CN101448267A - Wireless sensor network node coverage optimization method based on particle swarm algorithm - Google Patents

Abstract

A particle swarm algorithm is applied to an optimum coverage problem of wireless sensor network nodes, and the invention provides a coverage mechanism of the particle swarm optimization algorithm based on a discrete binary edition, and the mechanism is used for performing optimal solution on the wireless sensor network node coverage problem. The method comprises the following steps: defining the wireless sensor network node coverage problem as a 0/1 planning problem, encoding individuals of a binary particle swarm algorithm as a 0/1 binary string and then performing optimization by an evolutionism of the particle swarm algorithm. In the method, two important indexes are defined to evaluate the solution results, one index is 'coverage' of a region, and the other index is 'consumption rate' of the sensor. Validity and high efficiency of the algorithm are verified by a simulation experiment.

Description

Wireless sensor network node coverage optimization method based on particle cluster algorithm

Technical field:

The present invention relates to wireless sensor technology and intelligence computation two big fields, be specifically related to a kind of wireless sensor network node coverage optimization method based on particle cluster algorithm.

Technical background:

Since the nineties in 20th century, fast development along with technology such as embedded system, radio communication, network and microelectromechanical systemss, (Wireless Sensor Networks WSN) has caused people's very big concern to have the wireless sensor network of perception, calculating and wireless communication ability.Wireless sensor network is a kind of emerging technology for information acquisition, is a kind ofly can independently finish appointed task according to environment, has the intelligent network system of cognitive ability.It by be deployed in a large amount of cheapnesss in the induction region, the sensor node of low-power consumption constitutes.By various informative transducer built-in in the node, object information in sensor node perception synergistically, communication and the processing network's coverage area, can also carry out simple calculation task, implement route and other control strategies, can be at any time, obtain a large amount of accurate and reliable information under most place and the multiple environmental condition.These outstanding characteristics are that wireless sensor network has been given wide application space and development prospect.As a new field in the information technology, wireless sensor network has all been obtained in the military and civilian field widely and has been used, as military surveillance, environmental monitoring, medical monitoring, space exploration, urban traffic control, warehousing management etc.Wireless sensor network is considered to indispensable a kind of technology in the following message transmission.

Yet, because wireless sensor network node is numerous, energy constraint, change in topology is frequent, and works under no worker monitor or rugged environment usually.Agreement and algorithm that these characteristics make tradition be used for networks such as ad hoc are difficult to be applicable to wireless sensor network.The energy demand of how in zone induction and routing procedure, to save sensor node, take into account the energy consumption balance, prolong network life, and improve cognition and the swarm intelligence collaboration capabilities of sensor node, become the wireless sensor network design and used the key issue of the necessary consideration of institute.In the various researchs of WSN and using, at first to solve the deployment issue of node, because it directly affects the accuracy of monitoring result and comprehensive.Rational and effective node deployment scheme can significantly reduce the networking time, quick coverage goal zone, and can also prolong network life by coordinating control, adapt to the topological structure that changes.Therefore, the node deployment of wireless sensor network is the first step of setting up wireless sensor network, has only finished the deployment of sensor node at guarded region, could further carry out other work and optimization.How disposing node and make it to reach optimum covering, is the key issue during sensor network nodes is disposed.This also is the emphasis problem that this paper studies.

Generally, the covering problem of wireless sensor network is a np complete problem, and present most of algorithms all are merely able to find its near-optimization to cover.Though these algorithms and agreement provide some solutions for the covering problem of WSN, but they exist the shortcoming of the flow process complexity of algorithm own, and the covering problem of WSN is as a kind of np complete problem, the optimum that these didactic deterministic algorithms often can only obtain being similar to covers the result, and the shortcoming of this local optimum is also limiting the further popularization and the use of these algorithms and agreement.

Particle cluster algorithm is as a kind of novel intelligence computation method, its basic thought is exactly that the flock of birds of simulation living nature is looked for food and the shoal of fish phenomenon of looking for food, to carrying out global search in the space of finding the solution of problem, finally find globally optimal solution by simulating these swarm intelligence behaviors.Compare with other intelligence computation methods, particle cluster algorithm is more prone to realize that operational efficiency is higher, can obtain gratifying solving result quickly.Therefore, particle cluster algorithm has obtained paying close attention to widely in recent years.But, particle cluster algorithm proposes as a kind of optimizer of continuous field when proposing, so it has natural advantage to continuous optimization problems, and often is difficult to direct use on the discrete combination problem.Given this, some researchers are on the basis that keeps predecessor group's basic idea and swarm intelligence behavioral trait, relevant discrete version particle cluster algorithm has been proposed, for the application of particle cluster algorithm on the discrete combination optimization problem provides a very effective solution.Particularly the presenter Kennedy of predecessor group algorithm and Eberhart had proposed discrete binary version particle cluster algorithm (Binary PSO afterwards, BPSO) have and predecessor group algorithm similar operation, algorithm operating is simple, and has proved its validity and high efficiency in the numerical experiment test.Therefore, the present invention will use BPSO that the node covering problem of WSN is optimized, thereby expand the application of PSO.

Summary of the invention:

The present invention applies to particle cluster algorithm in the wireless sensor network node coverage optimization.The concrete steps of algorithm comprise:

(1) generation of problem.Random scatter N sensing radius is the wireless senser of R in D * D zone.

(2) each parameter of initialization algorithm, and set up the population of the first generation, wherein, the coded system of particle is a binary coding.Binary length equals all number of sensors N, shown in following formula:

$X_i=[x_i^1,x_i^2,\·\·\·,x_i^N],x_i^j=\left\{\mathrm{0,1}\right\}$

Wherein, $x_i^j=1$ Transducer j is used in expression, $x_i^j=0$ Transducer j is not used in expression.

(3), assess its adaptive value according to the current location of each particle.The adaptive value function is:

max?f(X)＝w ₁f ₁(X)+w ₂(1-f ₂(X))

$f_1\left(X\right)=\frac{\underset{S_j\∈S^{*}}{\mathrm{\Σ}}A\left(S_j\right)}{A}$

$f_2\left(X\right)=\frac{\left|S^{*}\right|}{\left|S\right|}$

Wherein, f ₁The expression coverage rate, f ₂Represent consumption rate.S={S ₁, S ₂..., S _NThe set of expression wireless sensor node, the node subclass of X representative is S ^*={ S _j, | x _j=1}.A represents the area in whole zone.| S| represents the number of all the sensors node.| S ^*| the number of sensors that expression is opened.w ₁And w ₂Represent f respectively ₁And f ₂Weight.

(4) upgrade the individual optimal location of each particle, and the global optimum position of all particles.

(5) each particle's velocity $V_i=[v_i^1,v_i^2,\·\·\·,v_i^D]$ Upgrade.More new formula is:

$v_i^d=v_i^d+c_1\×{\mathrm{rand}}_1^d\×({\mathrm{pBest}}_i^d-x_i^d)+c_2\×{\mathrm{rand}}_2^d\×({\mathrm{gBest}}^d-x_i^d)$

C wherein ₁And c ₂Be accelerator coefficient (also claiming the study factor),

With Be two random numbers on [0,1] interval, pBest _iRepresent the global optimum position of the historical optimal location and all particles of particle self respectively with gBest.

(6) according to V _iSet the reposition X of particle _iTo each dimension j, order $p=\mathrm{sigmoid}\left(v_i^j\right)=\frac{1}{1+e^{-v_i^j}},$ If r=random (0,1)＜p, then $x_i^j=1,$ Otherwise $x_i^j=0.$

(7) if satisfy end condition, termination routine then, otherwise get back to step (3).

Algorithm of the present invention has good intuitive.At first, the node covering problem of WSN is defined as one 0/1 planning problem, and the individuality of BPSO is encoded to 0/1 binary string, the evolutionary mechanism by PSO is optimized then.That is to say, N the transducer of having supposed random scatter, so individual coding is exactly 0/1 binary string that the N position is long.Then, these transducers are carried out optimal selection.If certain transducer is selected, the corresponding positions in the individuality coding is set to 1 so, otherwise is set to 0.Therefore, the network configuration of individual coding and reality is corresponding intuitively gets up, and not only thought is understood easily, and the algorithm realization is also easy.

The algorithm of invention has good flexibility, has defined two important index the result who finds the solution is assessed, and one is regional " coverage rate ", and one is " consumption rate " of transducer.Coverage rate is meant under the effect of the transducer of current use, the percentage of territory, the area occupied area area that is covered by a transducer at least in the whole zone; Consumption rate refers to the percentage of the number of sensors and the total quantity of required use.Generally speaking, coverage rate is high more good more, and consumption rate is the smaller the better.A good optimized Algorithm is exactly can be by selecting best sensor combination, makes to finish high as far as possible areal coverage under the condition of few transducer is tried one's best in use.The present invention is weighted these two indexs, thereby has constructed final evaluation criteria.The flexibility of algorithm is embodied in the selection of weighting weight, and network designer can be adjusted the weight of two indexs, thereby make algorithm be applicable to different environment according to different demands.For example,, can suitably lay particular stress on the index of " consumption rate ", allow algorithm seek as far as possible and utilize transducer covering scheme still less if to the not too application of sensitivity of some zones; If some are covered requirement than higher application, can suitably lay particular stress on the index of " coverage rate ", the coverage rate that the covering scheme that allows algorithm find reaches is high as far as possible, and the number of sensors of using only is less important index.

Description of drawings:

Fig. 1 particle cluster algorithm is optimized the flow chart that wireless sensor network node covers

Embodiment:

Further the method for invention is described below in conjunction with accompanying drawing.

The model of wireless sensor network covering problem is normally set up like this: suppose N the identical sensor node of parameter configuration rendered to monitored area A.Sensor node set C={c ₁, c ₂..., c _N, c wherein _i={ x _i, y _i, R}, (x _i, y _i) be node distribution coordinate, R is the monitoring radius.A is a two dimensional surface, is often turned to m * n grid point by discrete, and (x, y), 1≤x≤m, 1≤y≤n have not by node c by following formula computation grid point then _iCover (wherein 1 expression is capped, and 0 expression is not capped).

To pixel arbitrarily (x, y), as long as exist an integer i ∈ [1 ..., N] make P (x, y, c _i)=1, promptly this point is present in a sensor node c _iMonitoring range in, just think that it is capped.Further can count the total node that is capped in the regional A thus and count D.And definition

$f=\frac{D}{m\×n}$

Coverage rate for this wireless sensor network.

Particle cluster algorithm is a branch of evolutionary computation, is a kind of a kind of random search algorithm of biological activity of simulating nature circle.Algorithm requires each individuality (particle) to safeguard two vectors in the process of evolving, and is exactly velocity vector $V_i=[v_i^1,v_i^2,\·\·\·,v_i^D]$ And position vector $X_i=[x_i^1,x_i^2,\·\·\·,x_i^D],$ Wherein i represents the numbering of particle, and D is the dimension of finding the solution problem.The speed that particle had has determined its travel direction and speed, and the position in solution space of separating of particle representative has then been embodied in the position, be assessment this separate the basis of quality.Also require each particle to safeguard a historical optimal location vector (representing) simultaneously with pBest, that is to say in the process of evolving, if particle has arrived certain and has made the better position of adaptive value, then this position is recorded in the historical optimal vector, if particle can constantly find more excellent position, this vector also can constantly upgrade.In addition, colony also safeguards a global optimum, represents with gBest, and this is exactly optimum that among the pBest of all particles, and the guiding particle plays to this global optimum's zone convergence in this global optimum.In each generation, particle upgrades self speed and position according to pBest and gBest, and concrete formula is:

$v_i^d={\mathrm{\ω}\×v}_i^d+c_1\×{\mathrm{rand}}_1^d\×({\mathrm{pBest}}_i^d-x_i^d)+c_2\×{\mathrm{rand}}_2^d\×({\mathrm{gBest}}^d-x_i^d)$

$x_i^d=x_i^d+v_i^d$

Wherein ω represents inertia weight, c ₁And c ₂Be accelerator coefficient (also claiming the study factor),

With

Be two random numbers on [0,1] interval.

Particle cluster algorithm puts forward as the optimizer of a continuous field question, therefore it is very suitable for the optimization of the problem in continuous field, but the researcher is also attempting by changing some forms of particle cluster algorithm, the more new formula of speed and position for example, and algorithm is applied among the discrete field (Combinatorial Optimization).Wherein, (Binary PSO, it is simple BPSO) to have a form, realizes the advantage that is easy to, and shown good performance in emulation experiment for the discrete binary version particle cluster algorithm that Kennedy and Eberhart proposed in 1997.Therefore, will adopt BPSO in the present invention as optimization tool, realize to the optimization of wireless sensor network covering problem with find the solution.

BPSO has changed the coded system of traditional PS O, particle position is defined as a binary zero, 1 string, and particle's velocity will be the probability of location revision variable by being transformed of Sigmoid function.When particle when carrying out Velocity Updating, still operating speed more new formula operate, but, each dimension in the study vector sum particle current location vector at this time

Can only value 0 or 1.After the speed of each dimension is upgraded according to formula,

Will be limited on [Vmax, Vmax], wherein Vmax=6.By using Vmax that particle's velocity is limited, BPSO does not need operation parameter ω.The maximum difference of BPSO and traditional PS O is the particle position update mode.In BPSO, can't directly the arithmetic addition be carried out in position and speed.The solution of BPSO is by using the Sigmoid function, with speed

Be converted into a real number p on the interval [0,1], shown in following formula:

$p=\mathrm{sigmoid}\left(v_i^d\right)=\frac{1}{1+e^{-v_i^d}}$

By the effect of Sigmoid function, speed will be interpreted as a probable value.BPSO generates a real number r between [0,1] at random, if r＜p, then in addition corresponding position $x_i^d=1,$ Otherwise, $x_i^d=0.$

The present invention adopts discrete binary particle swarm optimization algorithm (BPSO) as instrument, and the optimum covering problem in the wireless sensor network is found the solution and optimized.The covering problem form is turned to one 0/1 planning problem, and what need find is exactly to select in numerous transducers which, so that under the prerequisite that guarantees high coverage rate, enable minimum transducer.Therefore, covering problem is very suitable for using BPSO to find the solution.In BPSO, the position encoded of each individuality is exactly a Binary Zero, 1 string, and wherein binary length equals all number of sensors N, shown in following formula:

$X_i=[x_i^1,x_i^2,\·\·\·,x_i^N],x_i^j=\left\{\mathrm{0,1}\right\}$

Wherein, $x_i^j=1$ Transducer j is used in expression, $x_i^j=0$ Transducer j is not used in expression.

No matter be to use traditional PS O, also be to use BPSO, an important problem is exactly the form that the target of optimizing need be converted into fitness function.By fitness function, we can assess individuality, and that retains separates, and make separating of making good use of guide poor separating, and finally reach whole purpose of evolving.

In the covering problem of wireless sensor network, generally can relate to " coverage rate " and " consumption rate " two indexs.Coverage rate is meant under the effect of the transducer of current use, the percentage of territory, the area occupied area area that is covered by a transducer at least in the whole zone; Consumption rate refers to the percentage of the number of sensors and the total quantity of required use.Generally speaking, coverage rate is high more good more, and consumption rate is the smaller the better.A good optimized Algorithm is exactly can be by selecting best sensor combination, makes to finish high as far as possible areal coverage under the condition of few transducer is tried one's best in use.

The present invention uses f ₁Indicate coverage rate, use f ₂Indicate consumption rate.Suppose to exist wireless sensor node S set={ S ₁, S ₂..., S _N, the node subclass of X representative is S ^*={ S _j, | x _j=1}, then f ₁And f ₂Definition respectively as follows:

$f_1\left(X\right)=\frac{\underset{S_j\∈S^{*}}{\mathrm{\Σ}}A\left(S_j\right)}{A}$

$f_2\left(X\right)=\frac{\left|S^{*}\right|}{\left|S\right|}$

In above formula, denominator A represents the area in whole zone, the area stack of the overlay area of the transducer of all unlatchings of branch subrepresentation, | S| represents the number of all the sensors node, | S ^*| the number of sensors that expression is opened.

In fact, if Direct calculation formulas may run into and is difficult to calculate

Difficulty.In order to address this problem, the present invention is by the mode of discrete regionization, from another one angle calculation f ₁The sensing radius of supposing each transducer all is R, and each transducer S _jThe position be { x _j, y _j, whole zone evenly is divided into the grid of m * n, can obtain the coordinate T of each grid point like this _i={ x _i, y _i.Shown in the following formula of formula, if Z _Ij=1, grid point T then _iBy transducer S _jCover.

Can obtain grid point T thus _iThe condition that is covered by wireless sensor network is to have a transducer S at least _jWith its covering, can be expressed as with formula:

$Z_i={\mathrm{\Π}}_{S_j\∈S^{*}}{Z}_{\mathrm{ij}}$

So Z _i=1 or Z _i=0, Z wherein _i=1 expression grid point T _iCovered Z by wireless sensor network _i=0 expression grid point T _iCan not be covered by wireless sensor network.Based on above definition, can calculate coverage rate f by following formula ₁:

$f_1\left(X\right)=\frac{{\mathrm{\Σ}}_{i=1}^{m\×n}{Z}_i}{m\×n}$

Obtained f ₁And f ₂Afterwards, just can obtain final fitness function f by the method for weighting:

maxf(X)＝w ₁f ₁(X)+w ₂(1-f ₂(X))

W in the following formula ₁And w ₂Represent f respectively ₁And f ₂Weight, its value depends on the composite request of designer to this network index.Generally require w ₁+ w ₂=1.0.Overall functional value f is between 0～1, and this value is big more, shows that scheme is excellent more.By the adaptive value function just can every kind of node deployment scheme of comprehensive assessment quality.

The flow chart that particle cluster algorithm optimization wireless sensor network node covers as shown in Figure 1.

With an emulation experiment is that example is tested the algorithm of invention, supposes that experiment is to carry out in the zone of a 100m * 100m.At first be to throw in N=100 wireless senser at random, the sensing radius R of each transducer is 11m.Use BPSO that sensor node is optimized selection then.In BPSO, population scale is M=40, parameter c ₁=c ₂=2.0, Vmax=6, evolutionary generation were 200 generations.For convenience's sake, with area dividing be the grid of m * n=100 * 100.In addition, the w in the fitness function formula ₁Be made as 0.8, w ₂Be made as 0.2.

The 1st generation of optimizing process, the adaptive value of the individuality that adaptive value is the highest in the colony is 0.79, and the sensor node number of corresponding coverage rate and use is respectively 87.57% and 53.Along with the increase of evolutionary generation, the adaptive value of colony is also rising, and to time the 50th generation, adaptive value has reached 0.87, and coverage rate has at this time also reached 94.8%.The sensor node number that use on the contrary, has dropped to 47.This explanation, the optimized choice of sensor node all has significant effects to reducing number of sensors to greatest extent and increasing regional area coverage.To the 200th generation, adaptive value has reached 0.91, and coverage rate is also up to 98.83%, and number of sensors further drops to 42.The algorithm of experimental result proof invention is highly effective in wireless sensor network node coverage optimization.

Claims (4)

1, a kind of wireless sensor network node coverage optimization method based on particle cluster algorithm is characterized in that, this method may further comprise the steps:

(1) generation of problem.Random scatter N sensing radius is the wireless senser of R in D * D zone.

(2) each parameter of initialization algorithm, and set up the population of the first generation.

(3), assess its adaptive value according to the current location of each particle.

(4) upgrade the individual optimal location of each particle, and the global optimum position of all particles.

(5) each particle's velocity $V_i=[v_i^1,v_i^2,\·\·\·,v_i^D]$ Upgrade.More new formula is:

$v_i^d=v_i^d+c_1\×{\mathrm{rand}}_1^d\×({\mathrm{pBest}}_i^d-x_i^d)+c_2\×{\mathrm{rand}}_2^d\×({\mathrm{gBest}}^d-x_i^d)$

C wherein ₁And c ₂Be accelerator coefficient (also claiming the study factor),

With

Be two random numbers on [0,1] interval, pBest _iRepresent the global optimum position of the historical optimal location and all particles of particle self respectively with gBest.

(6) according to V _iSet the reposition X of particle _iTo each dimension j, order $p=\mathrm{sigmoid}\left(v_i^j\right)=\frac{1}{1+e^{-v_i^j}},$ If r=random (0,1)＜p, then $x_i^j=1,$ Otherwise $x_i^j=0.$

(7) if satisfy end condition, termination routine then, otherwise get back to step (3).

2, based on the described a kind of wireless sensor network node coverage optimization method based on particle cluster algorithm of claim 1, the coded system that it is characterized in that particle is a binary coding.Binary length equals all number of sensors N, shown in following formula:

$X_i=[x_i^1,x_i^2,\·\·\·,x_i^N],x_i^j=\left\{\mathrm{0,1}\right\}$

Wherein, $x_i^j=1$ Transducer j is used in expression, $x_i^j=0$ Transducer j is not used in expression.

3, based on the described a kind of wireless sensor network node coverage optimization method of claim 1, it is characterized in that the adaptive value function is based on particle cluster algorithm:

maxf(X)＝w ₁f ₁(X)+w ₂(1-f ₂(X))

$f_1\left(X\right)=\frac{\underset{S_j\∈S^{*}}{\mathrm{\Σ}}A\left(S_j\right)}{A}$

$f_2\left(X\right)=\frac{\left|S^{*}\right|}{\left|S\right|}$

Wherein, f ₁The expression coverage rate, f ₂Represent consumption rate.S={S ₁, S ₂..., S _NThe set of expression wireless sensor node.The node subclass of X representative is S ^*={ S _j, | x _j=1}.A represents the area in whole zone.| S| represents the number of all the sensors node.| S ^*| the number of sensors that expression is opened.w ₁And w ₂Represent f respectively ₁And f ₂Weight.