CN105554873B - A kind of Wireless Sensor Network Located Algorithm based on PSO-GA-RBF-HOP - Google Patents
A kind of Wireless Sensor Network Located Algorithm based on PSO-GA-RBF-HOP Download PDFInfo
- Publication number
- CN105554873B CN105554873B CN201510757614.9A CN201510757614A CN105554873B CN 105554873 B CN105554873 B CN 105554873B CN 201510757614 A CN201510757614 A CN 201510757614A CN 105554873 B CN105554873 B CN 105554873B
- Authority
- CN
- China
- Prior art keywords
- node
- particle
- anchor node
- pso
- formula
- 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.)
- Expired - Fee Related
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W64/00—Locating users or terminals or network equipment for network management purposes, e.g. mobility management
- H04W64/003—Locating users or terminals or network equipment for network management purposes, e.g. mobility management locating network equipment
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W4/00—Services specially adapted for wireless communication networks; Facilities therefor
- H04W4/02—Services making use of location information
- H04W4/025—Services making use of location information using location based information parameters
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W84/00—Network topologies
- H04W84/18—Self-organising networks, e.g. ad-hoc networks or sensor networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Theoretical Computer Science (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Genetics & Genomics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Physiology (AREA)
- Position Fixing By Use Of Radio Waves (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of Wireless Sensor Network Located Algorithms based on PSO-GA-RBF-HOP.The algorithm comprises the following steps: Step 1: anchor node initializes, and calculating item number average distance；Step 2: hop count updates；Step 3: generating dummy node at random, training sample is generated；Step 4: carrying out variation update to the parameter in PSO with GA；Step 5: with the specific steps of GA-PSO optimization RBF parameter；Step 6: unknown node positions.The present invention is on the basis of the DV-HOP localization method of wireless sensor network non-range, introduce RBF neural network algorithm, optimizing is carried out to parameters such as center, the errors of neural network using GA and PSO algorithm simultaneously, keeps positioning more accurate, effectively increases positioning accuracy.
Description
Technical field
The present invention relates to a kind of algorithms of non-ranging wireless sensor network positioning neural network based, while utilizing grain
Subgroup and genetic algorithm improve radial basis function neural network, realize the raising of precision.
Background technique
Wireless sensor network (Wireless Sensor Networks, WSN) is made of several sensor nodes
Network topology structure, these nodes be widely distributed monitoring region in, they are communicated using wireless communication protocol,
The network topology structure that will form a multi-hop in region, by a large amount of sensor to the monitoring object in institute overlay area into
The acquisition of row data, while cooperating with each other between sensor and information is finally fed back to observer.In WSN network, sensor positioning
Technology is mostly important technology, this technology can reflect the location information of each sensor node, to be more advantageous to sight
Survey person grasps the situation in monitoring region, and location algorithm can be divided into: (range-based) location algorithm based on distance and away from
From unrelated (range-free) location algorithm.
(range-based) location algorithm based on distance is mainly based upon distance value or angle value between sensor node
Come what is calculated, the most typical one algorithm has TOA, TDOA, AOA, RSSI etc., but based on the location algorithm of distance to the external world
Environment and hardware supported requirement are very high, such as in the algorithm based on Distance positioning, TOA algorithm needs time synchronization very smart
Really, the selling at exorbitant prices for the sensor that TDOA algorithm needs, AOA algorithm is interfered by outside extremely serious.Apart from unrelated
(range-free) location algorithm, cardinal principle are positioned by the connection between network, and other phases are not needed
Hardware device is closed, but precision is frequently not very high, common distance that unrelated location algorithm has convex programming, mass center, DV-Hop, A-
morphous,APIT.In summary, it would be highly desirable to develop it is a kind of with apart from unrelated location in sensors network algorithm, can be more accurate
Ground positioning can also improve precision, become the target that the public studies.
It is various in unrelated localization method, using more and what is be easily modified surely belongs to DV-Hop algorithm, the calculation
Method is fewer and positioning accuracy is relatively high between the proportion requirement calibration node, thus in recent years to the research of this algorithm more
Extensively, to DV-Hop location algorithm in wireless sensor network, there is numerous improved methods, to the property of DV-Hop location algorithm
It especially accuracy can all increase significantly, thus in WSN location algorithm, based on DV-Hop algorithm, compound one kind
Or a variety of approximate algorithms after optimizing operates, become emphasis concerned by people.
Summary of the invention
In order to overcome the above-mentioned deficiencies of the prior art, the present invention provides one kind based on radial basis function neural network and
The Wireless Sensor Network Located Algorithm that DV-Hop is combined, while using population and genetic algorithm to the ginseng in neural network
Number optimizes.
Radial basis function (Radial Basis Function, RBF) is to belong to artificial neural network (Artificial
Neural Networks, ANN) scope, its maximum feature is most preferably to approach with uniqueness, while can be to avoid part
The problem of approaching exists, and RBF neural has intuitive and good " input-output " mapping function, uses in the present invention
WSN location algorithm of the RBF neural in conjunction with DV-Hop.But the convergence rate of RBF neural is very slow but also exists
Multiple ill-conditioning problems such as generalization ability is weaker, model parameter and structure are difficult to be arranged, the solution of connection weight, based on improving precision
The considerations of, some parameters in RBF are introduced global optimization approach and are optimized, to reach the property for improving RBF neural
Can effect, in numerous global optimization approaches, particle swarm algorithm (particle swarm optimization, PSO) due to
Its mathematic(al) structure model is easy to use, to be widely used.But PSO algorithm equally exists some disadvantages,
Such as convergence rate is slow, precision is low, the easy precocity of population etc..It therefore will be in genetic algorithm (Genetic Algorithm, GA)
Select, intersect, make a variation these core algorithms operation be introduced into particle swarm algorithm, while accelerating population convergence rate and also
Solves the premature convergence problem of population.
The present invention provides a kind of location algorithm of wireless sensor network based on PSO-GA-RBF-HOP, and main includes such as
Under several steps:
Step 1: anchor node initialization and the calculating of hop count average distance；
Step 2: hop count updates；
Step 3: generating dummy node at random, training data is generated；
Step 4: carrying out variation update to the parameter in PSO with GA；
Step 5: with the specific steps of GA-PSO optimization RBF parameter；
Step 6: unknown node positions.
The present invention has the advantages that
1. although positioning method effect of the non-ranging mode location in sensors network in performance not as good as ranging is good,
It is non-ranging independent of external environment and hardware condition, non-ranging method neural network based has more good performance.
2. improving particle swarm algorithm using genetic algorithm, by population crossover operation and mutation operation, seek iteration essence
The situation that degree is maximum or fitness is best, the disadvantage that can effectively avoid the slow precision of the convergence rate of PSO low improve particle populations
Diversity, also solve the premature convergence problem of population,
3. be introduced into modified particle swarm optiziation for determine sample data center in RBF neural, sample variance with
And the connection weight width of hidden layer and output layer, the sample data of the neural network more adaptive expectation after making training, thus
Improve the precision of location in sensors network.
Detailed description of the invention
Fig. 1 is the flow chart of sensor positioning method in the present invention；
Fig. 2 is RBF neural schematic diagram in the present invention；
Fig. 3 is the coordinate position figure of anchor node in case study on implementation；
Fig. 4 is that the error of coordinate of location algorithm RBF, GA-RBF and PSO-GA-RBF in case study on implementation compare figure；
Specific embodiment
The present invention is described in detail with reference to the accompanying drawings and examples.
The present invention is a kind of Wireless Sensor Network Located Algorithm based on PSO-GA-RBF-HOP, system block diagram such as Fig. 1
It is shown, specifically include that steps are as follows:
Step 1: anchor node initialization and the calculating of hop count average distance；
System needs hop count data information corresponding between each anchor node carrying out initialization return-to-zero first, often
Using the communication mode broadcasted, downstream anchor node provides the location information of oneself to a anchor node, and subsequent system is further according in WSN
The location information of each anchor node itself, by the corresponding hop count and distance of statistics origin to each anchor node, to count
The average distance of hop count is calculated, each anchor node A is according to the location information (x, y) and corresponding hop count of each anchor node of collection
Value h calculates the average distance c of hop count according to formula (1)i。
Step 2: hop count updates；
In WSN network, each unknown sensor node can seek nearest anchor node by way of broadcast, this is recently
The hop count numerical information recorded on anchor node adds 1, then nearest anchor node can be transmitted to its corresponding downstream anchor node, downstream anchor section
After point receives hop count information, hop count also will be updated, while forwarding downstream anchor node, this step is repeated, until all unknown sections
The hop count of point to each anchor node is fixed, until being no longer changed.Assuming that anchor node quantity is M, unknown node quantity is N, then
The hop count matrix for needing to generate a N × M is inputted as sample.
Step 3: dummy node is generated at random, training data cleaning；
In the already fixed region WSN, V dummy node coordinate, dummy node number are generated at random using software
Respectively 1,2,3 ..., V, node coordinate Vxyi=(xi, yi), the anchor node hop count update method according to step 2,
The dummy node of each generation is calculated separately out to the hop count matrix for each anchor node fixed, remembers dummy node ViTo anchor
The hop count matrix of node is VMi=[hi1, hi2, hi3..., hiM], using the matrix as neural with the improved RBF of GA and PSO
The training sample of network training inputs, and corresponding training sample desired output is the actual coordinate of the dummy node, i.e. Vxyi。
After obtaining training sample, data screening is also carried out to training sample, the screening technique chosen in the present invention is
Principle component analysis.Principle component analysis (principal component analysis, PCA) is also referred to as factor analysis,
It is very common data screening method.For given training sample set H=[VMi|Vxyi]=[hij, hij..., hV(M+2)]T, instruction
Practicing sample is V × (M+2) matrix.Enable hij'=hij- E (h), H '=[h11', h12' ..., hV(M+2)'] make center of a sample,
As shown in formula (2):
Shown in the covariance matrix of input sample such as formula (3):
CH=E (H ' (H ')T) (3)
C is calculated using formula (4)HEigenvalue λ1, λ2..., λ(M+2)With corresponding feature vector u1, u2...,
u(M+2):
CH*ui=λi*ui, i=1,2 ..., M+2 (4)
If eigenvalue λ1≥λ2≥...≥λ(M+2), then ui TIt is exactly projection of the input to feature vector.Enable U=[u1,
u2..., u(M+2)], then sample set is restructural as shown in formula (5):
Y=UT*H′ (5)
(M+2) of input is tieed up by sample set X by eigenvectors matrix U in formula (5) and is transformed to (the M+ in feature space
2) sample set Y, sample y in Y are tieed upiA certain component yijX in as XiJ-th of principal component of sample.In order in feature space
Selection main component gives up secondary component, defines shown in variance contribution ratio such as formula (6):
Step 4: GA carries out variation update to the parameter in PSO.
In particle swarm algorithm, it is assumed that the space of the algorithm search is n dimension, and num is enabled to indicate total number of particles, i-th of particle
Position vector is expressed as Xi=[xi1, xi2..., xin], velocity vector is expressed as Vi=[vi1, vi2..., vin], what particle lived through
Desired positions are denoted as Pi=(pi1, pi2..., pin), also referred to as Pibest.The desired positions that all particles live through in group
Call number indicates with symbol g, i.e. Pg=(pg1, pg2..., pgn), also referred to as gbest.Each particle is in iterative process in PSO algorithm
In the speed of particle and position are updated according to such as formula (7), (8):
Vi(k+1)=w Vi(k)+c1r1[pi-xi(k)]+c2r2[pg-xi(k)] (7)
Xi(k+1)=Xi(k)+Vi(k+1) (8)
In formula: k indicates that evolutionary generation, w are inertia weight coefficient；c1、c2For accelerated factor, also referred to as perception factor and society
It can the factor；r1、r2For the random number between [0,1].
The variation thought that genetic algorithm is used for reference in GA-PSO algorithm is implemented to intersect and make a variation in the motion process of population
Operation, improves the search capability of particle.Carry out crossover operation first: one group of preferable particle of fitness of selection is matched at random two-by-two
It is right, according to certain probability PCIntersected.For the particle X of pairingi、Xj, shown in crossover process such as formula (9) and (10):
In formula: α1、α2Random number respectively between [0,1], formula (10) and (11) respectively indicate position and speed to particle
Degree is intersected.Then the fitness value for detecting each particle, the particle poor for fitness is with certain probability PvIt carries out
Random initializtion improves the population diversity of particle, effectively overcomes the precocious phenomenon of population.
Step 5: GA-PSO optimizes RBF parameter.
Have the characteristics that Multi-layered Feedforward Networks in RBF neural.In classical RBF neural system, include
Three input layer, hidden layer, output layer parts form, and network topology structure is as shown in Figure 2.The section of input layer in a network
Point number is determined by the data dimension of input signal；It is hidden layer among input layer and output layer, hidden layer is by defeated
Enter one group of radial basis function that data obtain to constitute, is in the application typically all the meter for selecting Gaussian function as radial basis function
Calculation method；The data output of hidden layer to output layer follows weight superposition principle, finally calculates the output data of RBF network.
Assuming that the node number in RBF neural input layer is I, the node number of hidden layer is J, the node number of output layer is K,
When input is H=[hij, hij..., hV(M+2)]TWhen, shown in the output such as formula (11) of j-th of neuron of hidden layer:
In formula: CjFor the sample data center of Gaussian bases, it is the vector of I dimension；σjIt is then Gaussian bases
Sample variance.Output layer follows the principle of linear convergent rate, it is to be obtained by the output of hidden layer by linear combination.Output layer
Each neuron output data such as formula (12) shown in:
In formula: wjkThe connection weight for indicating hidden layer each node and each node of output layer, is the matrix of a J × K.
GA-PSO is used for the parameter optimisation procedure of RBF neural, search for RBF neural sample data center,
The connection weight width of sample variance and hidden layer and output layer.At the same time, by RBF neural to learning sample data
Fitness function of the error as population, as shown in formula (13):
In formula: M is number of samples, dikFor the desired output of node.yikIt is the corresponding reality output of input sample.
Specific optimization process is as follows:
1) according to the hop count information matrix of dummy node to each anchor node, the data training sample of neural network is obtained, by
This determines the node number of the input layer of applied RBF neural, hidden layer, output layer, while determining and nerve is arranged
The connection weight value range of the data center of Gaussian function, variance and hidden layer to output layer in network hidden layer.
2) it first has to initialize return-to-zero to the parameters in particle swarm algorithm.According to the practical feelings of neural network
The number of population and the position and speed coding of each particle is arranged in condition.So that it is determined that particle is most in particle swarm algorithm
Big pace of learning is Vmax, Studying factors are then the random number c between 0 to 11With c2, inertia weight w, maximum cycle is
mmax, the permission trueness error value E that is solved.Determine that the crossover probability in genetic algorithm is P againcIt is P with mutation probabilityv。
3) iterative learning.The position and speed information of each particle is updated according to formula (7), (8), while according to formula
(13) fitness value of more new particle.
4) crossover operation.Using the thought of genetic algorithm, according to formula (9), the principle of (10), from population calculating process
In select the preferable particle of partial fitness value, carry out crossover operation two-by-two.Compare the progeny after intersecting and first for particle
Fitness function value, the good particle of fitness value, which can be put into population, carries out next iteration.
5) mutation operation.Using the thought of genetic algorithm, the particle poor to those fitness function values in population with
Probability PvCarry out random initializtion.
6) according to step 4) and 5) in mutation operation, angle value and best position are preferably adapted to particle each in population
Set PibestValue carries out optimizing operation；Then angle value and global desired positions g are being preferably adapted to the overall situation in populationbestIt carries out
Optimizing operation.
7) if the number of iterations of population has reached best in set maximum number of iterations or population
Global fitness value reaches set precision, then jumps out loop iteration termination, on the contrary then continue to execute step 3) and be iterated.
Step 6: unknown node positions.
During positioning to unknown node, optimal searching principle is also followed, every iteration once all selects optimal section
Point input network is positioned, and the node having had good positioning is incorporated as anchor node, further according to new unknown node and anchor node
Hop count relationship, repeat the above steps and carry out cycle calculations, until all nodes are completed to position.By constantly optimal section
Point is positioned, and oriented node is added in anchor node, while updating unknown node to the hop count between anchor node, circulation
Back and forth until all unknown nodes complete positioning work.
Embodiment one
Step 1: anchor node initialization and the calculating of hop count average distance；
According to demand, in 200 × 200 region, 20 anchor nodes are fixed, add origin, altogether 21 anchor nodes,
Coordinate of 21 anchor nodes in region is as shown in table 1.
21 anchor node coordinates in 1 region of table
| Anchor node number | 1 | 2 | 3 | 4 | 5 | 6 | 7 |
| The X-axis left side | 116.45 | 108.15 | 173.99 | 52.96 | 63.61 | 23.84 | 187.97 |
| Y axis coordinate | 80.92 | 89.67 | 73.16 | 152.70 | 125.58 | 154.40 | 186.57 |
| Anchor node number | 8 | 9 | 10 | 11 | 12 | 13 | 14 |
| The X-axis left side | 129.11 | 95.89 | 127.86 | 108.94 | 129.46 | 108.78 | 144.21 |
| Y axis coordinate | 194.55 | 38.41 | 27.77 | 139.25 | 18.76 | 105.08 | 106.07 |
| Anchor node number | 15 | 16 | 17 | 18 | 19 | 20 | 21 |
| The X-axis left side | 104.50 | 198.74 | 43.74 | 21.16 | 21.94 | 12.72 | 0 |
| Y axis coordinate | 172.23 | 96.97 | 78.69 | 134.29 | 148.25 | 104.01 | 0 |
21 anchor nodes are defined in region, the hop count relationship between anchor node is as shown in figure 3, each anchor node arrives according to statistics
The hop count of origin is as shown in table 2.
Hop count of each anchor node of table 2 to origin
| Anchor node number | 1 | 2 | 3 | 4 | 5 | 6 | 7 |
| Hop count | 4 | 2 | 5 | 4 | 3 | 5 | 6 |
| Anchor node number | 8 | 9 | 10 | 11 | 12 | 13 | 14 |
| Hop count | 7 | 5 | 6 | 3 | 7 | 3 | 4 |
| Anchor node number | 15 | 16 | 17 | 18 | 19 | 20 | 21 |
| Hop count | 4 | 5 | 1 | 3 | 4 | 2 | 0 |
According to formula (1), calculating hop count average distance is 9.38.
Step 2: hop count updates；
Unknown node 50 is shared in region, anchor node 21, anchor node distribution map is as shown in figure 3, collect each unknown section
Point to the hop count between each anchor node, if encounter certain unknown node to certain anchor node hop count there are multiple values the case where, without exception
It is subject to minimum value.
Step 3: dummy node is generated at random, training neural network；
In 200 × 200 region, 150 dummy nodes are generated at random, by the analysis of PCA method to 150 virtual sections
Point carries out dimensionality reduction, has obtained 101 nodes, inputs using the hop count of this 101 nodes to 21 anchor nodes as training sample,
The coordinate of this 101 dummy nodes is exported as training sample, and the training sample of generation is normalized.
Step 4: carrying out variation update to the parameter in PSO with GA；
It is obtained according to emulation, carries out parameter update using only PSO, needing the number of iteration is to set in advance in the present embodiment
The maximum times set 1000 times, and be introduced into the cross and variation in genetic algorithm, then iteration 512 times are that can reach in the present embodiment in advance
The optimal adaptation degree first set.
Step 5: with the specific steps of GA-PSO optimization RBF parameter；
Step 4: five operation, the optimizing effect carried out using GA-PSO to RBF, can determine this according to the present invention
The input layer number of RBF neural is 21, and output layer number of nodes is 2, and the number of hidden nodes 21, hidden layer Gaussian function approaches
Center is the 37th group of data of training sample, and sample variance is the average variance of 37 groups of data.
Step 6: unknown node positions.
In order to examine the location algorithm of this paper, to DV-Hop algorithm, RBF-Hop algorithm, PSO-RBF-HOP algorithm, GA-
PSO-RBF-HOP algorithm carries out simulation comparison under Matlab platform, as shown in Figure 4.According to simulation result, four kinds of algorithms are compared
Range error value, RBF-Hop ratio DV-Hop improves 21% or so positioning accuracy, and PSO-RBF-Hop ratio RBF-Hop is improved
23% or so positioning accuracy, GA-PSO-RBF-Hop ratio PSO-RBF-Hop improve 49% positioning accuracy.Thus may be used
See, node positioning method provided by the invention can effectively increase the positioning accuracy of wireless sensor network.
Claims (3)
1. a kind of Wireless Sensor Network Located Algorithm based on PSO-GA-RBF-HOP, which is characterized in that specifically include step
It is as follows:
Step 1: anchor node initializes, and calculate item number average distance；
System needs hop count data information corresponding between each anchor node carrying out initialization return-to-zero first, secondly every
Using the communication mode broadcasted, downstream anchor node provides the location information of oneself to a anchor node, and subsequent system is further according in WSN
The location information of each anchor node itself, by the corresponding hop count and distance of statistics origin to each anchor node, to count
Calculate the average distance of hop count；
Step 2: hop count updates；
In WSN network, each unknown sensor node can seek nearest anchor node by way of broadcast, the nearest anchor section
The hop count numerical information recorded on point adds 1, which can be transmitted to his corresponding downstream anchor node, and downstream anchor node receives jump
After number information, hop count also will be updated, while forwarding downstream anchor node, this step is repeated, until all unknown nodes to each anchor
The hop count of node is fixed, until being no longer changed；Assuming that anchor node quantity is M, unknown node quantity is N, then needs to generate
The hop count matrix of one N × M is inputted as sample；
Step 3: generating dummy node at random, training sample is generated；
In the already fixed region WSN, V dummy node coordinate, dummy node number difference are generated at random using software
It is 1,2,3 ..., V, node coordinate Vxyi=(xi,yi), the anchor node hop count update method according to step 2, respectively
The dummy node of each generation is calculated to the hop count matrix for each anchor node fixed, remembers dummy node ViTo anchor node
Hop count matrix be VMi=[hi1,hi2,hi3,…,hiM], it is instructed using the matrix as with the improved RBF neural of GA and PSO
Experienced training sample input, corresponding training sample desired output are the actual coordinate of the dummy node, i.e. Vxyi；
Step 4: carrying out variation update to the parameter in PSO with GA；
In particle swarm algorithm, it is assumed that the space of the algorithm search is n dimension, and num is enabled to indicate total number of particles, i-th of particles position
Vector is expressed as Xi=[xi1,xi2,…,xin], velocity vector is expressed as Vi=[vi1,vi2,…,vin], particle lives through best
Position is denoted as Pi=(pi1,pi2,…,pin), also referred to as Pibest.The index for the desired positions that all particles live through in group
It number is indicated with symbol g, i.e. Pg=(pg1,pg2,…,pgn), also referred to as gbest；Each particle is pressed in an iterative process in PSO algorithm
The speed of particle and position are updated according to such as formula (7), (8):
Vi(k+1)=wVi(k)+c1r1[pi-xi(k)]+c2r2[pg-xi(k)] (7)
Xi(k+1)=Xi(k)+Vi(k+1) (8)
In formula: k indicates that evolutionary generation, w are inertia weight coefficient；c1、c2For accelerated factor, also referred to as perception factor and society because
Son；r1、r2For the random number between [0,1]；
The variation thought that genetic algorithm is used for reference in GA-PSO algorithm is implemented to intersect and be grasped with variation in the motion process of population
Make, improves the search capability of particle；Carry out crossover operation first: one group of preferable particle of fitness of selection is matched at random two-by-two
It is right, according to certain probability PCIntersected；For the particle X of pairingi、Xj, shown in crossover process such as formula (9) and (10):
In formula: α1、α2Random number respectively between [0,1], formula (10) and (11) respectively indicate to the position and speed of particle into
Row intersects；Then the fitness value for detecting each particle, the particle poor for fitness is with certain probability PvIt carries out random
Initialization, improves the population diversity of particle, effectively overcomes the precocious phenomenon of population；
Step 5: optimizing RBF parameter with GA-PSO；
Have the characteristics that Multi-layered Feedforward Networks in RBF neural；In classical RBF neural system, include input
Three layer, hidden layer, output layer parts form；The interstitial content of input layer in a network is the data dimension by input signal
It determines；It is hidden layer, one group of radial basis function structure that hidden layer is obtained by input data among input layer and output layer
At being in the application typically all to select calculation method of the Gaussian function as radial basis function；Data of the hidden layer to output layer
Output follows weight superposition principle, finally calculates the output data of RBF network；
Step 6: unknown node positions；
During positioning to unknown node, optimal searching principle is also followed, it is defeated that every iteration once all selects optimal node
Enter network to be positioned, the node having had good positioning is incorporated as anchor node, further according to the jump of new unknown node and anchor node
Number relationship, repeats the above steps and carries out cycle calculations, until all nodes are completed to position；By constantly optimal node into
Row positioning, oriented node is added in anchor node, while updating unknown node to the hop count between anchor node, is moved in circles
Until all unknown nodes complete positioning work.
2. the Wireless Sensor Network Located Algorithm according to claim 1 based on PSO-GA-RBF-HOP, feature exist
In: in step 3 after obtaining training sample, data screening, the screening side chosen in the present invention are also carried out to training sample
Method is principle component analysis；For given training sample set H=[VMi|Vxyi]=[hij,hij,…,hV(M+2)]T, training sample is
V × (M+2) matrix；Enable hij'=hij- E (h), H '=[h11′,h12′,…,hV(M+2)'] make center of a sample；Wherein such as formula
(2) shown in:
Shown in the covariance matrix of input sample such as formula (3):
CH=E (H'(H')T) (3)
C is calculated using formula (4)HEigenvalue λ1,λ2,…,λ(M+2)With corresponding feature vector u1,u2,…,u(M+2):
CH*ui=λi*ui, i=1,2 ..., M+2 (4)
Eigenvalue λ might as well be set1≥λ2≥…≥λ(M+2), then ui TIt is exactly projection of the input to feature vector；Enable U=[u1,u2,…,
u(M+2)], then sample set is restructural as shown in formula (5):
Y=UT*H' (5)
(M+2) of input is tieed up by sample set X by eigenvectors matrix U in formula (5) and is transformed to (M+2) in feature space dimension
Sample y in sample set Y, YiA certain component yijX in as XiJ-th of principal component of sample；In order to be selected in feature space
Main component gives up secondary component, defines shown in variance contribution ratio such as formula (6):
。
3. the Wireless Sensor Network Located Algorithm according to claim 1 based on PSO-GA-RBF-HOP, feature exist
In: specific optimization process is as follows in step 5:
1. obtaining the data training sample of neural network, thus really according to the hop count information matrix of dummy node to each anchor node
Determine the input layer of applied RBF neural, the node number of hidden layer, output layer, while determining and neural network is set
The connection weight value range of the data center of Gaussian function, variance and hidden layer to output layer in hidden layer；
2. first having to initialize return-to-zero to the parameters in particle swarm algorithm；According to the actual conditions of neural network, if
Set the number of population and the position and speed coding of each particle；So that it is determined that in particle swarm algorithm particle most university
Habit speed is Vmax, Studying factors are then the random number c between 0 to 11With c2, inertia weight w, maximum cycle mmax、
The permission trueness error value E solved；Determine that the crossover probability in genetic algorithm is P againcIt is P with mutation probabilityv；
3. iterative learning；The position and speed information of each particle is updated according to formula (7), (8), while more according to formula (13)
The fitness value of new particle, formula (13 are):
4. crossover operation；It is selected from population calculating process using the thought of genetic algorithm according to formula (9), the principle of (10)
The preferable particle of partial fitness value out carries out crossover operation two-by-two；Compare the progeny after intersecting and first for the suitable of particle
Response functional value, the good particle of fitness value, which can be put into population, carries out next iteration；
5. mutation operation；Using the thought of genetic algorithm, the particle poor to those fitness function values in population is with probability Pv
Carry out random initializtion；
6. according to step 4. and 5. in mutation operation, angle value and desired positions are preferably adapted to particle each in population
PibestValue carries out optimizing operation；Then angle value and global desired positions g are being preferably adapted to the overall situation in populationbestIt is sought
Excellent operation；
7. if the number of iterations of population has reached the best overall situation in set maximum number of iterations or population
Fitness value reaches set precision, then jumps out loop iteration termination, on the contrary then continue to execute step and 3. continue iteration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510757614.9A CN105554873B (en) | 2015-11-10 | 2015-11-10 | A kind of Wireless Sensor Network Located Algorithm based on PSO-GA-RBF-HOP |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510757614.9A CN105554873B (en) | 2015-11-10 | 2015-11-10 | A kind of Wireless Sensor Network Located Algorithm based on PSO-GA-RBF-HOP |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105554873A CN105554873A (en) | 2016-05-04 |
| CN105554873B true CN105554873B (en) | 2019-03-15 |
Family
ID=55833747
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510757614.9A Expired - Fee Related CN105554873B (en) | 2015-11-10 | 2015-11-10 | A kind of Wireless Sensor Network Located Algorithm based on PSO-GA-RBF-HOP |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105554873B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106060927A (en) * | 2016-08-15 | 2016-10-26 | 江西理工大学 | Location method for wireless sensor network node based on saddle-shaped terrain mountain area |
| CN107124696B (en) * | 2017-03-20 | 2020-05-26 | 东华大学 | Keyless system positioning method based on Bluetooth RSSI |
| CN106993273A (en) * | 2017-03-29 | 2017-07-28 | 江南大学 | Based on distance weighted and genetic optimization DV Hop localization methods |
| CN108366408A (en) * | 2017-04-26 | 2018-08-03 | 南京理工大学 | A kind of DV-Hop location technologies based on neural network |
| CN107528659B (en) * | 2017-09-20 | 2019-05-31 | 复旦大学 | The high-accuracy network topology equivalence algorithm of clock nonsynchronous network |
| CN108989098B (en) * | 2018-08-24 | 2021-06-01 | 福建师范大学 | Time delay optimization-oriented scientific workflow data layout method in hybrid cloud environment |
| CN109799765A (en) * | 2019-02-02 | 2019-05-24 | 中国民航大学 | Steering engine electrohydraulic servo system intelligent control method |
| CN110267279A (en) * | 2019-05-31 | 2019-09-20 | 暨南大学 | Three-dimensional ECDV-Hop localization method based on differential evolution |
| CN111277968B (en) * | 2020-01-20 | 2022-05-03 | 中山大学 | Wireless sensor network non-ranging positioning method based on stack self-encoder |
| CN111432368B (en) * | 2020-03-31 | 2022-10-21 | 中国人民解放军国防科技大学 | Ranging and positioning method suitable for sparse anchor node WSN |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101917762A (en) * | 2010-08-09 | 2010-12-15 | 哈尔滨工程大学 | Node positioning method of particle swarm optimization sensor with penalty function |
| CN102223711A (en) * | 2011-06-23 | 2011-10-19 | 杭州电子科技大学 | Method for positioning wireless sensor network node based on genetic algorithm |
| CN103096338A (en) * | 2011-11-02 | 2013-05-08 | 无锡物联网产业研究院 | Sensor network node location method |
| CN103929810A (en) * | 2014-05-09 | 2014-07-16 | 浙江师范大学 | DV-Hop wireless sensor network node locating method based on wavelet neural network |
-
2015
- 2015-11-10 CN CN201510757614.9A patent/CN105554873B/en not_active Expired - Fee Related
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101917762A (en) * | 2010-08-09 | 2010-12-15 | 哈尔滨工程大学 | Node positioning method of particle swarm optimization sensor with penalty function |
| CN102223711A (en) * | 2011-06-23 | 2011-10-19 | 杭州电子科技大学 | Method for positioning wireless sensor network node based on genetic algorithm |
| CN103096338A (en) * | 2011-11-02 | 2013-05-08 | 无锡物联网产业研究院 | Sensor network node location method |
| CN103929810A (en) * | 2014-05-09 | 2014-07-16 | 浙江师范大学 | DV-Hop wireless sensor network node locating method based on wavelet neural network |
Non-Patent Citations (1)
| Title |
|---|
| 无线传感器网络中DV_Hop算法的改进研究;陈琳;《中国优秀硕士学位论文全文数据库》;20140915;全文 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105554873A (en) | 2016-05-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105554873B (en) | A kind of Wireless Sensor Network Located Algorithm based on PSO-GA-RBF-HOP | |
| Gharghan et al. | Accurate wireless sensor localization technique based on hybrid PSO-ANN algorithm for indoor and outdoor track cycling | |
| CN106131797B (en) | A kind of water-saving irrigation monitoring network localization method based on RSSI ranging | |
| Bae et al. | Large-scale indoor positioning using geomagnetic field with deep neural networks | |
| Arora et al. | Nature inspired range based wireless sensor node localization algorithms | |
| KR102116824B1 (en) | Positioning system based on deep learnin and construction method thereof | |
| Chagas et al. | An approach to localization scheme of wireless sensor networks based on artificial neural networks and genetic algorithms | |
| Payal et al. | Artificial neural networks for developing localization framework in wireless sensor networks | |
| CN106912105A (en) | 3-D positioning method based on PSO_BP neutral nets | |
| CN110062327A (en) | The wireless sensor network node locating method of microhabitat grey wolf optimization DV-Hop algorithm | |
| CN108924755B (en) | DV-HOP indoor positioning method based on immune particle swarm optimization | |
| CN110049433A (en) | A kind of positioning performance optimization method based on EDW-DPSO algorithm | |
| Moussa et al. | Localization of wireless sensor network using bees optimization algorithm | |
| CN104902565A (en) | Distributed wireless sensor network three-dimensional multi-dimensional scaling (MDS) location method | |
| Sulaiman et al. | Towards a better indoor positioning system: A location estimation process using artificial neural networks based on a semi-interpolated database | |
| CN108460462A (en) | A kind of Interval neural networks learning method based on interval parameter optimization | |
| Liu et al. | An indoor WLAN location algorithm based on fingerprint database processing | |
| CN113219408A (en) | Improved RBF neural network indoor visible light positioning method and system | |
| CN107613561A (en) | A kind of method for three-dimensionally positioning network node of wireless sensor based on exchange SCE PSO algorithms | |
| CN105636102B (en) | A kind of localization method and device based on Bayes posterior probability | |
| Sadhukhan et al. | A novel weighted fusion based efficient clustering for improved Wi-Fi fingerprint indoor positioning | |
| CN115550837B (en) | DV-Hop positioning method based on chaotic mapping and gray wolf algorithm optimization | |
| Oh et al. | DNN based WiFi positioning in 3GPP indoor office environment | |
| CN117241215A (en) | Wireless sensor network distributed node cooperative positioning method based on graph neural network | |
| CN108668254B (en) | WiFi signal characteristic area positioning method based on improved BP neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| DD01 | Delivery of document by public notice |
Addressee: Wang Song Document name: Notification of Passing Examination on Formalities |
|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190315 Termination date: 20211110 |