CN110442143A - A kind of unmanned plane situation data clustering method based on combination multiple target dove group's optimization - Google Patents
A kind of unmanned plane situation data clustering method based on combination multiple target dove group's optimization Download PDFInfo
- Publication number
- CN110442143A CN110442143A CN201910603461.0A CN201910603461A CN110442143A CN 110442143 A CN110442143 A CN 110442143A CN 201910603461 A CN201910603461 A CN 201910603461A CN 110442143 A CN110442143 A CN 110442143A
- Authority
- CN
- China
- Prior art keywords
- node
- data
- pigeon
- collection
- dove group
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention discloses a kind of unmanned plane situation data clustering method based on combination multiple target dove group's optimization: load data set；Calculate the difference matrix and adjacency matrix of data set；Minimum spanning tree is solved using Prim algorithm；By the collection when collection is divided into strong ties and Weak link side collection in minimum spanning tree；The precomputation that strong ties side collection is decoded, and is clustered；It generates initial dove group and decodes and obtain cluster result, then using compactness and successive Assessment initial time dove group；Non-dominated ranking is carried out to initial dove group, determines global history optimal location and the center of current time dove group；The position and speed for updating dove group obtains cluster result to dove group's position decoding, updates dove group's global history optimal location and center, persistently the process is until meet termination condition.The inventive method is easy to realize, reduces calculated load, reduces the dimension of decision space, it is easier to search optimal solution, have the ability for preferably adapting to different clustered demands.
Description
Technical field
The present invention relates to a kind of unmanned plane situation data clustering methods based on combination multiple target dove group's optimization, belong to number
According to digging technology field.
Background technique
Unmanned plane has low cost, good economy performance, hang time length, can execute task in the presence of a harsh environment and can keep away
The advantages that exempting from casualties, in power grid inspection, search and rescue aid, civil fields and target reconnaissance, tracking, the strike such as take photo by plane
Etc. military fields suffer from extensive use.Single rack unmanned plane is due to by factors such as sensing range, weapon load, computing capabilitys
Limitation, it is difficult to complete complicated task.Multiple UAVs are shared by the interaction of information, cooperateed in the form of function distribution
The tasks such as search, scouting and strike are executed, system survival rate and overall execution efficiency can be effectively improved.Multiple no-manned plane collaboration is held
During row task, Situation Awareness is the foundation of unmanned plane behaviour decision making, and every frame unmanned plane in unmanned plane cluster requires root
Data mining is carried out to the integrated informations such as the perception information of ambient enviroment and the information for closing on friendly machine received according to it, to it
Knowledge excavation and pattern classification are carried out, to extract valuable knowledge and important information.But multiple unmanned planes are simultaneously
The mode deposited and environmental information it is fast changing so that information content to be processed increases severely, the information processing capability of unmanned plane and
Computing capability is limited.Therefore it is most important to design a kind of rationally efficient data digging method.The present invention is directed to pass through design one
Data clusters analysis method of the kind based on combination multiple target dove group's optimization, improves unmanned plane and handles the energy that large-scale data excavates
Power reduces calculated load, valuable knowledge and important information can be extracted from the data of magnanimity, is conducive to nothing
Man-machine Situation Assessment and behaviour decision making.
Data clusters are a common methods in data mining, belong to unsupervised study.Common data clusters
Method mainly includes the cluster (K- mean value and K- central point algorithm) based on division, (Chameleon is calculated the cluster based on level
Method), density clustering (DBSCAN algorithm) etc..The it is proposed of these methods efficiently solves the clustering problem of data, but
It is the increase with data dimension, performance can correspondingly decline.Further, since data clusters belong to unsupervised study,
Data are not no labels, and different evaluation indexes may result in different cluster knots under conditions of lacking priori knowledge
Fruit.Current clustering method mostly uses single evaluation index to instruct cluster process, such as: compactedness or separability, when given
When different information collection, single evaluation index tends not to all effective to most data set.Therefore, for existing cluster
Method there are the problem of, data clusters problem is converted optimization problem by the present invention, and choose simultaneously multiple may deposit according to demand
In the suitable Cluster Assessment index of mutual the constraint relationship, be then based on combination multiple target dove group optimizing method design one kind can
For the data clusters analysis method of large-scale data, by optimizing multiple evaluation indexes simultaneously to obtain the difference of data set
Feature provides advantageous support to realize that unmanned plane obtains more valuable information from the situation information of complicated magnanimity.
Summary of the invention
The present invention provides a kind of unmanned plane situation data clustering method based on combination multiple target dove group's optimization, mesh
Be to convert optimization problem for data clusters problem, and design combination multiple target dove colony optimization algorithm and solve the problem, from
And realize the clustering of data.
Present invention clustering problem for data provides a kind of unmanned plane state based on combination multiple target dove group's optimization
Gesture data clustering method, the realization block diagram of this method is as shown in Figure 1, main realize that steps are as follows:
Step 1: load unmanned plane situation data set
Data set to be processed is loaded, and calculates the number N of data intensive datadataAnd the dimension M of datadata, useTo indicate i-th of back end in data set.
Step 2: the difference matrix and adjacency matrix of data set are calculated
The purpose of data clusters is that the inwardness and rule according to some standard or data draw the high data of similarity
Same cluster is assigned to, and the lower data of similarity are divided into different clusters.With the Euclidean distance between nodeTo indicate data centralized node i and data
Otherness between node j, Euclidean distance is smaller, and the otherness of two nodes is smaller in data set, is divided into same cluster
Possibility is bigger.Calculate the Euclidean distance d in data set between all nodesij, and it is normalizedThe difference matrix D of node can then be obtaineddataAre as follows:
Ascending sort is carried out by row to the difference matrix as shown in formula (1), obtains the adjacency matrix as shown in formula (2):
Wherein,Indicate according to the otherness between nodes all in data set and node j by it is small to
The number for each adjacent node being ranked up greatly.
Step 3: minimum spanning tree is solved
According to the adjacency matrix as shown in formula (2), minimum spanning tree is solved using Prim algorithm.Random selection one
The node is added in vertex set S=[i] by node i as starting point, selection and the smallest node j of its otherness, then S
=S ∪ [i]=[i, j].In addition, by generation when j → i is added in collection V, V=[j → i].In j → i, by rising for side
Point j has been stored in point set SspIn, terminal i is stored in terminal collection SepIn, Sep(j)=i indicates that node j is connected to node i, and i is
Terminal collection SepElement, and j is known as element i in terminal collection SepIn element index label.The step is repeated, it will not be on vertex
The node of concentration and in vertex set the smallest node of any node otherness successively choose, generate new side, and more
New summit collection, side collection, point set and a terminal collection terminate the step when all nodes in data set are added into vertex set
Suddenly.Successively by terminal collection SepAnd its element index label is attached the minimum spanning tree for then producing data set.
Step 4: by the collection when collection is divided into strong ties and Weak link side collection in minimum spanning tree
According to the adjacency matrix N of formula (3) and data setnearestWith difference matrix DdataCalculate the weight of each side j → i
Value Wji:
Wherein, nn (i, j) is used for which arest neighbors that calculate node j is node i.According to adjacency matrix NnearestIt can
, if nn (i, j)=nik, then it represents that node j is k-th of arest neighbors of node i.Weighted value WjiIt is bigger, indicate node j
A possibility that difference between node i is bigger, is divided into same cluster is smaller.
The weighted value W on all sides of the minimum spanning tree of data set is calculated according to formula (3), and by the corresponding power in these sides
Weight values carry out descending sort, by the biggish λ of weighted value × (Ndata- 1), 0≤λ≤1 collection E when Weak link is addeds, remaining
(1-λ)×(Ndata- 1) the collection E when strong ties are addedw.Point set S is played simultaneouslyspAlso it is correspondingly divided into Weak link and plays point set
SspwPoint set S is played with strong tiessps。
During following steps solve data clusters optimization problem using dove colony optimization algorithm, all strong ties sides
Connection relationship do not change, and Weak link side collection is then rejected, by giving Weak link to play point set SspsIn all sections
Point j redistributes connection endpoint k (k is the adjacent node in the m field of node j), to form new connection relationship j → k
To replace former Weak link side j → i, whole Weak link terminal k to form the position of i-th pigeonIts
Middle NwsFor SspsThe number of interior joint.
Step 5: strong ties side collection is decoded to obtain pre- cluster result
It, can be first by strong ties side since strong ties side collection does not change in subsequent cluster Optimization Solution
Collection decoding obtains the pre- cluster result of data.The point set S from strong tiesspsOne node of middle random selection is as starting point, by institute
There is the node division to link together to same class, introduces group indication vectorIndicate each node in data set
The label for the class being assigned to plays point set S for Weak link if A (2)=3 indicate that node 2 is divided into the 3rd classspwIn section
Point, group indication are initialized as A (i)=- 1, i ∈ Sspw, the step is repeated until all strong ties play point set SspsIn
Node is assigned.
Step 6: initial dove group is generated
It is random to generate psizeA pigeon, every pigeon include spatial positionAnd speedWherein i is the number of pigeon, the spatial position of pigeonIt indicates all
The connection endpoint k on new Weak link side.Different Weak link side collection can be obtained in spatial position by updating pigeon.Setting
Current simulation time is t=0.R is map and compass operator, β1And β2For the random of the Gaussian distributed that is randomly generated
Number, σ is transfer factor, TmaxFor maximum number of iterations,For 1 × NswThe row vector of dimension.
Step 7: the objective function of evaluation t=0 moment dove group
According to the spatial position of pigeonWith terminal collection SepCarry out the spatial position of pigeon
The reconstruct of full coding length, the connection relationship for later pigeon will be reconstructed being decoded between obtaining each node, then according between node
Connection relationship by all class indications be still A (i)=- 1, i ∈ SspwNode be assigned in corresponding class, until all sections
The class indication A (i) ≠ -1, i ∈ S of pointspw.The cluster result that data concentrate all data can be obtained according to group indication vector A
Ci=[c1,c2,…,cγ], wherein γ is the number for being formed by class, cτ, τ=1,2 ..., γ is τ class.
Two Cluster Assessment indexs (objective function) of compactness and continuity are selected to evaluate the quality of cluster result, are used
Clustering distance characterizes continuity (the objective function f of cluster in class1):
Compactness (the objective function f of cluster is characterized with clustering distance in class2).Every one kind is calculated according to formula (5) first
In each node to cluster centre average distance.
Wherein, cdτIndicate average distance of each node to cluster centre, c in τ classτFor the node collection of τ class, | cτ
| it is the number of nodes that τ class includes,For the cluster centre of τ class.This step is repeated until calculating
The corresponding target function value of all pigeons in dove group, wherein pigeon i represents ith cluster result C=[c1,c2,…,cγ],
Its corresponding target function value is Fi(t)=[fi1(t),fi2(t)]。
Step 8: to initial dove, group carries out non-dominated ranking, determine current time dove group global history optimal location and
Center
The dominance relation for comparing pigeon i Yu pigeon j based on target function value passes through Pareto (Pareto) non-dominant row
Sequence algorithm is layered entire dove group.If all target function value F of pigeon ii(t)=[fi1(t),fi2(t)] it is superior to
The target function value F of pigeon jj(t)=[fj1(t),fj2(t)], i.e. fi1(t)≤fj1(t) and fi2(t)≤fj2(t), then claim dove
Sub- i dominates pigeon j, if pigeon i is not dominated by other pigeons, which is referred to as non-dominant pigeon, is divided into first
The non-dominant layer of grade.
All non-dominant pigeons positioned at the non-dominant layer of the first order are saved in external archival collection AS, from external archival
Collect and randomly choose a pigeon in AS, as the global optimum position p of t momentbest(t), by position in external archival collection AS
In the non-dominant layer of the first order all pigeons mean place as pcenter(t)。
Step 9: the position and speed of pigeon is updated
Introduce auxiliary vector ζi=[ζi1,ζi2,…,ζiNws], ζij∈ [- 1,0,1] is converted into continuous dove colony optimization algorithm
Combinatorial optimization algorithm allows to for solving cluster optimization problem:
Wherein, pcenterjIt (t) is t moment dove group center position pcenter(t) j-th of element, pgbestjIt (t) is t moment
The history optimal location p of dove groupgbest(t) j-th of element.The speed more new formula of pigeon can be obtained according to formula (7) are as follows:
Wherein, β1And β2Random number for the Gaussian distributed being randomly generated,For 1 × NswThe row vector of dimension.
Then according to the speed v of t+1 moment pigeoni(t+1) ζ is calculatedi(t+1):
Wherein, δ is scheduled constant, by updated ζi(t+1) formula (10) Lai Gengxin pigeon is substituted at the t+1 moment
Position:
Wherein, λ is randomly selected pij(t) arest neighbors.The step is repeated until the position and speed of all pigeons updates
It completes.
Step 10: the fitness function of assessment t+1 moment dove group
Pigeon is decoded according to step 6, forms cluster, and calculates the fitness function F of t+1 moment dove groupi(t+
1)=[fi1(t+1),fi2(t+1)].Compare pigeon p based on target function valuei(t+1) with pigeon pj(t+1) dominance relation,
All non-dominant pigeons positioned at the non-dominant layer of the first order are saved in external archival collection AS.
Step 11: non-dominated ranking is carried out to external archival collection AS, and selects to need in AS to give up according to crowding distance
The pigeon of abandoning
Non-dominated ranking is carried out to all pigeons saved in the external archival collection AS at t+1 moment, it is non-to give up the non-first order
The non-dominant pigeon of layer is dominated, and calculates the crowding distance of the non-dominant pigeon of the non-dominant layer of the first order, it is big to give up crowding distance
Pigeon.
Step 12: updating the global optimum position and center of pigeon, and updates dove group's quantity
A pigeon is randomly choosed from external archival collection AS, using its position as the global optimum position at t+1 moment
pbest(t+1), using the center of the non-dominant pigeon of the non-dominant layer of the first order as the center p of dove groupcenter(t+
1).In each iterative process, the number of pigeon can be gradually reduced.
Psize(t+1)=Psize(t)-Pdec (11)
Wherein, PdecFor the pigeon number given up.
Step 13: judge whether to stop iteration
Iteration of simulations number t=t+1.If t is greater than maximum iteration of simulations number Tmax, then emulation terminates, and enters step ten
Four；Otherwise, return step eight.
Step 14: output data cluster result
Cluster result is exported, and draws the forward position Pareto curve.
The invention proposes a kind of unmanned plane situation data clustering methods based on combination multiple target dove group's optimization, pass through
Optimization problem is converted by data clusters problem, then design combination multiple target dove colony optimization algorithm carrys out solving optimization problem, can
For solving the data analysis problems of large-scale data.The advantage of clustering method proposed by the invention is mainly reflected in: first
First, pigeon position (cluster result) encoding mechanism effectively reduces the calculated load of large-scale data, reduces decision space
Dimension, so that combination multiple target dove colony optimization algorithm is easier to search optimal solution (optimal cluster result)；Secondly, set
The auxiliary vector of meter effectively converts combinatorial optimization algorithm for continuous dove colony optimization algorithm, has original dove colony optimization algorithm
The standby ability for solving discrete optimization problems of device, has widened the application field of dove colony optimization algorithm；Finally, in the process optimized
In, while compactness and continuity the two Cluster Assessment indexs are considered, can obtain and preferably adapt to different clusters and need
The ability asked, so that the different characteristic of data set is obtained, to realize that unmanned plane obtains more from the situation information of complicated magnanimity
Valuable information advantageous support is provided.
Detailed description of the invention
Data clusters analysis flow chart diagram of the Fig. 1 based on combination multiple target dove group's optimization
Fig. 2 a, b, c, d cluster the position encoded figure of pigeon, wherein Fig. 2 a is minimum spanning tree and its expression of data set；
Fig. 2 b is to be ranked up according to the weighted value of connection to the side of minimum spanning tree；Fig. 2 c is to remove Weak link side collection, by strong ties
Side collection generates the pre- cluster result of data；Fig. 2 d is to generate new connection (storing connection endpoint in pigeon position) to replace Weak link side.
Fig. 3 clusters pigeon position decoding figure
The cluster result figure and the forward position Pareto curve of Fig. 4 a-e data set 1
The cluster result figure and the forward position Pareto curve of Fig. 5 a-f data set 2
The cluster result figure and the forward position Pareto curve of Fig. 6 a-f data set 3
Figure label and symbol description are as follows:
T --- iteration of simulations number
psize--- the sum of pigeon in dove group
The label of i --- pigeon
Tmax--- maximum iteration of simulations number
U --- terminal corresponding to Weak link side starting point to be determined
f1--- objective function indicates Cluster Assessment index continuity
f2--- objective function indicates Cluster Assessment index compactness
The abscissa of X --- two-dimentional data set
The ordinate of Y --- two-dimentional data set
A --- different clustering target value (f1, f2) corresponding cluster result
B --- different clustering target value (f1, f2) corresponding cluster result
C --- different clustering target value (f1, f2) corresponding cluster result
D --- different clustering target value (f1, f2) corresponding cluster result
Specific embodiment
The validity of method proposed by the invention is verified below by specific data clusters example.This method
Specific step is as follows:
Step 1: load unmanned plane situation data set
Data set to be processed is loaded, clustering is carried out to three different types of data sets altogether, with first number
It is illustrated according to collection.The data set includes Ndata=12 data, are respectively as follows: x1=[1,1], x2=[1,1.2], x3=
[1.2,1]、x4=[1.2,1.2], x5=[2,2], x6=[2.2,2], x7=[2,2.2], x8=[2.2,2.2], x9=[3,
1]、x10=[3.2,1], x11=[3,1.2], x12=[3.2,1.2], the dimension M of datadata=2.
Step 2: the difference matrix and adjacency matrix of data set are calculated
Calculate the Euclidean distance d in data set between all nodesij=| | xi-xj| |, and it is normalizedNode difference matrix D can then be obtaineddataAre as follows:
To difference matrix DdataAscending sort is carried out by row, adjacency matrix N can be obtainednearestAre as follows:
Its every a line indicate the otherness in data set between first node of all nodes and the row by it is ascending into
Row sequence.
Step 3: minimum spanning tree is solved
According to adjacency matrix Nnearest, minimum spanning tree is solved using Prim algorithm.Random selection node 5 is used as
Initial point, selection are added in vertex set S=[5] with the smallest node 6 of its otherness, then S=S ∪ [6]=[5,6].Side collection V
=[6 → 5].Play point set SspIn=[5,6], terminal collection SepSep(6)=5.The step is repeated, by the section not in vertex set
Point and in vertex set the smallest node of any node otherness successively choose, generate new side, and more new summit
Collection, side collection, point set and a terminal collection terminate the step when all nodes in data set are added into vertex set, at this point,
Playing point set is Ssp=[5,6,7,8,4,2,3,1,11,9,12,10], terminal integrate as Sep=[2,4,4,5,5,5,5,6,11,
12,6,11].According to SepAnd its element index label (i.e. the start node on each side) can obtain connection relationship are as follows: node 1 →
Node 2,2 → node of node 4,3 → node of node 4,4 → node of node 5,5 → node of node 5,6 → node of node 5, node 7
→ node 5,8 → node of node 6,9 → node of node 11,10 → node of node 12,11 → node of node 6,12 → node of node
11, the minimum spanning tree generated is as shown in Figure 2 a.
Step 4: calculating the weighted value on each side in minimum spanning tree, according to weighted value by the side in minimum spanning tree
Collection is divided into strong ties collection in collection and Weak link
According to formula (3) and adjacency matrix NnearestAnd difference matrix DdataCalculate the weighted value W of each side j → iji。
As shown in Figure 2 a, each side (by point set S of minimum spanning treesp=[5,6,7,8,4,2,3,1,11,9,12,10], end
Point set Sep=[2,4,4,5,5,5,5,6,11,12,6,11] and terminal collection element index Sep_index=[1,2,3,4,5,6,7,
8,9,10,11,12] indicate) corresponding weighted value be W=[2.1,0,2.1,2.1,5.5,2.1,3.1,2.1,5.5,2.1,
3.1,2.1].By by all weighted value WjiAfter descending sort selection, λ=0.4 is taken, then the side of minimum spanning tree can be drawn
It is divided into the strong ties as shown in Figure 2 b collection in collection and Weak link.Wherein, it is S that Weak link, which plays point set,spw=[3,4,11,12],
Weak link side integrates as Es={ 3 → 4,4 → 5,11 → 6,12 → 11 }, it is S that strong ties, which play point set,sps=[1,2,6,7,8,9,
10], strong ties side integrates as Ew={ 1 → 2,2 → 4,6 → 5,7 → 5,8 → 6,9 → 11,10 → 12 }.By Weak link edge contract
Afterwards, terminal integrates variation as Sep=[2,4, U, U, 5,5,5,6,11,12, U, U], wherein U indicates Weak link starting point to be confirmed
Collect SspwThe connection endpoint of=[3,4,11,12], such as Fig. 2 d.
Step 5: strong ties side collection is decoded to obtain pre- cluster result
The point set S from strong tiesspsOne node of random selection is as starting point in=[1,2,6,7,8,9,10], by institute
There is the node division to link together to same class, introduces group indication vectorIndicate each node in data set
The label for the class being assigned to plays point set S for Weak link if A (2)=3 indicate that node 2 is divided into the 3rd classspwIn section
Point, group indication are initialized as A (i)=- 1, i ∈ Sspw, the step is repeated until all strong ties play point set SspsIn
Node is assigned.The result clustered in advance is as shown in Figure 2 c.
Step 6: initial dove group is generated
It is random to generate psize=100 pigeons, every pigeon include spatial positionAnd speedWherein NwsFor the number of Weak link side starting point centralized node, and Nws=4, i are the number of pigeon.
Setting current simulation time is t=0, maximum number of iterations Tmax=50, map and compass operator are R=0.3, transfer because
Son is σ=0.45, threshold value δ=3.6.
Step 7: the objective function of evaluation t=0 moment dove group
By pi=[pi1,pi2,pi3,pi4] and terminal collection Sep=[2,4, U, U, 5,5,5,6,11,12, U, U] carries out carry out dove
The full coding length reconstruct of the spatial position of son can obtain Sep=[2,4, pi1,pi2,5,5,5,6,11,12,pi3,pi4], by its into
Row decoding can obtain node connection relationship as shown in Figure 3, then still be by all class indications according to the connection relationship between node
A (i)=- 1, i ∈ SspwNode be assigned in corresponding class class indication A (i) ≠ -1, i ∈ S until all nodesspw。
The cluster result C that data concentrate all data can be obtained according to group indication vector Ai=[c1,c2,…,cγ], wherein γ is institute's shape
At class number, cτ, τ=1,2 ..., γ is τ class.
According to cluster result CiWith formula (4) to formula (6) calculating target function fi1(cluster continuity parameter, i-th pigeon
Corresponding objective function f1) and objective function fi2(cluster compactness index, the corresponding objective function f of i-th pigeon2), it repeats
This step is until calculate the corresponding target function value F of all pigeons in dove groupi(t)=[fi1(t),fi2(t)]。
Step 8: non-dominated ranking is carried out to initial dove group
Compare pigeon p based on target function valuei(t) with pigeon pj(t) dominance relation passes through Pareto non-dominated ranking
Algorithm is layered entire dove group.All non-dominant pigeons positioned at the non-dominant layer of the first order are saved in external archival collection
In AS, a pigeon is randomly choosed from external archival collection AS, as the global optimum position p of t momentbest(t), will
Center of the mean place of the non-dominant pigeon of the non-dominant layer of the first order as t moment dove group in external archival collection AS
pcenter(t)。
Step 9: the position and speed of pigeon is updated
By pbest(t)、pcenter(t) and the position of all pigeons of t moment substitutes into formula (7) and is calculated and pigeon position
pi(t) corresponding auxiliary vector ζi(t) then value is updated according to the speed of pigeon more new formula (8), location assistance vector
Formula (9) and location update formula (10) can calculate the position p of t+1 speed pigeoni(t+1) and speed vi(t+1).Repeating should
Step is until the position and speed of all pigeons is completed to update.
Step 10: the target function value of assessment t+1 moment dove group
Pigeon is decoded according to step 6, forms cluster, and calculates the fitness function F of t+1 moment dove groupi(t+
1)=[fi1(t+1),fi2(t+1)].Compare pigeon p based on target function valuei(t+1) with pigeon pj(t+1) dominance relation,
All non-dominant pigeons positioned at the non-dominant layer of the first order are saved in external archival collection AS.
Step 11: non-dominated ranking is carried out to external archival collection AS, and selects to need in AS to give up according to crowding distance
The pigeon of abandoning
Non-dominated ranking is carried out to all pigeons saved in the external archival collection AS at t+1 moment, it is non-to give up the non-first order
The non-dominant pigeon of layer is dominated, and calculates the crowding distance of the non-dominant pigeon of the non-dominant layer of the first order, it is big to give up crowding distance
Pigeon.
Step 12: updating the global optimum position and center of pigeon, and updates dove group's quantity
A pigeon is randomly choosed from external archival collection AS, using its position as the global optimum position at t+1 moment
pbest(t+1), using the center of the non-dominant pigeon of the non-dominant layer of the first order as the center p of dove groupcenter(t+
1).The number P of pigeon is updated according to formula (11)size(t+1)。
Step 13: judge whether to stop iteration
Iteration of simulations number t=t+1.If t is greater than maximum iteration of simulations number Tmax=50, then emulation terminates；Otherwise, it returns
Return step 8.
Step 14: output data cluster result
Cluster result is exported, draws the forward position Pareto curve, the cluster result of three different data sets is respectively such as Fig. 4
To shown in Fig. 6.Fig. 4 a is the given cluster result of data set 1, and Fig. 4 e is to be obtained using combination multiple target dove colony optimization algorithm
The forward position the cluster Pareto curve of data set 1 indicates the set of cluster objective function optimal solution, according to the requirement of clustering target,
Available different optimal data cluster result.Fig. 4 b is data clusters knot corresponding to A point on Pareto curve in Fig. 4 e
Fruit, Fig. 4 c be in Fig. 4 e on Pareto curve data clusters corresponding to B point as a result, Fig. 4 d is C on Pareto curve in Fig. 4 e
The corresponding data clusters result of point.Fig. 5 a is the given cluster result of data set 2, and Fig. 5 f is using combination multiple target dove group
The forward position the cluster Pareto curve for the data set 2 that optimization algorithm obtains.Fig. 5 b, Fig. 5 c, Fig. 5 d and Fig. 5 e are respectively in Fig. 5 f
A point on Pareto curve, B point, data clusters result corresponding to four points of C point and D point.Fig. 6 a is the given poly- of data set 3
Class is as a result, Fig. 6 f is the forward position the cluster Pareto curve of the data set 3 obtained using combination multiple target dove colony optimization algorithm.Figure
6b, Fig. 6 c, Fig. 6 d and Fig. 6 e are respectively A point on Pareto curve in Fig. 6 f, B point, data corresponding to four points of C point and D point
Cluster result.
Demonstrated by the cluster result of three different types of data sets proposed through the invention based on combine it is more
The unmanned plane situation data clustering method of target dove group's optimization can effectively realize the clustering of data.
Claims (9)
1. a kind of unmanned plane situation data clustering method based on combination multiple target dove group's optimization, it is characterised in that: this method step
It is rapid as follows:
Step 1: load unmanned plane situation data set
Data set to be processed is loaded, and calculates the number N of data intensive datadataAnd the dimension M of datadata, useTo indicate i-th of back end in data set；
Step 2: the difference matrix and adjacency matrix of data set are calculated；
Step 3: minimum spanning tree is solved
According to adjacency matrix, minimum spanning tree is solved using Prim algorithm；
Step 4: by the collection when collection is divided into strong ties and Weak link side collection in minimum spanning tree
Step 5: strong ties side collection is decoded to obtain pre- cluster result；
Step 6: initial dove group is generated
It is random to generate psizeA pigeon, every pigeon include spatial position and speed；
Step 7: the objective function of evaluation t=0 moment dove group；
Step 9: the position and speed of pigeon is updated；
Step 8: carrying out non-dominated ranking to initial dove group, determines global history optimal location and the center of current time dove group
Position；
Step 10: the fitness function of assessment t+1 moment dove group
Pigeon is decoded according to step 6, forms cluster, and calculates the fitness function of t+1 moment dove group；Based on target
Functional value compares pigeon pi(t+1) with pigeon pj(t+1) dominance relation, by all positioned at the non-dominant of the non-dominant layer of the first order
Pigeon is saved in external archival collection AS；
Step 11: carrying out non-dominated ranking to external archival collection AS, and selects to need to give up in AS according to crowding distance
Pigeon；
Step 12: updating the global optimum position and center of pigeon, and updates dove group's quantity；
Step 13: judge whether to stop iteration
Iteration of simulations number t=t+1；If t is greater than maximum iteration of simulations number Tmax, then emulation terminates, and enters step 14；It is no
Then, return step eight；
Step 14: output data cluster result
Cluster result is exported, and draws the forward position Pareto curve.
2. a kind of unmanned plane situation data clustering method based on combination multiple target dove group's optimization according to claim 1,
It is characterized by: detailed process is as follows for the step 2: with the Euclidean distance between nodeTo indicate data centralized node i and data
Otherness between node j；Calculate the Euclidean distance d in data set between all nodesij, and it is normalizedThe difference matrix D of node can then be obtaineddataAre as follows:
Ascending sort is carried out by row to the difference matrix as shown in formula (1), obtains the adjacency matrix as shown in formula (2):
Wherein,Indicate according to the otherness between nodes all in data set and node j it is ascending into
The number for each adjacent node that row sequence obtains.
3. according to a kind of unmanned plane situation data clusters side based on combination multiple target dove group's optimization according to claim 1
Method, it is characterised in that: detailed process is as follows for the step 3:
A node i is randomly choosed as starting point, selection and the smallest node j of its otherness, and the node is added to top
In point set S=[i], then S=S ∪ [i]=[i, j]；In addition, by generation when j → i is added in collection V, V=[j → i]；In
In j → i, the starting point j on side point set S has been stored inspIn, terminal i is stored in terminal collection SepIn, Sep(j)=i indicates that node j connects
It is connected to node i, i is terminal collection SepElement, and j is known as element i in terminal collection SepIn element index label；Repeat the step
Suddenly, by the node not in vertex set and in vertex set the smallest node of any node otherness successively choose, it is raw
The side of Cheng Xin, and update vertex set, Bian Ji, play point set and terminal collection, when all nodes in data set are added into vertex set
When, terminate the step；Successively by terminal collection SepAnd its element index label is attached the minimum generation for then producing data set
Tree.
4. according to a kind of unmanned plane situation data clusters side based on combination multiple target dove group's optimization according to claim 1
Method, it is characterised in that: detailed process is as follows for the step 4: according to the adjacency matrix N of formula (3) and data setnearestAnd difference
Different matrix DdataCalculate the weighted value W of each side j → iji:
Wherein, nn (i, j) is used for which arest neighbors that calculate node j is node i；According to adjacency matrix NnearestIt can obtain, if
Nn (i, j)=nik, then it represents that node j is k-th of arest neighbors of node i；
The weighted value W on all sides of the minimum spanning tree of data set is calculated, and the corresponding weighted value in these sides is subjected to descending row
Sequence, by the biggish λ of weighted value × (Ndata- 1), 0≤λ≤1 collection E when Weak link is addeds, remaining (1- λ) × (Ndata-1)
The collection E when strong ties are addedw；Point set S is played simultaneouslyspAlso it is correspondingly divided into Weak link and plays point set SspwWith strong ties starting point
Collect Ssps。
5. according to a kind of unmanned plane situation data clusters side based on combination multiple target dove group's optimization according to claim 4
Method, it is characterised in that: it is further, point set S is played to Weak linkspsIn all node j redistribute connection endpoint k, k is section
Adjacent node in the m field of point j, to form new connection relationship j → k to replace former Weak link side j → i, whole is weak
Connection endpoint k forms the position of i-th pigeonWherein NwsFor SspsThe number of interior joint.
6. according to a kind of unmanned plane situation data clusters side based on combination multiple target dove group's optimization according to claim 1
Method, it is characterised in that: detailed process is as follows for the step 5: the point set S from strong tiesspsMiddle one node of random selection is made
All node divisions to link together to same class are introduced into group indication vector for starting pointIndicate data
The label for the class for concentrating each node to be assigned to plays point set S for Weak linkspwIn node, group indication initializes
For A (i)=- 1, i ∈ Sspw, the step is repeated until all strong ties play point set SspsIn node be assigned.
7. according to a kind of unmanned plane situation data clusters side based on combination multiple target dove group's optimization according to claim 1
Method, it is characterised in that: detailed process is as follows for the step 6:
According to the spatial position of pigeonWith terminal collection SepCarry out the full volume of the spatial position of pigeon
Code length reconstruct, the connection relationship for later pigeon will be reconstructed being decoded between obtaining each node, then according to the company between node
It is still A (i)=- 1, i ∈ S that relationship, which is connect, by all class indicationsspwNode be assigned in corresponding class, until all nodes
Class indication A (i) ≠ -1, i ∈ Sspw；The cluster result C that data concentrate all data can be obtained according to group indication vector Ai=
[c1,c2,…,cγ], wherein γ is the number for being formed by class, cτ, τ=1,2 ..., γ is τ class；
Two Cluster Assessment indexs of compactness and continuity are selected to evaluate the quality of cluster result, with clustering distance in class come table
Levy the continuity of cluster:
The compactness of cluster is characterized with clustering distance in class；First according to each node in the every one kind of formula (5) calculating to cluster
The average distance at center；
Wherein, cdτIndicate average distance of each node to cluster centre, c in τ classτFor the node collection of τ class, | cτ| it is the
The number of nodes that τ class includes,For the cluster centre of τ class；This step is repeated until calculating in dove group
The corresponding target function value of all pigeons, wherein pigeon i represents ith cluster result C=[c1,c2,…,cγ], it is corresponding
Target function value is Fi(t)=[fi1(t),fi2(t)]。
8. according to a kind of unmanned plane situation data clusters side based on combination multiple target dove group's optimization according to claim 1
Method, it is characterised in that: detailed process is as follows for the step 8:
The dominance relation for comparing pigeon i Yu pigeon j based on target function value, by Pareto non-dominated ranking algorithm to entire dove
Group is layered；All non-dominant pigeons positioned at the non-dominant layer of the first order are saved in external archival collection AS, are deposited from outside
A pigeon is randomly choosed in shelves collection AS, as the global optimum position p of t momentbestIt (t), will be in external archival collection AS
Positioned at the non-dominant layer of the first order all pigeons mean place as pcenter(t)。
9. according to a kind of unmanned plane situation data clusters side based on combination multiple target dove group's optimization according to claim 1
Method, it is characterised in that: detailed process is as follows for the step 9: auxiliary vector is introducedζij∈[-
1,0,1] continuous dove colony optimization algorithm is converted into combinatorial optimization algorithm, allowed to for solving cluster optimization problem:
Wherein, pcenterjIt (t) is t moment dove group center position pcenter(t) j-th of element, pgbestj(t) for t moment dove group's
History optimal location pgbest(t) j-th of element；The speed more new formula of pigeon can be obtained according to formula (7) are as follows:
Wherein, β1And β2Random number for the Gaussian distributed being randomly generated,For 1 × NswThe row vector of dimension；
Then according to the speed v of t+1 moment pigeoni(t+1) ζ is calculatedi(t+1):
Wherein, δ is scheduled constant, by updated ζi(t+1) formula (10) Lai Gengxin pigeon is substituted into the position at t+1 moment:
Wherein, λ is randomly selected pij(t) arest neighbors；The step is repeated until the position and speed of all pigeons is updated and completed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910603461.0A CN110442143B (en) | 2019-07-05 | 2019-07-05 | Unmanned aerial vehicle situation data clustering method based on combined multi-target pigeon swarm optimization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910603461.0A CN110442143B (en) | 2019-07-05 | 2019-07-05 | Unmanned aerial vehicle situation data clustering method based on combined multi-target pigeon swarm optimization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110442143A true CN110442143A (en) | 2019-11-12 |
| CN110442143B CN110442143B (en) | 2020-10-27 |
Family
ID=68429412
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910603461.0A Active CN110442143B (en) | 2019-07-05 | 2019-07-05 | Unmanned aerial vehicle situation data clustering method based on combined multi-target pigeon swarm optimization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110442143B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111367315A (en) * | 2020-03-11 | 2020-07-03 | 北京邮电大学 | Trajectory planning method and device applied to information collection of unmanned aerial vehicle |
| CN112596574A (en) * | 2020-12-21 | 2021-04-02 | 广东工业大学 | Photovoltaic maximum power tracking method and device based on layered pigeon swarm algorithm |
| CN113075648A (en) * | 2021-03-19 | 2021-07-06 | 中国舰船研究设计中心 | Clustering and filtering method for unmanned cluster target positioning information |
| CN113110595A (en) * | 2021-05-12 | 2021-07-13 | 中国人民解放军陆军工程大学 | Heterogeneous unmanned aerial vehicle group cooperation method for target verification |
| CN113741482A (en) * | 2021-09-22 | 2021-12-03 | 西北工业大学 | Multi-agent path planning method based on asynchronous genetic algorithm |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102983948A (en) * | 2012-12-19 | 2013-03-20 | 山东黄金集团有限公司 | Adaptive clustering transmission method and device for wireless sensor network |
| CN103440275A (en) * | 2013-08-08 | 2013-12-11 | 南京邮电大学 | Prim-based K-means clustering method |
| CN103699933A (en) * | 2013-12-05 | 2014-04-02 | 北京工业大学 | Traffic signal timing optimization method based on minimum spanning tree clustering genetic algorithm |
| CN103985112A (en) * | 2014-03-05 | 2014-08-13 | 西安电子科技大学 | Image segmentation method based on improved multi-objective particle swarm optimization and clustering |
| CN105066998A (en) * | 2015-08-03 | 2015-11-18 | 北京航空航天大学 | Quantum-behaved pigeon inspired optimization-based unmanned aerial vehicle autonomous aerial refueling target detection method |
| CN105654500A (en) * | 2016-02-01 | 2016-06-08 | 北京航空航天大学 | Unmanned aerial vehicle target detection method for optimizing visual attention mechanism based on bionic pigeons |
| CN108256553A (en) * | 2017-12-20 | 2018-07-06 | 中国人民解放军国防科技大学 | Construction method and device for double-layer path of vehicle-mounted unmanned aerial vehicle |
-
2019
- 2019-07-05 CN CN201910603461.0A patent/CN110442143B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102983948A (en) * | 2012-12-19 | 2013-03-20 | 山东黄金集团有限公司 | Adaptive clustering transmission method and device for wireless sensor network |
| CN103440275A (en) * | 2013-08-08 | 2013-12-11 | 南京邮电大学 | Prim-based K-means clustering method |
| CN103699933A (en) * | 2013-12-05 | 2014-04-02 | 北京工业大学 | Traffic signal timing optimization method based on minimum spanning tree clustering genetic algorithm |
| CN103985112A (en) * | 2014-03-05 | 2014-08-13 | 西安电子科技大学 | Image segmentation method based on improved multi-objective particle swarm optimization and clustering |
| CN105066998A (en) * | 2015-08-03 | 2015-11-18 | 北京航空航天大学 | Quantum-behaved pigeon inspired optimization-based unmanned aerial vehicle autonomous aerial refueling target detection method |
| CN105654500A (en) * | 2016-02-01 | 2016-06-08 | 北京航空航天大学 | Unmanned aerial vehicle target detection method for optimizing visual attention mechanism based on bionic pigeons |
| CN108256553A (en) * | 2017-12-20 | 2018-07-06 | 中国人民解放军国防科技大学 | Construction method and device for double-layer path of vehicle-mounted unmanned aerial vehicle |
Non-Patent Citations (4)
| Title |
|---|
| QIU HUAXIN.ECT: "《Multi-objective pigeon-inspired optimization for brushless direct current motor parameter design》", 《SCIENCE CHINA TECHNOLOGICAL SCIENCES》 * |
| 叶军伟: "《普里姆算法和克鲁斯卡尔算法构造最小生成树》", 《河南科技》 * |
| 吴登磊等: "《基于欧氏距离的K均方聚类算法研究与应用》", 《数字技术与应用》 * |
| 王晓柱等: "《最小生成树的prim算法及minimum函数》", 《山东轻工业学院学报》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111367315A (en) * | 2020-03-11 | 2020-07-03 | 北京邮电大学 | Trajectory planning method and device applied to information collection of unmanned aerial vehicle |
| CN111367315B (en) * | 2020-03-11 | 2021-06-11 | 北京邮电大学 | Trajectory planning method and device applied to information collection of unmanned aerial vehicle |
| CN112596574A (en) * | 2020-12-21 | 2021-04-02 | 广东工业大学 | Photovoltaic maximum power tracking method and device based on layered pigeon swarm algorithm |
| CN112596574B (en) * | 2020-12-21 | 2022-06-24 | 广东工业大学 | Photovoltaic maximum power tracking method and device based on layered pigeon swarm algorithm |
| CN113075648A (en) * | 2021-03-19 | 2021-07-06 | 中国舰船研究设计中心 | Clustering and filtering method for unmanned cluster target positioning information |
| CN113075648B (en) * | 2021-03-19 | 2024-05-17 | 中国舰船研究设计中心 | Clustering and filtering method for unmanned cluster target positioning information |
| CN113110595A (en) * | 2021-05-12 | 2021-07-13 | 中国人民解放军陆军工程大学 | Heterogeneous unmanned aerial vehicle group cooperation method for target verification |
| CN113110595B (en) * | 2021-05-12 | 2022-06-21 | 中国人民解放军陆军工程大学 | Heterogeneous unmanned aerial vehicle group cooperation method for target verification |
| CN113741482A (en) * | 2021-09-22 | 2021-12-03 | 西北工业大学 | Multi-agent path planning method based on asynchronous genetic algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110442143B (en) | 2020-10-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110442143A (en) | A kind of unmanned plane situation data clustering method based on combination multiple target dove group's optimization | |
| CN109815886A (en) | A kind of pedestrian and vehicle checking method and system based on improvement YOLOv3 | |
| CN102413029B (en) | Method for partitioning communities in complex dynamic network by virtue of multi-objective local search based on decomposition | |
| CN109063911A (en) | A kind of Load aggregation body regrouping prediction method based on gating cycle unit networks | |
| CN105488528A (en) | Improved adaptive genetic algorithm based neural network image classification method | |
| CN109102124B (en) | Dynamic multi-target multi-path induction method and system based on decomposition and storage medium | |
| CN108573303A (en) | It is a kind of that recovery policy is improved based on the complex network local failure for improving intensified learning certainly | |
| CN107133695A (en) | A kind of wind power forecasting method and system | |
| CN101482876B (en) | Weight-based link multi-attribute entity recognition method | |
| CN102262702B (en) | Decision-making method for maintaining middle and small span concrete bridges | |
| CN102024179A (en) | Genetic algorithm-self-organization map (GA-SOM) clustering method based on semi-supervised learning | |
| CN103324954A (en) | Image classification method based on tree structure and system using same | |
| CN110070116A (en) | Segmented based on the tree-shaped Training strategy of depth selects integrated image classification method | |
| CN110322075A (en) | A kind of scenic spot passenger flow forecast method and system based on hybrid optimization RBF neural | |
| CN108573274A (en) | A kind of selective clustering ensemble method based on data stability | |
| CN110070228A (en) | BP neural network wind speed prediction method for neuron branch evolution | |
| CN110020712A (en) | A kind of optimization population BP neural network forecast method and system based on cluster | |
| CN109582714A (en) | A kind of government affairs item data processing method based on time fading correlation | |
| CN109902808A (en) | A method of convolutional neural networks are optimized based on floating-point numerical digit Mutation Genetic Algorithms Based | |
| CN115912502A (en) | Intelligent power grid operation optimization method and device | |
| CN112200391B (en) | Power distribution network edge side load prediction method based on k-nearest neighbor mutual information feature simplification | |
| CN112711985B (en) | Fruit identification method and device based on improved SOLO network and robot | |
| CN106919783B (en) | A kind of multiple target degree of association division processing method of buoy data | |
| CN109116300A (en) | A kind of limit learning position method based on non-abundant finger print information | |
| CN117458480A (en) | Photovoltaic power generation power short-term prediction method and system based on improved LOF |
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 |