CN109543818A - A kind of link evaluation method and system based on deep learning model - Google Patents
A kind of link evaluation method and system based on deep learning model Download PDFInfo
- Publication number
- CN109543818A CN109543818A CN201811221654.1A CN201811221654A CN109543818A CN 109543818 A CN109543818 A CN 109543818A CN 201811221654 A CN201811221654 A CN 201811221654A CN 109543818 A CN109543818 A CN 109543818A
- Authority
- CN
- China
- Prior art keywords
- link
- deep learning
- model
- learning model
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Traffic Control Systems (AREA)
Abstract
The present invention relates to a kind of link evaluation methods based on deep learning model, comprising: setting deep learning model structure；The degree data of the data on flows and the link that acquire link in network are data pair；To all data to being divided to establish training set and test set；The deep learning model is trained with the training set, determines the parameter of the deep learning model；The deep learning model is tested with the test set, if test accuracy is less than test threshold, adjusts the parameter and to the deep learning model re -training, it is on the contrary then using the deep learning model as link evaluation model；The current degree of the link is obtained, with the current traffic data of the link by the link evaluation model to assess the link.
Description
Technical field
The invention belongs to information technology fields, distribute field especially suitable for wireless network resource.
Background technique
With the rapid development of wireless communication technique, wireless telecom equipment is in explosive growth, and frequency spectrum resource becomes more next
It is more nervous, while being interfered with each other between communication equipment, so that original limited frequency spectrum resource utilization rate wretched insufficiency.Wireless network
Communication equipment between interfere with each other, will cause the loss of data, after loss of data generally use retransmission mechanism, though the mechanism
So data transmission success can be improved to a certain extent but also result in the waste of energy.If the upper-layer protocol of network is selecting
When routing diameter carries out transmission data select the preferable link of link-quality transmitted, then can improve data transfer rate, and also
Energy can be saved because reducing the re-transmission of data.Link-quality between node is periodically variable, that is, last moment link matter
The quality for measuring preferable link subsequent time not necessarily might as well.If being able to know that the link-quality of subsequent time, upper-layer protocol
The selection preferable link of subsequent time quality carries out data transmission, and can further submit data transmission success.In addition,
Under same wireless network, network is made of multiple data link, and the flow of the transmission of different communication equipment is different, on upper layer
Under the control of Data Transport Protocol, it often will appear link and occupy the very big situation of degree difference, it is most of that flow is by small
What part of links was transmitted.Therefore, it excavates network flow and data link occupies the non-linear relation of degree, realize network money
The balance dispatching distribution in source is the effective way for improving network spectrum resource utilization.
It is the common method of link-quality assessment based on Optimum Theory, but these methods are mostly heuritic approaches,
It is difficult to be suitable for continually changing network flow and link occupies the relation excavation of degree.Therefore, the present invention proposes that one kind can be with
Link-quality is assessed under heterogeneous networks flow, the adaptive calculation for excavating network flow and link occupancy Degree of Accord Relation
Method.This algorithm occupies degree to link using deep learning model and assesses, it can be from continually changing network flow
The abstract feature of extraction height, excavates network flow and link occupies degree directly potential relationship.
Summary of the invention
To solve the above problems, the invention discloses a kind of link evaluation methods based on deep learning model, comprising: set
Set deep learning model structure；Acquire link in network data on flows and the link degree data to constitute data pair,
And by all data to dividing to establish training set and test set；The deep learning model is instructed with the training set
Practice, determines the parameter of the deep learning model；The deep learning model is tested with the test set, if test accuracy is small
In test threshold, then adjust the parameter and to the deep learning model re -training, it is on the contrary then using the deep learning model as chain
Road assessment models；The current degree of the link is obtained with the data on flows of current ink by the link evaluation model.
Wherein the deep learning model is multilayer neural network structure, including input layer, multiple hidden layers and output layer,
The input layer includes N number of input neuron, which corresponds in N number of node of the link, each hidden layer
Including L calculating neuron, which meets y=φ (ω x+b), which includes M output neuron, this is defeated
Neuron is corresponded in M of the network links out, whereinφ is nonlinear function, and x is should
The input vector of neuron is calculated, y is the output vector of the calculating neuron, and ω, b are the parameter of the deep learning model, ω
For weight vector, b is offset vector, and κ is correction constant, and L, M, N are positive integer, and 1≤κ≤10, φ are hyperbolic tangent function.
Link evaluation method of the present invention is trained the link evaluation model by back-propagation algorithm, and
As the loss function l of the training link evaluation modelsWhen less than loss threshold value, deconditioning simultaneously obtains ω and b, whereinFor the reality output that the data on flows of the data pair is obtained by the link evaluation model, yiFor the number
According to pair degree data.
Link evaluation method of the present invention is adjusted weight vector ω by damped manner, i.e.,Wherein ω ' is the adjusted value that weight vector ω is obtained in last round of adjustment, and η is that the link is trained to comment
Estimate the autoadapted learning rate of model.
Link evaluation method of the present invention acquires the data using Poisson traffic model or self similarity traffic model
It is right.
The invention further relates to a kind of link evaluation systems based on deep learning model, comprising:
Model construction module, for deep learning model structure to be arranged；
Data acquisition module, for acquiring the data on flows of link in network and the degree data of the link to constitute number
According to right, and by all data to dividing to establish training set and test set；
Model training module determines the deep learning mould for being trained with the training set to the deep learning model
The parameter of type；
Model measurement module, for being tested with the test set the deep learning model, if test accuracy is less than
Test threshold, then adjust the parameter and to the deep learning model re -training, it is on the contrary then using the deep learning model as link
Assessment models；
Link evaluation module, for obtaining working as the link by the link evaluation model with the data on flows of current ink
Preceding degree.
Link evaluation system of the present invention, the deep learning model are multilayer neural network structure, including input
Layer, multiple hidden layers and output layer, the input layer include N number of input neuron, which corresponds in the link
N number of node, each hidden layer includes L calculating neuron, which meets y=φ (ω x+b), the output layer
Including M output neuron, which is corresponded in M of the network links, whereinφ is nonlinear function, and x is the input vector of the calculating neuron, and y is the defeated of the calculating neuron
Vector out, ω are weight vector, and b is offset vector, and ω, b are the parameter, and κ is correction constant, and L, M, N are positive integer, 1≤κ≤
10。
Link evaluation system of the present invention, the model training module specifically include: by back-propagation algorithm pair
The link evaluation model is trained, wherein training the loss function of the link evaluation model to beWork as lsIt is small
When losing threshold value, deconditioning simultaneously obtains ω and b,It is obtained for the data on flows of the data pair by the link evaluation model
Reality output, yiFor the degree data of the data pair.
Link evaluation method of the present invention, the model measurement module further include: parameter adjustment module, for passing through
Damped manner is adjusted weight vector ω, whereinω ' is weight vector ω in last round of adjustment
The adjusted value of acquisition, η are the autoadapted learning rate of the training link evaluation model.
Detailed description of the invention
Fig. 1 is a kind of Deep Learning model knot of link evaluation method based on Deep Learning model of the embodiment of the present invention
Structure schematic diagram.
Fig. 2 is a kind of link evaluation method flow diagram based on Deep Learning model of the embodiment of the present invention.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing, the present invention is mentioned
A kind of link evaluation method method and system based on Deep Learning model out is further described.It should be appreciated that herein
Described specific implementation method is only used to explain the present invention, is not intended to limit the present invention.
In the wireless network, the relationship for excavating network flow and link occupancy degree is extremely complex, by link-quality, source
It is influenced with many factors such as the position of purpose and network interferences.The invention proposes a kind of link evaluation methods.The method is based on
Deep learning model analyzes link occupancy situation, occupies the hiding relationship of degree for excavating network flow and link.
In a wireless network, there are two types of the models of network flow: Poisson flow, Self-Similar Traffic.
Poisson traffic model is most widely used network flow descriptive model in current wireless network.In this model,
Data packet interarrival time t meets the exponential distribution of parameter lambda, and the mean value and variance of this model are all λ.Poisson traffic model
Shown in probability density function such as formula (1).
F (t)=λ e-λt (1)
Self similar processes are the complicated processes with self-similarity nature.Self-similarity nature is in internet traffic
Universal feature, it includes the long-rang dependence of flow, sudden and on different time scales correlation.The present invention borrows
Pareto distribution is helped to generate Self-Similar Traffic.Shown in the probability density function of self similarity traffic model such as formula (2).
Wherein, α, β are the shape and scale parameter of distribution, α > 0, β > 0 respectively.
Link occupancy situation for the method using deep learning to network under different flow model, is firstly introduced into net
Input the flow vector X, X of network are the vectors of a N*1, and each element in vector is xn, xnFor the flow number of each node
According to N is the total node number in network, and the expression formula of each entity is as shown in formula 3, wherein 0 < n≤N, n is positive integer.
Wherein d is the flow that node needs to input, and the size of d is by the density probability formula (1) of aforementioned flow model or (2)
It provides, i.e. d=f (t).
Fig. 1 is a kind of Deep Learning model knot of link evaluation method based on Deep Learning model of the embodiment of the present invention
Structure schematic diagram.As shown in Figure 1, deep learning model of the invention is described in detail below by multitiered network structure composition
The composed structure of deep learning model.Deep learning model is made of multilayer, specifically in an embodiment of the present invention, using 4 layers
Structure is input layer, hidden layer 1, hidden layer 2 and output layer respectively, and the present invention is not limited thereto.Other than input layer
What each layer was all made of multiple calculating neurons, calculating neuron is the calculator for having nonlinear function function.
Neuron passes through the input of the multiple Weights of linear combination, then obtains single output by certain nonlinear activation function.Individually
Shown in the operation of neuron such as formula (4).
Y=φ (ω x+b) (4)
Wherein x is the input vector for calculating neuron, and y is the output vector for calculating neuron, and ω is weight vector, and b is
Offset vector, φ are activation primitives.It is in network that input vector x, which is the length N of input flow vector X, input vector x, in the present invention
Number of nodes.Output vector y is the occupancy degree vector of link, and the length M of output vector y is the number of link in network.Chain
Road appraisal procedure includes training and two stages of test, and weight vector ω and offset vector b in deep learning model are to pass through
The training of model determines.
The training set of deep learning model is 100000 grabbed under different flow model by passing through Technology of Network Sniffer
Data occupy degree vector to its corresponding link including inlet flow vector sum to composition, each data.In the embodiment of the present invention
In, activation primitive φ uses tanh function, that is, hyperbolic tangent function, and it is nonlinear for exporting, as shown in formula (5).
Shown in the number L such as formula (6) of 2 neuron of hidden layer 1 and hidden layer.
Wherein N is input layer number, that is, nodes number, and M is output layer neuron number, that is, link in network
Number, κ are correction constant, κ ∈ [1,10].
In the training stage, the present invention uses cross entropy as loss function, as shown in formula (7).
WhereinFor the reality output of training stage, yiFor the desired output of training sample, the i.e. data pair for training
In degree data.
The present invention is trained depth model using back-propagation algorithm (BackPropagation, BP).BP algorithm sheet
It is to ask its objective function to reach the algorithm of minimum value by gradient descent method, such as formula (8) institute using cross entropy as objective function in matter
Show.First to each training sample set a desired output, then to mode input signal by input layer through hidden layer
To output Es-region propagations, the i.e. forward-propagating of signal.The weight of network is fixed and invariable in the transmittance process forward of signal, such as
Fruit cannot obtain desired output in output layer, then carry out the back-propagation process of error signal.Error is reality output and phase
It hopes the difference of output, is provided in the present invention by cross entropy.According to this minimum principle of cross entropy, from output layer toward middle layer successively to
Preceding propagation, the in the process connection weight of corrective networks, the i.e. inverse propagation of error signal.With working signal forward-propagating and
The alternate repetition of the inverse communication process of error signal carries out, and the reality output of network is gradually approached to corresponding desired output, net
Network constantly rises the accuracy that training sample exports.
Wherein ω ' is the adjusted value that weight vector ω is obtained in last round of adjustment, and η is autoadapted learning rate.
In the training stage, the condition of present invention training cut-off is loss function ls≤ 0.01, factor of momentum 0.09, study
Rate η is autoadapted learning rate, as shown in formula (9).
giIt is the partial derivative in the m times iteration for determining weight parameter ω using gradient descent methodIt is possible thereby to see
Increasing with iteration decline number m out, learning rate η is gradually reduced.
In test phase, 5000 test sample data pair are grabbed under different flow model by Technology of Network Sniffer, are surveyed
Trained deep learning model is tried, checks whether deep learning model over-fitting occurs and test accuracy lower than 90%, such as
There is over-fitting in fruit, then re-starts training to model using the method for weight decaying.Weight decaying refers to as gradient declines
Number constantly reduce the ratio of weight, it is specific as shown in formula (10).
If deep learning model, which does not occur over-fitting, tests accuracy not less than 90%, to complete the depth of test
Habit model is link evaluation model.
Fig. 2 is a kind of link evaluation method flow diagram based on Deep Learning model of the embodiment of the present invention.Such as Fig. 2 institute
Show, main flow of the invention includes the following steps:
Step S1 determines deep learning model structure；
Step S2 acquires the data on flows of link in network and the degree of respective links, by data on flows and degree
Constitute data pair；
Step S3, by data to being divided into training set and test set；
Step S4, using training set, with loss function lsIt is stop condition less than 0.01, using back-propagation algorithm to depth
It spends learning model and carries out repetition training, determine the weight vector and offset vector of deep learning model；
Step S5 tests trained deep learning model using test set；
Step S6, judge deep learning model whether over-fitting, i.e., if test accuracy be greater than 90%, it is excessively not quasi-
It closes, completes test, turn to step S8, otherwise turn to step S7；
Step S7, decays to weight vector, and turns to step 4；
Step S8, by the deep learning model for completing to test as link evaluation model, with the current traffic data of link
Corresponding current degree is obtained by link evaluation model, to complete to assess the degree of link.
The present invention has the advantage that
The concurrency of height, deep learning is formed by connecting by many identical simple process unit the parallel combineds, although often
The function of a processing unit is simple, but the concurrent activities of the simple unit of a large amount of functions, but has very powerful processing capacity,
This enables deep learning model to analyze the link occupancy situation under complex network flow；
The nonlinear interaction of height, each neuron in deep learning model can receive a large amount of other neurons
Input, and be non-linear relation between the input and output of each neuron.This between neuron conditions each other and mutually
The relationship mutually influenced, realizes whole network from the Nonlinear Mapping for being input to output state space, therefore it can be with any essence
Degree approaches arbitrary continuation nonlinear function；
Good error resilience performance and function of associate memory, deep learning model can be realized pair by the network structure of itself
The memory of information, the information remembered are stored in the connection weight between neuron.It does not see and being stored from single weight
The information content so that neural network has good fault-tolerance, and can be carried out cluster but due to distributed storage mode
Isotype information processing work is restored in analysis, feature extraction, memory.The good error resilience performance of deep learning model, which is also manifested by, works as
When the case where certain neurons failure occur in the hardware or software of system, entire model still can work on.Deep learning mould
The function of associate memory of type makes it have very strong uncertain information processing capacity.Even if inputting INFORMATION OF INCOMPLETE, inaccuracy
Or it is smudgy, deep learning model still be able to associative thinking be present in memory in things complete graph as.As long as input
Mode close to training sample, system can provide logic conclusion；
Adaptively, autonomous learning function, deep learning model obtain the weight and structure of network by training and study,
Very strong independent learning ability and the adaptability to environment are showed, can be very good the variation for adapting to Model of network traffic.
Claims (10)
1. a kind of link evaluation method based on deep learning model characterized by comprising
Deep learning model structure is set；
The data on flows of link in network and the degree data of the link are acquired to constitute data pair, and by all data pair
It is divided to establish training set and test set；
The deep learning model is trained with the training set, determines the parameter of the deep learning model；
The deep learning model is tested with the test set, if test accuracy is less than test threshold, adjusts the parameter
And to the deep learning model re -training, it is on the contrary then using the deep learning model as link evaluation model；
The current degree of the link is obtained with the current traffic data of the link by the link evaluation model.
2. link evaluation method as described in claim 1, which is characterized in that the deep learning model is multilayer neural network
Structure, including input layer, multiple hidden layers and output layer, the input layer include N number of input neuron, and the input neuron is one by one
Corresponding to N number of node of the link, each hidden layer includes L calculating neuron, which meets y=φ (ω x
+ b), which includes M output neuron, which corresponds in M of the network links, whereinφ is nonlinear function, and x is the input vector of the calculating neuron, and y is the defeated of the calculating neuron
Vector out, ω, b are the parameter of the deep learning model, and ω is weight vector, and b is offset vector, and κ is correction constant, L, M, N
For positive integer, 1≤κ≤10.
3. link evaluation method as claimed in claim 2, which is characterized in that φ is hyperbolic tangent function.
4. link evaluation method as claimed in claim 2 or claim 3, which is characterized in that commented by back-propagation algorithm the link
Estimate model to be trained, and as the loss function l of the training link evaluation modelsWhen less than loss threshold value, deconditioning is simultaneously obtained
Weight vector ω and offset vector b are taken, wherein Pass through the link evaluation for the data on flows of the data pair
The reality output that model obtains, yiFor the degree data of the data pair.
5. link evaluation method as claimed in claim 4, which is characterized in that adjusted by damped manner to weight vector ω
It is whole, i.e.,Wherein ω ' is the adjusted value that weight vector ω is obtained in last round of adjustment, and η is that training should
The autoadapted learning rate of link evaluation model.
6. link evaluation method as described in claim 1, which is characterized in that use Poisson traffic model or Self-Similar Traffic mould
Type acquires the data pair.
7. a kind of link evaluation system based on deep learning model, for assessing the pass between network flow and link degree
System characterized by comprising
Model construction module, for deep learning model structure to be arranged；
Data acquisition module, for acquiring the data on flows of link in network and the degree data of the link to constitute data
It is right, and by all data to dividing to establish training set and test set；
Model training module determines the deep learning model for being trained with the training set to the deep learning model
Parameter；
Model measurement module, for being tested with the test set the deep learning model, if test accuracy is less than test
Threshold value, then adjust the parameter and to the deep learning model re -training, it is on the contrary then using the deep learning model as link evaluation
Model；
Link evaluation module, for obtaining currently accounting for for the link by the link evaluation model with the data on flows of current ink
Expenditure.
8. link evaluation system as claimed in claim 7, which is characterized in that the deep learning model is multilayer neural network
Structure, including input layer, multiple hidden layers and output layer, the input layer include N number of input neuron, and the input neuron is one by one
Corresponding to N number of node of the link, each hidden layer includes L calculating neuron, which meets y=φ (ω x
+ b), which includes M output neuron, which corresponds in M of the network links, whereinφ is nonlinear function, and x is the input vector of the calculating neuron, and y is the defeated of the calculating neuron
Vector out, ω, b are the parameter of the deep learning model, and ω is weight vector, and b is offset vector, and κ is correction constant, L, M, N
For positive integer, 1≤κ≤10.
9. link evaluation system as claimed in claim 8, which is characterized in that the model training module specifically includes: passing through
Back-propagation algorithm is trained the link evaluation model, and as the loss function l of the training link evaluation modelsLess than damage
When losing threshold value, deconditioning simultaneously obtains weight vector ω and offset vector b, wherein For the data pair
The reality output that data on flows is obtained by the link evaluation model, yiFor the degree data of the data pair.
10. link evaluation method as claimed in claim 9, which is characterized in that the model measurement module further include: parameter tune
Mould preparation block, for being adjusted by damped manner to weight vector ω, whereinω ' is weight vector
The adjusted value that ω is obtained in last round of adjustment, η are the autoadapted learning rate of the training link evaluation model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811221654.1A CN109543818A (en) | 2018-10-19 | 2018-10-19 | A kind of link evaluation method and system based on deep learning model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811221654.1A CN109543818A (en) | 2018-10-19 | 2018-10-19 | A kind of link evaluation method and system based on deep learning model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109543818A true CN109543818A (en) | 2019-03-29 |
Family
ID=65844037
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811221654.1A Pending CN109543818A (en) | 2018-10-19 | 2018-10-19 | A kind of link evaluation method and system based on deep learning model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109543818A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110336754A (en) * | 2019-05-09 | 2019-10-15 | 北京邮电大学 | A kind of network flow configuration method and device |
| CN110598737A (en) * | 2019-08-06 | 2019-12-20 | 深圳大学 | Online learning method, device, equipment and medium of deep learning model |
| CN111783930A (en) * | 2019-04-03 | 2020-10-16 | 南京大学 | Neural network test sufficiency evaluation method based on path state |
| CN111800301A (en) * | 2020-08-20 | 2020-10-20 | 浙江璟锐科技有限公司 | Network security evaluation method and system in machine type communication |
| CN112019381A (en) * | 2020-08-12 | 2020-12-01 | 苏州浪潮智能科技有限公司 | Cluster link detection method and system based on deep learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107273971A (en) * | 2017-06-13 | 2017-10-20 | 石家庄铁道大学 | Architecture of Feed-forward Neural Network self-organizing method based on neuron conspicuousness |
| EP3264722A1 (en) * | 2014-12-26 | 2018-01-03 | System73, Inc. | Method and system for adaptive virtual broadcasting of digital content |
| CN108075974A (en) * | 2016-11-14 | 2018-05-25 | 中国移动通信有限公司研究院 | A kind of flow transmission control method, device and SDN architecture systems |
-
2018
- 2018-10-19 CN CN201811221654.1A patent/CN109543818A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3264722A1 (en) * | 2014-12-26 | 2018-01-03 | System73, Inc. | Method and system for adaptive virtual broadcasting of digital content |
| CN108075974A (en) * | 2016-11-14 | 2018-05-25 | 中国移动通信有限公司研究院 | A kind of flow transmission control method, device and SDN architecture systems |
| CN107273971A (en) * | 2017-06-13 | 2017-10-20 | 石家庄铁道大学 | Architecture of Feed-forward Neural Network self-organizing method based on neuron conspicuousness |
Non-Patent Citations (3)
| Title |
|---|
| 张鹏: "《基于Java 3D的虚拟现实技术研究与实践》", 31 August 2017, 苏州大学出版社 * |
| 文笃石: "基于数据仓库的客户挽留系统", 《微型机与应用》 * |
| 石爱中: "《国家审计信息化与国家治理》", 31 December 2012, 北京：中国时代经济出版社 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111783930A (en) * | 2019-04-03 | 2020-10-16 | 南京大学 | Neural network test sufficiency evaluation method based on path state |
| CN111783930B (en) * | 2019-04-03 | 2024-03-12 | 南京大学 | Neural network test sufficiency evaluation method based on path state |
| CN110336754A (en) * | 2019-05-09 | 2019-10-15 | 北京邮电大学 | A kind of network flow configuration method and device |
| CN110336754B (en) * | 2019-05-09 | 2020-04-21 | 北京邮电大学 | Network traffic configuration method and device |
| CN110598737A (en) * | 2019-08-06 | 2019-12-20 | 深圳大学 | Online learning method, device, equipment and medium of deep learning model |
| CN112019381A (en) * | 2020-08-12 | 2020-12-01 | 苏州浪潮智能科技有限公司 | Cluster link detection method and system based on deep learning |
| CN111800301A (en) * | 2020-08-20 | 2020-10-20 | 浙江璟锐科技有限公司 | Network security evaluation method and system in machine type communication |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109543818A (en) | A kind of link evaluation method and system based on deep learning model | |
| WO2021036414A1 (en) | Co-channel interference prediction method for satellite-to-ground downlink under low earth orbit satellite constellation | |
| CN106570597A (en) | Content popularity prediction method based on depth learning under SDN architecture | |
| CN109547431A (en) | A kind of network security situation evaluating method based on CS and improved BP | |
| US20220101167A1 (en) | Hybrid Quantum Computation Architecture for Solving Quadratic Unconstrained Binary Optimization Problems | |
| CN108460461A (en) | Mars earth shear parameters prediction technique based on GA-BP neural networks | |
| CN103105246A (en) | Greenhouse environment forecasting feedback method of back propagation (BP) neural network based on improvement of genetic algorithm | |
| CN109635763B (en) | Crowd density estimation method | |
| CN106650022A (en) | Method for predicting fault of complex electronic device | |
| CN115051929B (en) | Network fault prediction method and device based on self-supervision target perception neural network | |
| CN107295537A (en) | A kind of method and system for wireless sensor network reliability of testing and assessing | |
| CN110321493A (en) | A kind of abnormality detection of social networks and optimization method, system and computer equipment | |
| CN110135482B (en) | Network topology inference method and system based on convolutional neural network | |
| CN104092503A (en) | Artificial neural network spectrum sensing method based on wolf pack optimization | |
| CN110533109A (en) | A kind of storage spraying production monitoring data and characteristic analysis method and its device | |
| CN107038450A (en) | Unmanned plane policing system based on deep learning | |
| Reddy et al. | An optimal neural network model for software effort estimation | |
| El-Shafie et al. | Generalized versus non-generalized neural network model for multi-lead inflow forecasting at Aswan High Dam | |
| CN115526308A (en) | Time sequence prediction method based on width pruning echo state network | |
| Gudmundsdottir et al. | Inferring interwell connectivity in fractured geothermal reservoirs using neural networks | |
| CN106101035A (en) | The signal blind checking method of unity-feedback neutral network | |
| Balabanov et al. | Multilayer Perceptron Training Randomized by Second Instance of Multilayer Perceptron | |
| Chai et al. | Production Forecast of Coalbed Methane Based on GA Optimized BP Neural Network | |
| Shu et al. | Link prediction based on 3D convolutional neural network | |
| CN113556266B (en) | Traffic matrix prediction method taking traffic engineering as center |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190329 |
|
| RJ01 | Rejection of invention patent application after publication |