective function；By the striked corresponding Slope Parameters value of minimum value, make
For the best configuration numerical value of the slope of the best hyperplane；By the best configuration parameter of the slope of the best hyperplane and incite somebody to action
The sample that the training dataset and prediction data obtained after normalized is concentrated substitutes into the constraint condition, obtains described best
The best configuration numerical value of the plane constant of hyperplane.Wherein, the best hyperplane constructed by the model training unit
The mesh of the relaxation vector for the sample that the training dataset and prediction data obtained after Slope Parameters and the normalized is concentrated
Scalar functions are as follows:Wherein, C indicates preset nonnegative constant, ξiIt indicates to obtain after the normalized
The relaxation vector for the i-th sample that training dataset and prediction data are concentrated；The dual function of the objective function are as follows:And:Wherein, L (λ, w, b)
Indicate that the dual function of the objective function, λ indicate Lagrange multiplier, λiIt is obtained after the expression expression normalized
The Lagrange multiplier for i-th of sample that training dataset and prediction data are concentrated, λiIt indicates to indicate at the normalization
The Lagrange multiplier for j-th of sample that the training dataset and prediction data obtained after reason is concentrated.
The embodiment of the invention also provides a kind of computer readable storage mediums, are stored thereon with computer instruction, described
The step of mobile computing resource allocation methods are executed when computer instruction is run.Wherein, the mobile computing resource
Distribution method refers to being discussed in detail for preceding sections, repeats no more.
The embodiment of the invention also provides a kind of terminal, including memory and processor, energy is stored on the memory
Enough computer instructions run on the processor, the processor execute the movement when running the computer instruction
The step of computational resource allocation method.Wherein, the mobile computing resource allocation methods refer to detailed Jie of preceding sections
It continues, repeats no more.
It is preparatory by inputting obtained calculating task data to be predicted using the above scheme in the embodiment of the present invention
Trained computing resource prediction model obtains the classification results of the calculating task data to be predicted, and to obtained point
Class result makes decisions, so as to using the mobile terminal corresponding with the court verdict or mobile edge calculations clothes
The corresponding calculating task of the device processing computation requests of being engaged in, can be by computing resource in the mobile terminal and mobile edge service
Reasonable distribution is carried out between device, submits to mobile edge calculations server process with by the calculating task of all mobile terminals
It compares, so as to improve the treatment effeciency of mobile computing task, improves the utilization rate of resource.
Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of above-described embodiment is can
It is completed with instructing relevant hardware by program, which can store in computer readable storage medium, and storage is situated between
Matter may include: ROM, RAM, disk or CD etc..
Although present disclosure is as above, present invention is not limited to this.Anyone skilled in the art are not departing from this
It in the spirit and scope of invention, can make various changes or modifications, therefore protection scope of the present invention should be with claim institute
Subject to the range of restriction.
Claims (10)
1. a kind of mobile computing resource allocation methods characterized by comprising
To the computation requests that send of received mobile terminal analyze, obtain corresponding calculating task data to be predicted；
Obtained calculating task data to be predicted are inputted into trained computing resource prediction model in advance, are obtained described to pre-
Survey the classification results of calculating task data；
Obtained classification results are made decisions, and use the mobile terminal corresponding with the court verdict or movement
The corresponding calculating task of computation requests described in edge calculations server process.
2. mobile computing resource allocation methods according to claim 1, which is characterized in that the computing resource prediction model
Training obtains by the way of following:
Training dataset and predictive data set are constructed respectively using history computational resource allocation data；
Constructed training dataset and predictive data set are normalized respectively；
Construct the functional expression of best hyperplane so that the best hyperplane by the training dataset obtained after normalized and
The sample that prediction data is concentrated is divided into two classes；
The sample concentrated based on the training dataset obtained after normalized and prediction data to the best hyperplane away from
From minimum value, the constraint condition of the best hyperplane is set；
The sample concentrated based on the training dataset obtained after the normalized and prediction data and the constraint condition, really
The best configuration parameter of the fixed best hyperplane；
The functional expression that the best configuration parameter is substituted into the best hyperplane obtains the computing resource prediction model.
3. mobile computing resource allocation methods according to claim 2, which is characterized in that described based at the normalization
The sample and the constraint condition that the training dataset and prediction data obtained after reason is concentrated, determine the best hyperplane most
Good configuration parameter, comprising:
The sample that the training dataset and prediction data that obtain after the normalized are concentrated substitutes into the constraint condition, obtains
To the best configuration parameter of the best hyperplane.
4. mobile computing resource allocation methods according to claim 3, which is characterized in that the constraint condition are as follows:
yi[(w·xi)+b] >=1, i=1,2 ..., l；
Wherein, yiIndicate the feature mark for the i-th sample that the training dataset obtained after the normalized and prediction data are concentrated
Label, xiIndicate the training dataset obtained after the normalized and the i-th sample that prediction data is concentrated, wxi+ b indicates institute
Best hyperplane is stated, w indicates the Slope Parameters of the best hyperplane, and b indicates the plane constant parameter of the best hyperplane,
L indicates the training dataset obtained after normalized and the total sample number that prediction data is concentrated.
5. mobile computing resource allocation methods according to claim 2, which is characterized in that described based at the normalization
The sample and the constraint condition that the training dataset and prediction data obtained after reason is concentrated, determine the best hyperplane most
Good configuration parameter, comprising:
Corresponding relaxation is respectively set in the sample concentrated for the training dataset and prediction data that obtain after the normalized
Vector；
Between the sample and the best hyperplane concentrated based on the training dataset obtained after normalized and prediction data
Distance maximize and minimize the error, construct about the best hyperplane Slope Parameters and the normalized after
The objective function of the relaxation vector for the sample that the training dataset and prediction data arrived is concentrated；
The dual function of the objective function, and the training number that will be obtained after normalized are sought using method of Lagrange multipliers
The dual function is substituted into respectively according to the sample that collection and prediction data are concentrated, and the maximum value of the dual function is sought, as institute
State the minimum value of objective function；
By the striked corresponding Slope Parameters value of minimum value, the best configuration numerical value of the slope as the best hyperplane；
By the best configuration parameter of the slope of the best hyperplane and by the training dataset obtained after normalized and in advance
The sample that measured data is concentrated substitutes into the constraint condition, obtains the best configuration numerical value of the plane constant of the best hyperplane.
6. mobile computing resource allocation methods according to claim 5, which is characterized in that constructed is described best super flat
The relaxation vector for the sample that the training dataset and prediction data obtained after the Slope Parameters in face and the normalized is concentrated
Objective function are as follows:
Wherein, C indicates preset nonnegative constant, ξiIndicate the training dataset obtained after the normalized and prediction data
The relaxation vector for i-th of the sample concentrated.
7. mobile computing resource allocation methods according to claim 6, which is characterized in that the antithesis letter of the objective function
Number are as follows:
And:
λi>=0, λj≥0；
Wherein, L (λ, w, b) indicates that the dual function of the objective function, λ indicate Lagrange multiplier, λiReturn described in indicating
The Lagrange multiplier for i-th of sample that the training dataset and prediction data obtained after one change processing is concentrated, λiIt indicates
Indicate the Lagrange multiplier for j-th of sample that the training dataset obtained after the normalized and prediction data are concentrated.
8. mobile computing resource allocation methods according to claim 7, which is characterized in that using following formula to gained
To classification results make decisions:
Wherein, f (x) indicates that, to the court verdict for the unknown vector x for belonging to class, sgn () indicates return function, and: whenNumerical value f (x) at the top of support vector machines value be 1；WhenNumerical value exist
The value of f (x) is -1 when the lower section of support vector machines.
9. a kind of computer readable storage medium, is stored thereon with computer instruction, which is characterized in that the computer instruction fortune
Perform claim requires the step of 1 to 8 described in any item mobile computing resource allocation methods when row.
10. a kind of terminal, which is characterized in that including memory and processor, storing on the memory can be at the place
The computer instruction run on reason device, perform claim requires any one of 1 to 8 institute when the processor runs the computer instruction
The step of mobile computing resource allocation methods stated.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810933422.2A CN109343951B (en) | 2018-08-15 | 2018-08-15 | Mobile computing resource allocation method, computer-readable storage medium and terminal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810933422.2A CN109343951B (en) | 2018-08-15 | 2018-08-15 | Mobile computing resource allocation method, computer-readable storage medium and terminal |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109343951A true CN109343951A (en) | 2019-02-15 |
| CN109343951B CN109343951B (en) | 2022-02-11 |
Family
ID=65291591
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810933422.2A Active CN109343951B (en) | 2018-08-15 | 2018-08-15 | Mobile computing resource allocation method, computer-readable storage medium and terminal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109343951B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111400007A (en) * | 2020-03-13 | 2020-07-10 | 重庆特斯联智慧科技股份有限公司 | Task scheduling method and system based on edge calculation |
| CN112433852A (en) * | 2020-11-23 | 2021-03-02 | 广州技象科技有限公司 | Internet of things edge calculation control method, device, equipment and storage medium |
| CN113610303A (en) * | 2021-08-09 | 2021-11-05 | 北京邮电大学 | Load prediction method and system |
| CN114513506A (en) * | 2020-11-17 | 2022-05-17 | 中国联合网络通信集团有限公司 | Service processing method, access edge cloud server and service processing system |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104753718A (en) * | 2015-02-25 | 2015-07-01 | 重庆大学 | Deficiency service throughput rate complementing method and device based on non-negative polymerization |
| GB2535613A (en) * | 2014-12-19 | 2016-08-24 | Apical Ltd | Sensor noise profile |
| CN106534333A (en) * | 2016-11-30 | 2017-03-22 | 北京邮电大学 | Bidirectional selection computing unloading method based on MEC and MCC |
| CN106709820A (en) * | 2017-01-11 | 2017-05-24 | 中国南方电网有限责任公司电网技术研究中心 | Electrical power system load prediction method and device based on depth belief network |
| CN107231384A (en) * | 2017-08-10 | 2017-10-03 | 北京科技大学 | A kind of ddos attack detection defence method cut into slices towards 5g networks and system |
| CN107770263A (en) * | 2017-10-16 | 2018-03-06 | 电子科技大学 | A kind of internet-of-things terminal safety access method and system based on edge calculations |
| CN107766868A (en) * | 2016-08-15 | 2018-03-06 | 中国联合网络通信集团有限公司 | A kind of classifier training method and device |
| CN107846704A (en) * | 2017-10-26 | 2018-03-27 | 北京邮电大学 | A kind of resource allocation and base station service arrangement method based on mobile edge calculations |
| CN108011747A (en) * | 2017-10-30 | 2018-05-08 | 北京邮电大学 | Edge delamination social relationships cognitive method |
| CN108009089A (en) * | 2017-12-01 | 2018-05-08 | 中南大学 | A kind of increment machine learning method and system based on lucidification disposal |
| CN108304877A (en) * | 2018-02-02 | 2018-07-20 | 电子科技大学 | A kind of physical layer channel authentication method based on machine learning |
-
2018
- 2018-08-15 CN CN201810933422.2A patent/CN109343951B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2535613A (en) * | 2014-12-19 | 2016-08-24 | Apical Ltd | Sensor noise profile |
| CN104753718A (en) * | 2015-02-25 | 2015-07-01 | 重庆大学 | Deficiency service throughput rate complementing method and device based on non-negative polymerization |
| CN107766868A (en) * | 2016-08-15 | 2018-03-06 | 中国联合网络通信集团有限公司 | A kind of classifier training method and device |
| CN106534333A (en) * | 2016-11-30 | 2017-03-22 | 北京邮电大学 | Bidirectional selection computing unloading method based on MEC and MCC |
| CN106709820A (en) * | 2017-01-11 | 2017-05-24 | 中国南方电网有限责任公司电网技术研究中心 | Electrical power system load prediction method and device based on depth belief network |
| CN107231384A (en) * | 2017-08-10 | 2017-10-03 | 北京科技大学 | A kind of ddos attack detection defence method cut into slices towards 5g networks and system |
| CN107770263A (en) * | 2017-10-16 | 2018-03-06 | 电子科技大学 | A kind of internet-of-things terminal safety access method and system based on edge calculations |
| CN107846704A (en) * | 2017-10-26 | 2018-03-27 | 北京邮电大学 | A kind of resource allocation and base station service arrangement method based on mobile edge calculations |
| CN108011747A (en) * | 2017-10-30 | 2018-05-08 | 北京邮电大学 | Edge delamination social relationships cognitive method |
| CN108009089A (en) * | 2017-12-01 | 2018-05-08 | 中南大学 | A kind of increment machine learning method and system based on lucidification disposal |
| CN108304877A (en) * | 2018-02-02 | 2018-07-20 | 电子科技大学 | A kind of physical layer channel authentication method based on machine learning |
Non-Patent Citations (3)
| Title |
|---|
| EDITORIAL TEAM: ""Taking Machine Leaning to the Edge - insideBIGDATA"", 《HTTPS://INSIDEBIGDATA.COM/2017/09/25/TAKING-MACHINE-LEARNING-EDGE》 * |
| JIN QI: ""Collaborative Energy Management Optimization Toward a Green Energy Local Area Network"", 《IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS》 * |
| 于晓艺: ""空天地一体化网络接入网边缘计算的实时CPU调度算法"", 《第十四届卫星通信学术年会》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111400007A (en) * | 2020-03-13 | 2020-07-10 | 重庆特斯联智慧科技股份有限公司 | Task scheduling method and system based on edge calculation |
| CN114513506A (en) * | 2020-11-17 | 2022-05-17 | 中国联合网络通信集团有限公司 | Service processing method, access edge cloud server and service processing system |
| CN112433852A (en) * | 2020-11-23 | 2021-03-02 | 广州技象科技有限公司 | Internet of things edge calculation control method, device, equipment and storage medium |
| CN112433852B (en) * | 2020-11-23 | 2021-09-03 | 广州技象科技有限公司 | Internet of things edge calculation control method, device, equipment and storage medium |
| CN113610303A (en) * | 2021-08-09 | 2021-11-05 | 北京邮电大学 | Load prediction method and system |
| CN113610303B (en) * | 2021-08-09 | 2024-03-19 | 北京邮电大学 | Load prediction method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109343951B (en) | 2022-02-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11030484B2 (en) | System and method for efficient generation of machine-learning models | |
| CN109343951A (en) | Mobile computing resource allocation methods, computer readable storage medium and terminal | |
| CN106202431B (en) | A kind of Hadoop parameter automated tuning method and system based on machine learning | |
| US9111232B2 (en) | Portable workload performance prediction for the cloud | |
| CN110390345B (en) | Cloud platform-based big data cluster self-adaptive resource scheduling method | |
| Wan et al. | A selection method based on MAGDM with interval-valued intuitionistic fuzzy sets | |
| CN107220217A (en) | Characteristic coefficient training method and device that logic-based is returned | |
| CN110008259A (en) | The method and terminal device of visualized data analysis | |
| CN113037877B (en) | Optimization method for time-space data and resource scheduling under cloud edge architecture | |
| Zhao et al. | Toward SLA-constrained service composition: An approach based on a fuzzy linguistic preference model and an evolutionary algorithm | |
| Thonglek et al. | Improving resource utilization in data centers using an LSTM-based prediction model | |
| CN110097098A (en) | Data classification method and device, medium and electronic equipment based on base classifier | |
| Nadeem et al. | Optimizing execution time predictions of scientific workflow applications in the grid through evolutionary programming | |
| Vitali et al. | Learning a goal-oriented model for energy efficient adaptive applications in data centers | |
| CN109358962A (en) | The autonomous distributor of mobile computing resource | |
| Shi et al. | Location-aware and budget-constrained service brokering in multi-cloud via deep reinforcement learning | |
| CN113674087A (en) | Enterprise credit rating method, apparatus, electronic device and medium | |
| Wang et al. | Decomposition-based multi-objective evolutionary algorithm for virtual machine and task joint scheduling of cloud computing in data space | |
| CN113032367A (en) | Dynamic load scene-oriented cross-layer configuration parameter collaborative tuning method and system for big data system | |
| CN105740967A (en) | Manufacture cloud service execution time prediction method and manufacture cloud service execution time prediction device | |
| Liu et al. | An optimized speculative execution strategy based on local data prediction in a heterogeneous hadoop environment | |
| Zhuang et al. | Smart multi-tenant federated learning | |
| Abdelwahab et al. | Alleviating the sparsity problem of collaborative filtering using an efficient iterative clustered prediction technique | |
| Daraghmeh et al. | Incorporating Data Preparation and Clustering Techniques for Workload Segmentation in Large-Scale Cloud Data Centers | |
| Zhen et al. | Lean production and technological innovation in manufacturing industry based on SVM algorithms and data mining technology |
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 |