CN108770010A - A kind of service-oriented wireless network networking model intelligent reconstruction method - Google Patents
A kind of service-oriented wireless network networking model intelligent reconstruction method Download PDFInfo
- Publication number
- CN108770010A CN108770010A CN201810703460.9A CN201810703460A CN108770010A CN 108770010 A CN108770010 A CN 108770010A CN 201810703460 A CN201810703460 A CN 201810703460A CN 108770010 A CN108770010 A CN 108770010A
- Authority
- CN
- China
- Prior art keywords
- value
- networking
- networking model
- vector turns
- 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.)
- Granted
Links
- 230000006855 networking Effects 0.000 title claims abstract description 103
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000012549 training Methods 0.000 claims abstract description 44
- 230000005540 biological transmission Effects 0.000 claims abstract description 32
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 27
- 238000013528 artificial neural network Methods 0.000 claims abstract description 13
- 238000011156 evaluation Methods 0.000 claims description 26
- 238000012360 testing method Methods 0.000 claims description 23
- 230000008569 process Effects 0.000 claims description 12
- 238000010801 machine learning Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 4
- 230000033001 locomotion Effects 0.000 claims description 4
- 238000010606 normalization Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- 238000013139 quantization Methods 0.000 claims description 4
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000001514 detection method Methods 0.000 claims description 2
- 238000004088 simulation Methods 0.000 abstract description 3
- 210000004556 brain Anatomy 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- 101100172132 Mus musculus Eif3a gene Proteins 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000007726 management method Methods 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 241001269238 Data Species 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 210000004218 nerve net Anatomy 0.000 description 1
- 210000000653 nervous system Anatomy 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000003955 neuronal function Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000012559 user support system Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/06—Testing, supervising or monitoring using simulated traffic
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W74/00—Wireless channel access
- H04W74/02—Hybrid access
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L43/00—Arrangements for monitoring or testing data switching networks
- H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
- H04L43/0823—Errors, e.g. transmission errors
- H04L43/0829—Packet loss
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L43/00—Arrangements for monitoring or testing data switching networks
- H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
- H04L43/0852—Delays
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L43/00—Arrangements for monitoring or testing data switching networks
- H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
- H04L43/0876—Network utilisation, e.g. volume of load or congestion level
- H04L43/0888—Throughput
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Data Exchanges In Wide-Area Networks (AREA)
Abstract
The invention discloses a kind of service-oriented wireless network networking model intelligent reconstruction methods.This method user oriented demand for services carries out intelligent reconstruction by Learning Algorithm to user networking pattern.This method carries out vectorization expression by nine networking scale, portfolio, load capacity, transmission range, maximum hop count, bandwidth, mobility, chronesthesy and packet loss susceptibility attributes to user service first.Then, by changing user service attribute, emulation obtains the network performance parameter under different networking models.Network performance parameter under user service attribute and different networking models constitutes initial data.Initial data is quantified, empirical data set is formed.Later, by neural network algorithm, empirical data set is learnt, obtains training pattern.When user has new demand for services, this method is indicated according to the vectorization of demand for services, and training pattern is called to reconstruct the best networking model for meeting current service demand.The simulation results show validity of this method.
Description
Technical field
The invention belongs to field of wireless, more particularly to service-oriented wireless network networking model intelligent reconstruction side
Method.
Background technology
In modern wireless network, the demand for services of user is various and time-varying, a certain fixed networking model are difficult to even
It can not support changeable demand for services.Thus, it is badly in need of exploring under certain network condition, according to users service needs to user group
The method that net pattern carries out intelligent reconstruction.Pattern and the network parameter setting of networking are directly facing current demand for services, to
By the reconstruct of network structure and function, current users service needs are supported to the greatest extent.The judgement of reconstruction result quality
Foundation is whether the networking mode of reconstruct meets the demand for services of active user.It is changeable due to network condition and demand for services
Property, restructuring procedure allows for " improving the result of reconstruct using previous experience ", i.e. the intelligent reconstruction based on learning algorithm.
For the demand expansion research, by Learning Algorithm, energy is reconstructed towards new demand servicing for this method
Meet the best networking model of demand for services.In order to realize that the intelligent reconstruction of networking model, this method are serviced for the network user
The type characteristic of demand devises three kinds of networking models:Full-mesh dynamic Time Division networking model, multi-hop carrier wave detect networking model
With multi-hop ad hoc time division multiple acess networking model.
Full-mesh dynamic Time Division networking model (Dynamic Time Division Multiple Access, DTDMA),
The networking within the scope of full-mesh suitable for the network user supports network size variation, burst service transmission and full-mesh network to open up
Flutter control.As shown in Fig. 1, which, which synchronizes the time shaft of the whole network node, is divided into a series of continuous, random length networks
Time frame, each time frame include that node synchronization, time slot request, time slot distribution and data send four-stage.Node synchronous phase, entirely
Net node is completed distributed network and is synchronized, and generates the management node in current time frame.Node saves in the time slot request stage to management
Point sends itself time slot request information.Time slot allocated phase, management node are preferential according to the time slot request and business of the whole network node
Grade completes the unified distribution of time slot, and time slot allocation information is broadcasted remaining node in informing network.Each node knows itself
After the data slot being assigned to, the transmission of data grouping is completed in data transmission phase.Meanwhile the agreement can be propped up effectively
It holds new node to network and node logout, to meet the needs of network size dynamic expansion.
Multi-hop carrier wave detects networking model (Carrier Sense Multiple Access with Collision
Avoidance, CSMA/CA), there is high flexible, realize simple advantage, be suitable for user's networking within the scope of multi-hop, pass
Defeated business model is flexible, and bursty traffic is big.The pattern interacts the transmission/receiving node for meeting certain condition by control frame
It is transmitted based on the transmission definition that reservation is initiated, and is sent in main transmission and introduce one section between reservation process and Data frame transmission processes
Concurrent transmission gap.Main transmission receiving node uses the dynamic Adjusted Option based on exponential smoothing model to determine concurrent transmission gap
The length of period, it is main transmission CTS frame transmission ranges in remaining node in this gap according to certain rule attempt initiate or
Response sends reservation from transmission.After concurrent transmission gap, master/slave transmission initiates Data frame transmission processes simultaneously.Agreement uses
Concurrency conflict avoidance mechanism and ACK frames based on the estimation of tolerable jamming power acknowledgment strategy successively, ensure master/slave transmission
The transmitting of Data frames.
Multi-hop ad hoc time division multiple acess networking model (Enhanced Self-organized Time Division
Multiple Access, ESTDMA), it is suitable for user's networking within the scope of multi-hop transmission, transmission services pattern is fixed, and is happened suddenly
Portfolio is few.As shown in Fig. 2, the pattern is with self-organizing time division multiple protocol (Self-organized Time
Division Multiple Access, STDMA) based on, the time slot in STDMA agreements is divided into and distributes for the first time, is secondary
Distribution, data transmission and debit's response four-stage.Pass through RTR/ in the node of allocated phase for the first time, Yu Xianxuanding current time slots
CTR frames interaction reservation current time slots.In the secondary distribution stage, node uses time slot secondary distribution strategy competition-based.Node
By solving network throughput maximum value, it is informed in the optimal probability of secondary distribution stage contention reservation time slot, and with the probability
Allocated phase generates conflict or keeps idle time slot contention reservation for the first time.Success preengages the node of current time slots in data transmission
Stage transmission packet.After receiving node receives data packet, complete to pass to sending node response ACK frames in debit's acknowledgment phase
It is defeated.
Neural network be to the abstract of several fundamental characteristics of human brain and simulation, be it is a kind of by simple unit form being capable of mould
Quasi- biological nervous system makes real world objects the network structure of cross reaction.It is using the human brain operating mode of people as base
Plinth studies adaptive and nonprogrammed information processing method.The characteristics of this working mechanism, shows as measuring god by people in network
The processing function of own is embodied through the effect of member, from the structure of simulation human brain and single neuronal function, reaches mould
The purpose of anthropomorphic brain processing information.In general, neural network is made of input layer, hidden layer and output layer;Input layer corresponds to every
Several attributes of data;Hidden layer can be one or more layers, and every layer has several nodes;What output layer can be exported by several nodes
Different Results are classified.There are one threshold values for each node itself tool, and have weights between each node of next layer, lead to
It crosses and these parameters is finally determined to input sample data progress successive ignition, so as to predict new data.
Invention content
The purpose of the present invention is user oriented services, according to demand for services, between the networking model multiple users in network
Carry out intelligent reconstruction.It is basic thought of the present invention shown in attached drawing 3:By the way that input and two class of user demand totally nine kind will be emulated
Property and corresponding networking model vectorization, normalization, obtain empirical data set;With neural network algorithm to empirical data set
Learnt, obtains training pattern and accuracy rate;It calls training pattern to carry out intelligent reconstruction for new demand servicing, selects best networking
Pattern.Step of the present invention is:
Step 1:Empirical data set is constructed, includes nine attributes and a court verdict, nine attribute difference per data
It is networking scale, portfolio, load capacity, transmission range, maximum hop count, bandwidth, mobility, chronesthesy and packet loss susceptibility;
Attribute is divided into two classes:Emulation input and user demand, emulation input include networking scale, portfolio, load capacity, transmission range,
Seven maximum hop count, bandwidth and mobility attributes, user demand include three portfolio, chronesthesy and packet loss susceptibility categories
Property, wherein portfolio is both emulation input and user demand;Court verdict is the networking model finally to be selected, for
There are three types of the agreements of selection, is respectively:Full-mesh dynamic Time Division networking model, multi-hop carrier wave detect networking model and multi-hop from group
Knit time division multiple acess networking model;For every data, it is at random its nine attribute assignments, calculates each networking in the case
The evaluation score of pattern selects the networking model of highest scoring as the court verdict of the data;It will be taken per data is original
It is worth the empirically data set of the result after quantization.
Step 2:By neural network algorithm, empirical data set is learnt, learning process is divided into two steps:Training
And test;Empirical data set includes that enough data are used for machine learning algorithm, in each learning process, empirical data set
It is all divided into training set and test set at random;In the training process, the parameter that algorithm is determined by the data in training set, obtains
To training pattern;When test, calls training pattern to judge every data in test set, examine the accurate of training pattern
Rate;By repeatedly learning, the mean value and variance of accuracy rate are sought, observes the level and algorithmic stability degree of algorithm accuracy rate, and will
The maximum training result of accuracy rate preserves, the training pattern called when as prediction.
Step 3:Towards new demand servicing, the best networking model of intelligent reconstruction;By new demand servicing content, that is, service corresponding nine categories
Property actual numerical value as input, and according to the quantizing rule in step 1, the actual numerical value of input is quantified as nerve net
The data that network algorithm uses, then call training pattern, obtain three outputs under the conditions of new demand servicing, are exported according to three close
Classification is made decisions to new demand servicing like value, reconstructs the best networking model for meeting current new demand servicing.
According to the method for construction empirical data set of the present invention, has the experience for including more than 1000 valid data
Data set.Every time when study, empirical data set is all divided into two parts at random:More than 500 items determine that algorithm is joined for training
Number, obtains training pattern;Other more than 500 item will call the training pattern obtained in training process in test for testing, right
More than 500 datas are adjudicated one by one in test set, and contrast test result and actual result count final accuracy.By repeatedly learning
Verification is practised, algorithm accuracy rate is maintained at 80% or more.When towards new demand servicing, also can intelligent reconstruction quickly be carried out to new demand servicing.
Description of the drawings
Fig. 1 is full-mesh dynamic Time Division networking model frame structure;
Fig. 2 is self-organizing time division multiple acess networking model frame structure;
Fig. 3 is basic thought of the present invention;
Fig. 4 is neural network structure schematic diagram;
Fig. 5 is single neuron node schematic diagram.
Specific implementation mode
Present invention is further described in detail with reference to the accompanying drawings and examples.
This method is the user oriented service based on neural network algorithm, according to demand for services, to more in free space
Networking model between a node carries out intelligent reconstruction.In narration below, this specification moves the full-mesh that the present invention mentions
State time-division group network pattern is abbreviated as DTDMA (Dynamic Time Division Multiple Access), and multi-hop carrier wave is examined
It surveys networking model and is abbreviated as CSMA/CA (Carrier Sense Multiple Access with Collision
Avoidance), multi-hop ad hoc time division multiple acess networking model is abbreviated as ESTDMA (Enhanced Self-organized
Time Division Multiple Access)。
The specific implementation step of intelligent reconstruction method is given below:
Step 1:Construct empirical data set.
Empirical data designed by this method is concentrated, and includes nine attributes and a court verdict, nine categories per data
Property is that networking scale, portfolio, load capacity, transmission range, maximum hop count, bandwidth, mobility, chronesthesy and packet loss are quick respectively
Sensitivity.Attribute is divided into two classes:Emulation input and user demand.Emulation input includes networking scale, portfolio, load capacity, transmission
Seven distance, maximum hop count, bandwidth and mobility attributes, user demand includes portfolio, chronesthesy and packet loss susceptibility three
A attribute, wherein portfolio is both emulation input and user demand.Court verdict is the networking model finally to be selected,
There are three types of alternative networking models, is respectively:Full-mesh dynamic Time Division networking model, multi-hop carrier wave detection networking model and
Multi-hop ad hoc time division multiple acess networking model.For every data, it is at random its nine attribute assignments, calculates in the case
The evaluation score of each networking model selects the networking model of highest scoring as the court verdict of the data;By every number
Result empirically data set after quantifying according to original value.
The specific method is as follows for original data quantization in empirical data set:
(1) value range of network size is [1 ,+∞], and when value is [1,8], vector turns to 1, value be (8,20]
When, vector turns to 2, value be (20 ,+∞] when, vector turns to 3.
(2) value range of portfolio is [0 ,+∞], and when value is [0,200kbps], vector turns to 1, and value is
(200kbps, 2Mbps] when, vector turns to 2, value be (2Mbps ,+∞] when, vector turns to 3.
(3) value range of load capacity is [0,100%], and when value is [0,33%], vector turns to 1, and value is
(33%, 66%] when, vector turns to 2, value be (66%, 100%] when, vector turns to 3.
(4) value range of transmission range is [0 ,+∞], and when value is [0,1km], vector turns to 1, value be (1km,
20km] when, vector turns to 2, value be (20km ,+∞] when, vector turns to 3.
(5) maximum hop count is divided into two class of single-hop and multi-hop, and when being single-hop, vector turns to 1, when being multi-hop, by its vectorization
It is 2.
(6) value range of bandwidth is [12.8kbps ,+∞], and when value is [12.8kbps, 512kbps], vector turns to
1, value be (512kbps, 2Mbps] when, vector turns to 2, value be (2Mbps ,+∞] when, vector turns to 3.
(7) mobility is divided into " opposing stationary " and " random motion " two class, and when being " opposing stationary ", vector turns to 1, is
When " random motion ", vector turns to 2.
(8) chronesthesy be divided into "high", " in ", " low " three classes, be "high" when, vector turns to 1, for " in " when, vector turns to
2, when being " low ", vector turns to 3.
(9) packet loss susceptibility be divided into "high", " in ", " low " three classes, be "high" when, vector turns to 1, for " in " when, to
2 are quantified as, when being " low ", vector turns to 3.
(10) networking model is divided into three kinds, and when being " full-mesh dynamic Time Division networking model ", vector turns to 1, for " multi-hop carries
Wave detects networking model " when, vector turns to 2, and when being " multi-hop ad hoc time division multiple acess networking model ", vector turns to 3.
The specific configuration method of empirical data set is as follows:
(1) emulation obtains evaluation index
It is at random its nine attribute assignments by taking a data as an example.As shown in table 1, seven emulation of this data are defeated
Enter attribute to be emulated in emulation platform, respectively obtains the evaluation index of three kinds of networking models shown in table 2:Handling capacity, time delay
And packet loss.
1 original emulation input data of table
2 three kinds of networking model evaluation indexes of table
(2) evaluation index is normalized
For convenience of calculating, the evaluation index of obtained in (1) three kinds of networking models is normalized:For each
Evaluation index compares three kinds of networking models, by minimum conduct 1, obtained by the value that other two kinds of networking models are removed with its value
As a result the normalization result as other two kinds of networking models;If for a certain evaluation index, this of a certain networking model refers to
It is designated as 0, then is denoted as 0, and other two kinds of networking models are normalized after the same method.Normalize result
As shown in table 3.
Table 3 normalizes evaluation index
(3) vectorization initial data
The value mode and range of each attribute are different from initial data, and it is defeated can not to be directly used as machine learning algorithm
Enter, can not also carry out functional operation, therefore, it is necessary to which the value of each attribute is done vectorization processing, is allowed to become being suitable for
Machine learning algorithm and the data that functional operation can be carried out.Data after vectorization are as shown in table 4.
Data after 4 vectorization of table
(4) by functional operation Calculation Estimation score, networking model is selected
By above-mentioned steps, the user demand of the normalized evaluation index of three kinds of networking models and vectorization has been obtained,
That is portfolio, chronesthesy, packet loss susceptibility can calculate the evaluation score of each networking model by following formula:
Y=handling capacities × portfolio-time delay × chronesthesy-packet loss × packet loss susceptibility (1)
Y be each networking model evaluation score, handling capacity, time delay and packet loss be normalize after value, portfolio,
Chronesthesy and packet loss susceptibility are the value after original data vector.Each networking model it is expected obtain high-throughput,
Low time delay and low packet loss ratio.Therefore, it using three user demand attributes as weight, is multiplied, handles up with three evaluation indexes respectively
The product of amount and portfolio is just that time delay and chronesthesy and the product of packet loss and packet loss susceptibility are negative, and addition obtains
The evaluation score of each networking model.The evaluation score of three kinds of networking models is as shown in table 5.
The evaluation score of 5 three kinds of networking models of table
The networking model as this data of highest scoring is chosen, that is, chooses most terminations of the DTDMA as this data
Fruit, and court verdict vector is turned to 1.One complete empirical data is as shown in table 6.
6 one complete empirical datas of table
(5) it repeats the above steps, construction includes the empirical data set of a plurality of data.
Step 2:By neural network algorithm, empirical data set is learnt.
The learning process of machine learning algorithm is divided into two steps:Training and test.Algorithm can be determined by training
Parameter obtains training pattern;Test can be used to examine the accuracy rate of training pattern.Empirical data set includes that enough data are used
In machine learning algorithm, in each learning process, empirical data set is all divided into two parts at random, be respectively used to training and
Test.Empirical data set is carried out after repeatedly learning, the average value and variance of accuracy rate are asked.Every time when study, empirical data set
Again two parts are randomly divided into, it is different from the data of test for training to ensure every time.After completing repeatedly study, calculate
To the average value and variance of learning algorithm accuracy rate, the accuracy rate that average value can show algorithm is horizontal, and variance then embodies calculation
The stability of method.Meanwhile the maximum training result of accuracy rate being preserved, the training pattern called when as prediction.
The specific learning method of neural network algorithm is as follows:
(1) empirical data set is divided into two parts at random, is respectively used to training and tests
(2) training
As shown in Fig. 4, neural network is made of input layer, hidden layer and output layer:If input layer is corresponded to per data
Dry attribute;Hidden layer can be one or more layers, and every layer has several nodes;The Different Results that output layer can be exported by several nodes
Classify.For hidden layer and output layer, there are one threshold values for each node itself tool, and between each node of last layer all
There are weights, these parameters are finally determined by carrying out successive ignition to input sample data, it is pre- so as to be carried out to new data
It surveys.As shown in Fig. 5, it is a neuron node schematic diagram, which has n input, xiIt is defeated from i-th neuron
Enter, wiFor the connection weight of i-th of neuron, θ is node itself threshold value, then, the output y of the node is represented by:
For this method using single hidden layer configuration, nine attributes correspond to nine inputs, and three classification results correspond to three outputs.It is logical
It crosses training and determines the threshold value of each node and the connection weight with other nodes.In step 1, nine attributes are vectorial
1,2,3 are turned to, can be respectively represented with " 100 ", " 010 ", " 001 " final in output end directly as the numerical value of input
Tri- classification of DTDMA, CSMA/CA and ESTDMA.When starting to train, first have to all parameters in network, i.e., each node
Threshold value and the random assignment between section [0,1] of the weight between node two-by-two;Then, training intensive data is substituted into successively
Calculated in network, the numerical value of calculated output end will necessarily classification results corresponding with original data have error, according to this
Error can all be finely adjusted the parameter in network after every data calculates, and all data all substitute into training set
After being calculated once in network, it is primary to be denoted as iteration;It repeats the above process, carries out successive ignition calculating, gradually adjust network ginseng
Number, until error accuracy reaches requirement or iterations are enough, to obtain the training pattern of neural network.
(3) it tests
It takes out data from test set successively to be tested, every time when test, using the value of nine attributes as input, warp
Neural computing is crossed, three outputs are obtained in output end;It is approximately 1 by the maximum value in three outputs, other two value is close
Can be approximately " 100 ", " 010 " or " 001 " by three outputs, i.e., according to the approximation of output to this data accordingly like being 0
Make decisions classification;Whether the court verdict and actual result of contrast test data are identical.When all data tests in test set
After, the data bulk of mistake in judgment and the total quantity of test set are recorded, accuracy rate is calculated.
Step 3:Towards new demand servicing, the best networking model of intelligent reconstruction.
In the learning process to empirical data set, the highest training result of accuracy rate is preserved as training pattern.
When towards new demand servicing, by service content, i.e., the actual numerical value input of corresponding nine attributes is advised according to the quantization described in step 1
Then, the actual numerical value of input translates into the data used for machine learning algorithm, i.e., actual numerical value is quantified as 1,2,3;
Then training pattern can be called, obtain three outputs under the conditions of new demand servicing, according to three approximations " 100 " exported,
" 010 " or " 001 " makes decisions classification to new demand servicing, reconstructs the best networking model for meeting current new demand servicing.
The content not being described in detail in the present patent application book belongs to the prior art well known to professional and technical personnel in the field.
Claims (3)
1. a kind of service-oriented wireless network networking model intelligent reconstruction method, used step are:
Step 1:Empirical data set is constructed, includes nine attributes and a court verdict per data, nine attributes are group respectively
Network planning mould, portfolio, load capacity, transmission range, maximum hop count, bandwidth, mobility, chronesthesy and packet loss susceptibility;Attribute
It is divided into two classes:Emulation input and user demand, emulation input includes networking scale, portfolio, load capacity, transmission range, maximum
Seven hop count, bandwidth and mobility attributes, user demand include three portfolio, chronesthesy and packet loss susceptibility attributes,
In, portfolio is both emulation input and user demand;Court verdict is the networking model finally to be selected, available
Agreement there are three types of, be respectively:When full-mesh dynamic Time Division networking model, multi-hop carrier wave detection networking model and multi-hop ad hoc
Divide multiple access networking model;For every data, it is at random its nine attribute assignments, calculates each networking model in the case
Evaluation score, select the networking model of highest scoring as the court verdict of the data;It will be per the original value amount of data
Result after change empirically data set;
Step 2:By neural network algorithm, empirical data set is learnt, learning process is divided into two steps:Training and survey
Examination;Empirical data set include enough data be used for machine learning algorithm, in each learning process, empirical data set all by
It is divided into training set and test set at random;In the training process, the parameter that algorithm is determined by the data in training set, is instructed
Practice model;When test, calls training pattern to judge every data in test set, examine the accuracy rate of training pattern;
By repeatedly learning, the mean value and variance of accuracy rate are sought, observes the level and algorithmic stability degree of algorithm accuracy rate, and will be accurate
The maximum training result of rate preserves, the training pattern called when as prediction;
Step 3:Towards new demand servicing, the best networking model of intelligent reconstruction;By new demand servicing content, that is, service corresponding nine attributes
The actual numerical value of input is quantified as calculating for neural network by actual numerical value as input, and according to the quantizing rule in step 1
The data that method uses, then call training pattern, obtain three outputs under the conditions of new demand servicing, the approximation exported according to three
Classification is made decisions to new demand servicing, reconstructs the best networking model for meeting current new demand servicing.
2. a kind of service-oriented wireless network networking model intelligent reconstruction method according to claim 1, feature exist
The specific method of original data quantization is in empirical data set:
(1) value range of network size is [1 ,+∞], and when value is [1,8], vector turns to 1, value be (8,20] when, to
Be quantified as 2, value be (20 ,+∞] when, vector turns to 3;
(2) value range of portfolio is [0 ,+∞], and when value is [0,200kbps], vector turns to 1, and value is
(200kbps, 2Mbps] when, vector turns to 2, value be (2Mbps ,+∞] when, vector turns to 3;
(3) value range of load capacity is [0,100%], and when value is [0,33%], vector turns to 1, value be (33%,
When 66%], vector turns to 2, value be (66%, 100%] when, vector turns to 3;
(4) value range of transmission range is [0 ,+∞], and when value is [0,1km], vector turns to 1, value be (1km, 20km]
When, vector turns to 2, value be (20km ,+∞] when, vector turns to 3;
(5) maximum hop count is divided into two class of single-hop and multi-hop, and when being single-hop, vector turns to 1, and when being multi-hop, its vector is turned to 2;
(6) value range of bandwidth is [12.8kbps ,+∞], and when value is [12.8kbps, 512kbps], vector turns to 1, takes
Value for (512kbps, 2Mbps] when, vector turns to 2, value be (2Mbps ,+∞] when, vector turns to 3;
(7) mobility is divided into " opposing stationary " and " random motion " two class, and when being " opposing stationary ", vector turns to 1, is " random
When movement ", vector turns to 2;
(8) chronesthesy be divided into "high", " in ", " low " three classes, be "high" when, vector turns to 1, for " in " when, vector turns to 2, is
When " low ", vector turns to 3;
(9) packet loss susceptibility be divided into "high", " in ", " low " three classes, be "high" when, vector turns to 1, for " in " when, vectorization
It is 2, when being " low ", vector turns to 3;
(10) networking model is divided into three kinds, and when being " full-mesh dynamic Time Division networking model ", vector turns to 1, for " multi-hop carrier wave is examined
Survey networking model " when, vector turns to 2, and when being " multi-hop ad hoc time division multiple acess networking model ", vector turns to 3.
3. a kind of service-oriented wireless network networking model intelligent reconstruction method according to claim 1, feature exist
It is in the specific method of construction empirical data set:
(1) emulation obtains evaluation index, in nine attributes of empirical data set, networking scale, portfolio, load capacity, transmission
This seven attributes of distance, maximum hop count, bandwidth and mobility are to pass through the original to this seven attributes as the attribute of emulation input
Beginning numerical value is emulated, and three evaluation indexes of three kinds of networking models are obtained:Handling capacity, time delay and packet loss;
(2) evaluation index is normalized, three evaluation indexes are normalized, for each evaluation index, compares three kinds
Minimum conduct 1 is removed the value of other two kinds of networking models with its value, acquired results are as other two kinds by networking model
The normalization result of networking model;If for a certain evaluation index, the index of a certain networking model is 0, then is denoted as
0, and other two kinds of networking models are normalized after the same method;
(3) value of each attribute is done vectorization processing by vectorization initial data, and being allowed to, which becomes being suitable for machine learning, calculates
Method and the data that functional operation can be carried out;
(4) by functional operation Calculation Estimation score, networking model is selected, with evaluating for following formula three kinds of networking models of calculating
Point:
Y=handling capacities × portfolio-time delay × chronesthesy-packet loss × packet loss susceptibility (1)
Wherein, y is the evaluation score of each networking model, and handling capacity, time delay and packet loss are the value after normalization, business
Amount, chronesthesy and packet loss susceptibility are the value after original data vector;Each networking model it is expected that obtaining height handles up
Therefore amount, low time delay and low packet loss ratio using portfolio, chronesthesy and packet loss susceptibility as weight, are evaluated with three respectively
Index is multiplied, and the product of handling capacity and portfolio is denoted as just, and time delay and chronesthesy and packet loss and packet loss susceptibility multiply
Product be denoted as it is negative, addition obtain the evaluation score of each networking model;Finally, the networking model of highest scoring is chosen as every number
According to final result;
(5) it repeats the above steps, construction includes the empirical data set of a plurality of data.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810703460.9A CN108770010B (en) | 2018-06-26 | 2018-06-26 | Intelligent reconstruction method for service-oriented wireless network networking mode |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810703460.9A CN108770010B (en) | 2018-06-26 | 2018-06-26 | Intelligent reconstruction method for service-oriented wireless network networking mode |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108770010A true CN108770010A (en) | 2018-11-06 |
CN108770010B CN108770010B (en) | 2021-12-14 |
Family
ID=63975238
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810703460.9A Active CN108770010B (en) | 2018-06-26 | 2018-06-26 | Intelligent reconstruction method for service-oriented wireless network networking mode |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108770010B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111340205A (en) * | 2020-02-18 | 2020-06-26 | 中国科学院微小卫星创新研究院 | Anti-irradiation system and method of neural network chip for space application |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6606744B1 (en) * | 1999-11-22 | 2003-08-12 | Accenture, Llp | Providing collaborative installation management in a network-based supply chain environment |
WO2015180397A1 (en) * | 2014-05-31 | 2015-12-03 | 华为技术有限公司 | Method and device for recognizing data category based on deep neural network |
CN105578472A (en) * | 2016-01-19 | 2016-05-11 | 南京微传物联网科技有限公司 | Wireless sensor network performance online optimization and planning method based on immune theory |
CN105761488A (en) * | 2016-03-30 | 2016-07-13 | 湖南大学 | Real-time limit learning machine short-time traffic flow prediction method based on fusion |
CN107135041A (en) * | 2017-03-28 | 2017-09-05 | 西安电子科技大学 | A kind of RBF neural channel prediction method based on phase space reconfiguration |
CN107370676A (en) * | 2017-08-03 | 2017-11-21 | 中山大学 | Fusion QoS and load balancing demand a kind of route selection method |
-
2018
- 2018-06-26 CN CN201810703460.9A patent/CN108770010B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6606744B1 (en) * | 1999-11-22 | 2003-08-12 | Accenture, Llp | Providing collaborative installation management in a network-based supply chain environment |
WO2015180397A1 (en) * | 2014-05-31 | 2015-12-03 | 华为技术有限公司 | Method and device for recognizing data category based on deep neural network |
CN105578472A (en) * | 2016-01-19 | 2016-05-11 | 南京微传物联网科技有限公司 | Wireless sensor network performance online optimization and planning method based on immune theory |
CN105761488A (en) * | 2016-03-30 | 2016-07-13 | 湖南大学 | Real-time limit learning machine short-time traffic flow prediction method based on fusion |
CN107135041A (en) * | 2017-03-28 | 2017-09-05 | 西安电子科技大学 | A kind of RBF neural channel prediction method based on phase space reconfiguration |
CN107370676A (en) * | 2017-08-03 | 2017-11-21 | 中山大学 | Fusion QoS and load balancing demand a kind of route selection method |
Non-Patent Citations (2)
Title |
---|
G. VARAPRASAD: "《Power Aware and Signal Strength Based Routing Algorithm for Mobile Ad Hoc Networks》", 《2011 INTERNATIONAL CONFERENCE ON COMMUNICATION SYSTEMS AND NETWORK TECHNOLOGIES》 * |
王振邦: "《基于认知的无线网络自适应通信方法》", 《中国博士学位论文全文数据库信息科技辑》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111340205A (en) * | 2020-02-18 | 2020-06-26 | 中国科学院微小卫星创新研究院 | Anti-irradiation system and method of neural network chip for space application |
CN111340205B (en) * | 2020-02-18 | 2023-05-12 | 中国科学院微小卫星创新研究院 | Neural network chip anti-irradiation system and method for space application |
Also Published As
Publication number | Publication date |
---|---|
CN108770010B (en) | 2021-12-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN113873022A (en) | Mobile edge network intelligent resource allocation method capable of dividing tasks | |
CN104185298B (en) | Network load dynamic self-adapting parameter regulation means based on priority | |
CN109862610A (en) | A kind of D2D subscriber resource distribution method based on deeply study DDPG algorithm | |
Neel et al. | Game models for cognitive radio algorithm analysis | |
EP2043398A2 (en) | Method for reducing inter-cell interference in wireless ofdma networks | |
CN106454856A (en) | Spectrum allocation method based on graph coloring and analytic hierarchy process in cognitive radio | |
CN107040948A (en) | A kind of CSMA/CA optimization methods based on priority | |
CN112637883B (en) | Federal learning method with robustness to wireless environment change in electric power Internet of things | |
Gao et al. | Spectrum sharing for LTE and WiFi coexistence using decision tree and game theory | |
CN111314928A (en) | Wireless ad hoc network performance prediction method based on improved BP neural network | |
Valcarce et al. | Toward joint learning of optimal MAC signaling and wireless channel access | |
Balakrishnan et al. | Deep reinforcement learning based traffic-and channel-aware OFDMA resource allocation | |
CN107105455A (en) | It is a kind of that load-balancing method is accessed based on the user perceived from backhaul | |
Qin et al. | Pathway to intelligent radio | |
CN108770010A (en) | A kind of service-oriented wireless network networking model intelligent reconstruction method | |
CN111741450A (en) | Network flow prediction method and device and electronic equipment | |
Wilhelmi et al. | Machine learning for performance prediction of channel bonding in next-generation IEEE 802.11 WLANs | |
CN108882377A (en) | Resource allocation methods in LTE-R based on cognition | |
CN109548061A (en) | One kind being based on markovian cognition wireless network saturation throughput method for solving | |
CN112654063A (en) | Uplink capacity assessment method and device | |
Jiang et al. | Cognitive radio spectrum sharing schemes with reduced spectrum sensing requirements | |
CN103281701B (en) | A kind of collaboration frequency spectrum Forecasting Methodology of cognitive radio networks | |
CN106937326A (en) | Method of coordinated transmission and first base station between base station | |
CN104219679B (en) | A kind of improved method for channel allocation in cognitive radio networks | |
Sirhan et al. | Cognitive Radio Resource Scheduling using Multi agent QLearning for LTE |
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 |