journal of network and computer applications · 2019. 5. 12. · 2001), bacterial foraging...

Journal of Network and Computer Applications 34 (2011) 888–901

Contents lists available at ScienceDirect

Journal of Network and Computer Applications

1084-80

doi:10.1

n Corr

E-m

hukuny

journal homepage: www.elsevier.com/locate/jnca

RFID network planning using a multi-swarm optimizer

Hanning Chen n, Yunlong Zhu, Kunyuan Hu, Tao Ku

Key Laboratory of Industrial Informatics, Shenyang Institute of Automation, Chinese Academy of Sciences, Shenyang, 110016, China

a r t i c l e i n f o

Article history:

Received 20 October 2009

Received in revised form

24 March 2010

Accepted 28 April 2010Available online 7 May 2010

Keywords:

RFID network planning

Multi-swarm

Hierarchical interaction topology

PSO

PS2O

45/$ - see front matter & 2010 Elsevier Ltd. A

016/j.jnca.2010.04.004

esponding author. Tel.: +86 24 23970026; fa

ail addresses: [email protected] (H. Che

[email protected] (K. Hu), [email protected] (T. Ku).

a b s t r a c t

In this paper, we develop an optimization model for planning the positions of readers in the RFID

network based on a novel multi-swarm particle swarm optimizer called PS2O. The main idea of PS2O is

to extend the single population PSO to the interacting multi-swarms model by constructing hierarchical

interaction topology and enhanced dynamical update equations. This algorithm, which is conceptually

simple and easy to implement, has considerable potential for solving complex optimization problems.

With five mathematical benchmark functions, PS2O is proved to have significantly better performance

than five successful variants of PSO. PS2O is then used for solving the real-world RFID network planning

problem. Simulation results show that the proposed algorithm proves to be superior for planning RFID

networks than canonical PSO, multi-swarm cooperative PSO (MCPSO), and two evolutionary

algorithms, namely genetic algorithm with elitism (EGA) and self-adaptive evolution strategies

(SA-ES), in terms of optimization accuracy and computation robustness.

& 2010 Elsevier Ltd. All rights reserved.

1. Introduction

Radio Frequency Identification (RFID) technology is a type ofautomatic identification system. The purpose of an RFID system isto enable data to be transmitted by a portable device, called a tag,which is read by an RFID reader and processed according to theneeds of a particular application. In recent years, an enormousamount of technical and commercial development of RFID hasbeen demonstrated in many industrial applications, such asproduction, logistics, supply chain management and asset track-ing (Wan, 1999; Park et al., 2006; Eric and Fred, 2008). In manyapplications, the deployment of RFID system has generated theRFID network planning (RNP) problem that needs to be solved inorder to operate the large-scale network of RFID readers in anoptimal fashion (Guan et al., 2006). However, RNP is one of themost challenging problems that has to meet many requires of theRFID system. In general, the RNP aims to optimize a set ofobjectives (coverage, load balance, economic efficiency andinterference between readers, etc.) simultaneously by adjustingthe control variables (the coordinates of the readers, the numberof the readers, and the antenna parameters, etc.) of the system. Asa result, in the large-scale deployment environment, the RNPproblem is a high-dimensional nonlinear optimization problemwith a large number of variables and uncertain parameters.

In the past few decades, nature-inspired computation hasattracted significant attention. Among them, the most successful

ll rights reserved.

x: +86 24 23970013.

n), [email protected] (Y. Zhu),

are evolutionary algorithms (EA) and swarm intelligence (SI).Evolutionary algorithms are search methods that take theirinspiration from natural selection and survival of the fittest inthe biological world. Several different types of EA methods weredeveloped independently. These include genetic programming(GP) (Koza, 1992), evolutionary programming (EP) (Yao et al.,1999), evolution strategies (ES) (Bäck and Schwefel, 1995), andgenetic algorithm (GA) (Holland, 1975). Swarm intelligence isinspired by the collective behavior of social systems (such as fishschools, bird flocks and ant colonies), and has became aninnovative computational way to solving hard optimizationproblems. Currently, SI includes four different algorithms, namelyant colony optimization (ACO) (Dorigo et al., 1996; Dorigo andStützle, 2004), particle swarm optimization (PSO) (Eberchart andKennedy, 1995; Shi and Eberhart, 1998; Kennedy and Eberhart,2001), bacterial foraging algorithm (BFO) (Müeller et al., 2002;Passino 2002; Chen et al., 2009), and artificial bee colony (ABC)(Karaboga, 2005). Due to the simplicity and flexibility of EA and SIalgorithms, various evolutionary and swarm intelligence basedmethods have been developed to plan different wireless commu-nication networks (e.g., cellular radio network, wireless sensornetwork, and RFID network), such as Evolution Strategies(Sandalidis et al., (1998)), Genetic Algorithm (Jourdan and deWeck, 2004; Cerrum et al., 2004; Guan et al., 2006), and Particleswarm Optimization (Liang et al., 2006a, b; Elkamchouchi et al.,2007; Chen et al., 2010).

In this paper, a novel multi-swarm optimizer, called PS2O,which extend the single population PSO to interacting multi-swarm model by constructing hierarchical interaction topologiesand enhanced dynamical update equations, is proposed forsolving the network planning problem in RFID system. In PS2O,

www.elsevier.com/locate/jncadx.doi.org/10.1016/j.jnca.2010.04.004mailto:[email protected]:[email protected]:[email protected]:[email protected]/10.1016/j.jnca.2010.04.004

H. Chen et al. / Journal of Network and Computer Applications 34 (2011) 888–901 889

we implement a hierarchical interaction topology that consists oftwo levels, namely the individual level and swarm level, in whichinformation exchanges take place permanently. Each individual ofthe proposed model evolves based on the knowledge integrationof itself (associate with individual’s own cognition), its swarmmembers (associate social interaction within each swarm) and itssymbiotic partners from other swarm (associate heterogeneouscooperation between different swarms). That is, we extend thedynamic update equation of the classical PSO model by adding asignificant ingredient, which takes into account the symbioticmutualism coevolution between different swarms. By incorporat-ing this new degree of complexity, PS2O can accommodate aconsiderable potential for solving more complex problems. Herewe provide some initial insights into this potential by evaluatingPS2O on both mathematical benchmark functions and a real-world RNP case, which focusing on minimizing four specificobjective functions of a ten-reader RFID network. The simulationresults, which are compared to other methods, are reported in thispaper to show the merits of the proposed algorithm.

The rest of the paper is organized as follows. In Section 2, RFIDsystem models and the RNP problem definitions are presented.Section 3 first gives a review of the canonical PSO algorithm, andthen proposes the novel PS2O algorithm. In Section 4, it will beshown that PS2O outperforms the canonical PSO and its variantson five widely-used benchmark functions. The simulation resultsof PS2O on the RNP problem are presented in Section 5. Finally,Section 6 outlines the conclusions.

2. RFID network planning problem formulation

The key components of an RFID system are the tags and readers.Both tags and readers have an antenna for radio communicationwith each other. The RFID tag, which is attached to the item to betracked, stores the unique identification number of the item using asmall integrated circuit. The RFID readers communicate with thetags by reading/writing the information stored on them. The readerhas a limit on its interrogation range, within which the tags can beread. For example, RFID readers operating in the UHF bandtypically have an interrogation range of 3–5 m. As a result, theRFID reader network deployment is required to provide completecoverage for a given area with a large number of tags.The architecture of such an RFID system is illustrated in Fig. 1,where a central repository can gather data from readers throughmulti-hop wireless communication. In the figure, from a practicalpoint of view, the geographical working area on which the RFIDnetwork is deployed is considered as a flat square surface.

Fig. 1. An example RFID network deployment. Readers are represented by squarenodes, tags are represented by round nodes, and the interrogation zone of a reader

is represented by circles around readers.

However, in many applications, it is necessary to answer someimportant questions before deploying the RFID network, such as:(1) how many readers are needed; (2) where should the readersbe placed; (3) what is the efficient parameter setting for eachreader. Thus, the RFID network planning problem is a difficultproblem that needs to be solved in order to deploy and operatethe large-scale network of RFID readers in an optimal fashion. Inthis paper, the RNP problem concerns four principal RFID systemrequirements that are formulated as follows:

2.1. Optimal tag coverage

The first objective function represents the level of coverage,which is most important in an RFID system. In this paper, if theradio signal received at a tag is higher than the thresholdPd¼�10 dBm, the communication between reader and tag canbe established. Then the function is formulated as the sum of thedifference between the desired power level Pd and the actualreceived power Pi

r of each tag in the working area:

Minimize C ¼XNTi ¼ 1ðPri�PdÞ ð1Þ

where NT is the number of tags in the working area. This objectivefunction is in order to locate the RFID readers close to the regionswhere the desired coverage level is higher, while the areasrequiring lower coverage are taken into account by the properradiate power increases of the readers.

2.2. Reader collision avoidance

Reader collision mainly occurs in a dense reader environment,where several readers try to interrogate tags at the same time inthe same area. This result in an unacceptable level of misreads.The main feature of our approach is that the interference is notsolved by traditional ways, such as frequency assignment (Engelsand Sarma, 2002) and reader scheduling (Chen et al., 2010), but ina more precautionary way. This objective function is formulatedas:

Minimize I¼XM�1i ¼ 1

XMj ¼ iþ1

ðdistðRi,RjÞ�ðriþrjÞÞ ð2Þ

where M is the number of readers, dist() is the distance computefunction, Ri and Rj are the positions of readers, ri and rj are theinterference range of ith and jth reader respectively. By changingthe positions and radiated powers of readers, this objectivefunction is in order to locate readers far from each other in orderto reduce the interference.

2.3. Economic efficiency

This aspect could be approached from various points of view.From our perspective, the readers have to be positioned as closedas possible to the barycenter of the dense tags area. This objectivecan be reached by weighing the distances of each center of tagclusters from its best served reader. Here we employ K-meansclustering algorithm to find the tag cluster. It can be definedbelow:

Minimize E¼XMk ¼ 1ðdistðRk�CenterkÞÞ ð3Þ

where M is the number of readers, dist() is the distance betweenthe kth reader and the kth center, Rk and Centerk are the positionof kth cluster center and its best served reader respectively. In this

H. Chen et al. / Journal of Network and Computer Applications 34 (2011) 888–901890

way the algorithm tries to reduce the distance from the readers tothe elements with high tag densities.

2.4. Load balance

A network with a homogeneous distribution of reader cost cangives a better performance than an unbalanced configuration(Dong et al., 2008). Thus, in large-scale RFID system, the set of tagsto be monitored needs to be properly balanced among all readers.This objective function is formulated as:

Minimize L¼YM

k ¼ 1

1

Ck

� �ð4Þ

where Ck is the assigned tags number to reader k in the workingarea. In the process of the optimization algorithm, the numbers oftags served by kth reader changes as the function of the positionand radiate power of the reader.

2.5. Combined measure

In this paper, the overall optimal solution for RNP isrepresented by a linear combination of the four objectivefunctions:

Minimize A¼X4i ¼ 1

wifi=fimax; w1þw2þw3þw4 ¼ 1, wi40

ð5Þ

where fi is the objective function for the ith requirementnormalized to its maximum value fimax. The normalization isnecessary because these four objectives represent non-homo-geneous quantities and are very different in values.

3. PS2O algorithm

3.1. Canonical particle swarm optimization

The canonical PSO is a population-based technique, similar insome aspects to evolutionary algorithms, except that potentialsolutions (particles) move, rather than evolve, through the searchspace. The rules of particle dynamics that govern this movementare inspired by models of swarming and flocking (Kennedy andEberhart, 2001). In PSO population, each particle has a positionand a velocity, and experiences linear spring-like attractionstowards two attractors:

i.

Its previous best position.

ii.
Best position of its neighbors.
In mathematical terms, the ith particle is represented asxi¼(xi1,xi2,...xiD) in the D-dimensional space, wherexidA ½1d,ud� dA ½1,D�, ld, ud are the lower and upper bounds forthe dth dimension, respectively. The rate of velocity for particle iis represented as vi¼(vi1,vi2,...,viD) is clamped to a maximumvelocity Vmax which is specified by the user. In each time step t,the particles are manipulated according to the following equa-tions:

vidðtÞ ¼ wðvidðt�1ÞþR1c1ðpid�xidðt�1ÞÞþR2c2ðpgd�xidðt�1ÞÞÞ ð6Þ

xidðtÞ ¼ xidðt�1ÞþvidðtÞ ð7Þ

where R1 and R2 are random values between 0 and 1, c1 and c2 arelearning rates, which control how far a particle will move in asingle iteration, pid is the best position found so far of the ithparticle, pgd is the best position of any particles in its neighbor-

hood, and w is called constriction factor (Clerc and Kennedy,2002), give by:

w¼ 22�j�

ffiffiffiffiffiffiffiffiffiffiffiffij2�4jp�� ð8Þ

Where j¼ c1þc2,j44.

3.2. The proposed multi-swarm optimizer

Classical PSO uses the analogy of a single-species populationand the suitable definition of the particle dynamics and theparticle information network (interaction topology) to reflect thesocial evolution in the population. However, the situation innature is much more complex than what this simple metaphorseems to suggest. Indeed, in biological populations there is acontinuous interplay between individuals of the same species,and also encounters and interactions of various kinds with otherspecies (Tomassini, 2005). The points at issue can be clearly seenwhen one observes such ecological systems as symbiosis, host–parasite systems, and prey–predator systems, in which twoorganisms mutually support each other, one exploits the other,or they fight against each other. For instance, mutualistic relationsbetween plants and fungi are very common. The fungus invadesand lives among the cortex cells of the secondary roots and, inturn, helps the host plant absorb minerals from the soil. Anotherwell-known example is the ‘‘association’’ between the Nilecrocodile and the Egyptian plover, a bird that feeds on anyleeches attached to the crocodile’s gums, thus keeping them clean.This kind of ‘‘cleaning symbiosis’’ is also common in fish.

Inspired by the mutualism phenomenon in nature, we extendthe single population PSO to the interacting multi-swarms modelby constructing hierarchical information networks and enhancedparticle dynamics. In our multi-swarms approach, the interactionoccurs not only between the particles within each swarm but alsobetween different swarms. That is, the information exchanges ona hierarchical topology of two levels, namely the individual leveland the swarm level. Many patterns of connection can be used indifferent levels of our model. The most common ones are rings,two-dimensional and three-dimensional lattices, stars, and hy-percubes. Two example hierarchical topologies are illustrated inFig. 2. In Fig. 2(a), four swarms at the upper level are connected bya ring, while each swarm that possesses four individual particlesat the lower level is structured as a star. While in Fig. 2(b), bothlevels are structured as rings. Then, we suggest in the proposedmodel that each individual moving through the solution spaceshould be influenced by three attractors:

i.
Its own previous best position that is called ‘‘personal best’’(pbest).
ii.
Best position of its neighbors from its own swarm that iscalled ‘‘species best’’ (sbest).
iii.
Best position of its neighbor swarms that is called ‘‘commu-nity best’’ (cbest).
In mathematical terms, our multi-swarm model is defined as atriplet /P,T,CS, where P¼{S1,S2,...,Sn} is a collection of n swarms,and each swarm possesses a members set Sk ¼ fxk1,xk2,. . .,xkmg of mindividuals. T is the hierarchical topology of the multi-swarm. C isthe enhanced control low of the particle dynamics, which can beformulated as:

vkidðtÞ ¼ wðvkidðt�1ÞþR1c1ðp

kid�x

kidðt�1ÞÞþR2c2ðp

kgd�x

kidðt�1ÞÞ

þR3c3ðpygd�xkidðt�1ÞÞÞ ð9Þ

xkidðtÞ ¼ xkidðt�1Þþv

kidðtÞ ð10Þ

Fig. 2. Hierarchical topology of the multi-swarm.

Table 1Pseudocode for the PS2O algorithm.

Set t:¼0;INITIALIZE. Randomize n swarms each possesses m particles;WHILE (the termination conditions are not met)

FOR (each swarm k)Find in the kth swarm neighborhood, the point with the best fitness;

Set this point as pgdy ;

FOR (each particle i of swarm k)Find in the particle neighborhood, the point with the best fitness;

Set this point as pgdk ;

Update particle velocity using Eqs. (9);

Update particle position using Eqs. (10);

END FOREND FORSet t:¼t+1;

END WHILE


where xidk represents the position of the ith particle of the kth swarm,

pidk is the personal best position found so far by xid

k , pgdk is the best

position found so far by this particle’s neighbors within swarm k, pgdy

is the best position found so far by the other swarms in theneighborhood of swarm k (here y is the index of the swarm which thebest position belongs to), c1 are the individual learning rates, c2 arethe social learning rate between particles within each swarm; c3 arethe social learning rate between different swarms and R1,R2,R3AR

D

are random vectors uniformly distributed in [0, 1]. Here, the termR1c1ðpkid�x

kidÞ is associated with cognition since it takes into account

the individual’s own experiences; the term R2c2ðpkgd,xkidÞ represents

the social interaction within swarm k; the term R3c3ðpygd�xkidÞ takes

into account the symbiotic coevolution between dissimilar swarms.When constriction factor is implemented as in the canonical PSO

above, w is calculated from the values of the acceleration coefficients(i.e., the learning rate) c1 and c2, importantly, it is the sum f of thesetwo coefficients that determines what w to use (Clerc and Kennedy,2002). This fact implies that the particle’s velocity can be adjusted byany number of terms, as long as the acceleration coefficients sum toan appropriate value. Thus, the constriction factor w in velocityformula of PS2O can be calculated by:

w¼ 2

2�f�ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffif2�4f

q��

ð11Þ

where f¼ c1þc2þc3, f44 Then the algorithm will behaveproperly, at least as far as its convergence and explosioncharacteristics, whether all of j is allocated to one term, or it isdivided into thirds, fourths, etc.

The pseudocode for the PS2O algorithm is listed in Table 1. Theflowchart of the PS2O algorithm is presented in Fig. 3, and thecorresponding variables are summarized in Table 2.

4. Benchmark test

4.1. The benchmark function

A set of 5 benchmark functions, which are commonly used inevolutionary computation literature (Shi and Eberhart, 1999;Liang et al., 2006a, b) to show solution quality and convergencerate, was employed to evaluate the PS2O algorithm in comparisonto others. The first problem is the unimodal Sphere function thatis easy to solve. The second problem is the Rosenbrock function,which has a narrow valley from the perceived local optima to theglobal optimum and can be treat as a multimodal problem. Theremaining three functions are multimodal problem. Griewank’sfunction has a

QDi ¼ 1 cosðxi=

ffiffiipÞ component causing linkages

among variables, thereby making it difficult to reach the globaloptimum. The Weierstrass function is continuous but differenti-able only on a set of points. The composition functions are a set ofnovel challenging problems, which are constructed using somebasic benchmark function with a random located global optimumand several randomly located deep local optima. The Gaussianfunction is used to combine the basic functions and blur thefunction structures. CF1 is constructed using 10 sphere functionswhich is an asymmetrical multimodal function with 1 globaloptimum and 9 local optima (the landscape of CF1 is illustrated inFig. 4). The variables of the CF1 formulation can be referred to(Liang et al., 2005). The formulas of these functions are presentedbelow:

1.
Sphere function
f1ðxÞ ¼XDi ¼ 1

x2i ð12Þ

Rosenbrock function
2.
f2ðxÞ ¼XDi ¼ 1

100� ðxiþ1�x2i Þ2þð1�xiÞ2 ð13Þ

Griewank function
3.
f3ðxÞ ¼1

4000

XDi ¼ 1

x2i �YDi ¼ 1

cosxiffiffi

ip� �

þ1 ð14Þ

Weierstrass function
4.
f4ðxÞ ¼XDi ¼ 1

Xk maxk ¼ 0½ak cosð2pbkðxiþ0:5ÞÞ�

!

�nXk maxk ¼ 0½ak cosð2pbk�0:5Þ�

! ð15Þ

where a¼0.5, b¼3, kmax¼20.

Fig. 3. The flowchart of the PS2O algorithm.

Table 2List of variables used in PS2O.

n The number of swarms

m Population size of each swarm

k Swarm’s ID counter from 1 to n

i Individual’s ID counter from 1 to m

d Dimension of the problem

t Generation counter from 1 to max generation

y The index of the best neighbor swarm of the kth swarmxid

k The ith individual’s (of the kth swarm) dth dimension’s value

pidk The ith individual’s personal best (of the kth swarm)

pgdk The best neighbor position of xid

k in the kth swarm

pidy The best neighbor position of the kth swarm


5.
Composition function 1
f5ðxÞ ¼Xni ¼ 1fwi�½fiððx�oiþoioldÞ=li�MiÞþbiasi�gþ f _bias ð16Þ

where n, the number of basic function; wi, weight value foreach fi(x); fi(x), ith basic function used to construct thecomposition function (here f1–f10: Sphere Function); oi, newshifted optimum position for each fi(x); oiold, old optimumposition for each fi(x); li, used to stretch or compress the

function; Mi, orthogonal rotation matrix for each fi(x); biasi,define which optimum is global optimum.

4.2. Settings for the algorithms

Experiments were conducted with PS2O compared with fivesuccessful variants of PSO:

�

Local version of PSO with constriction factor (CPSO) (Kennedyand Mendes, 2002);

�

Fully informed particle swarm (FIPS) (Mendes et al., 2004);

�
Unified particle swarm (UPSO) (Parsopoulos and Vrahatis,
2004);

�
Fitness-distance-ratio based PSO (FDR-PSO) (Veeramachaneni
et al., 2003);

�
Multi-swarm cooperative PSO (MCPSO) (Niu et al., 2007).
Among these variations, UPSO combined the global version andlocal version PSO together to construct a unified particle swarmoptimizer; FIPS used all the neighbors’ knowledge of the particleto update the velocity; the FDR-PSO selects one other particle,which has a higher fitness value and is nearer to the particle being

-5

0

5

-5

0

50

500

1000

1500

Fig. 4. The landscape maps of CF1 function.


updated, to update each velocity dimension; MCPSO is also amultiple swarm optimizer based on the master-slave commu-nication model, in which the master swarm adjusts its trajectoryaccording to its own experience and the knowledge of slaveswarms. It should be noted that the major difference betweenPS2O and MCPSO is: in PS2O, each swarm performs the samesearching strategy and the searching information can transferamong these identical swarms according to different predefinedinteraction topologies; while in MCPSO, a population consists ofone master swarm and several slave swarms, and there is noinformation sharing between slave swarms.

The number of swarms n needs be tuned. Three 10-Dfunctions, namely Sphere, Rosenbrock, and Griewank, are usedto investigate the impact of this parameter. Experiments wereexecuted by changing the number of swarms and fixing eachswarm size at 10. The average test results obtained form 30 runsare plot in Fig. 5. From Fig. 5, we can observe that the performanceof PS2O is influenced by n. When n increases, we obtained fasterconvergence velocity and better results.

For fair comparison, the population size of all algorithms usedin our experiments was set at 100 (all the swarms of PS2O andMCPSO include the same particle numbers of 10). The maximumvelocity of all PSO variants was set to be 5% of the search space forunimodal functions and 50% for multimodal functions. Forcanonical PSO and UPSO, the learning rates c1 and c2 were both2.05 and the constriction factor w¼0.729. For FIPS, the constric-tion factor w equals to 0.729 and the U-ring topology thatachieved highest success rate is used. For FDR-PSO, the inertiaweight o started at 0.9 and ended at 0.5 and a setting of c1¼c2¼2.0 was adopted. For MCPSO, the decaying inertia weight ostarted at 0.9 and ended at 0.4 and a setting of c1¼c2¼2.05,c3¼2.0 was used. For PS2O, the interaction topology illustrated inFig. 2(b) is used; the constriction factor in PS2O is also used withw¼0.729 according to Clerc’s method; correspondingly, the jcoefficient must sum to 4.1 and then the learning ratesc1¼c2¼c3¼j/3E1.3667. The parameters setting for all algo-rithms are summarized in Table 3.

4.3. Simulation results

The experiment runs 50 times respectively for each algorithmon each benchmark function of 30 dimensions. The numbers ofgenerations were set to be 10000. The representative resultsobtained are presented in Table 4, including the best, worst, mean

and standard deviation of the function values found in 50 runs.Figs. 6–10 presents the evolution process for all algorithmsaccording to the reported results in Table 4.

From the results, we can observe that the PS2O algorithm obtainan obviously remarkable performance. We can see it clearly thatPS2O converged with greatly faster speed to significantly betterresults than the other PSO variants for both unimodal andmultimodal cases. It should be mentioned that the PS2O were theonly ones able to consistently find the minimum of the Spherefunction, Griewank’s function, Weierstrass function, and Composi-tion function 1, while the other algorithms generated poorerresults on them. The result on Rosenbrock obtained by PS2O is alsovery good. Since a result within 40.0 on 30-D Rosenbrock reportedin other EA and SI works is considered well, the PS2O algorithm’sperformance on Rosenbrock function is remarkable good.

With the hierarchical interaction topology, a suitable diversityin the whole population can be maintained. At the same time, theenhanced dynamical update rule significantly speeds up themulti-swarm to converge to the global optimum. Because of this,the PS2O performs considerably better than many PSO variants.

5. RFID network planning based on PS2O

The detailed design of RFID network planning algorithm basedon PS2O is introduced in this section.

5.1. RFID network planning procedure

The overall operating process can be described as follows:

(1)
Initialization phasea. Reader specification
This gives the details of the mobile RFID readers that includethe adjustable radiated power range; the according interroga-tion range – the distance up to which a tag can be read by thereader; the interference range – the distance within which iftwo readers transmit simultaneously their signals wouldinterfere; and the number of the mobile reader to be used.

b. Topology specificationThis gives the details of the working area to be covered byRFID network according to the application scenario. Itincludes the shape and dimension of the region; thenumber of the RFID tags to be used; the tag distribution

4 6 8 10 1210-36

10-34

10-32

10-30Sphere's Function

4 6 8 10 1210-2

10-1

100Rosenbrock's Function

3 6 9 12

10-1.4

10-1.3

10-1.2

10-1.1

Griewank's Function

fitn

ess

(log

)

fitn

ess

(log

)

fitn

ess

(log

)

n n

n

Fig. 5. PS2O’s results on 3 test functions with different n.

Table 3Parameter setting for all algorithms.

Type PS2O CPSO FIPS UPSO FDR-PSO MCPSO

n 10 NA NA NA NA 10

m 10 100 100 100 100 10

w 0.729 0.729 0.729 0.729 NA NAo NA NA NA NA 0.9–0.5 0.9–0.4c1 1.367 2.05 2.05 2.05 2.0 2.05

c2 1.367 2.05 2.05 2.05 2.0 2.05

c3 1.367 NA NA NA NA 2.0


(i.e., the tag position) in the working area; and the tagpower threshold – the minimum tag received power levelunder which the communication between reader and tagcan be established.

c. Population generationn�mðnZ2, mZ2Þindividuals forming the PS2O populationshould be randomly generated and the individuals can bedivided into n swarms (each swarm contains m individualswith random positions and velocities). In each swarm, eachparticle is characterized by real number representation andhas a dimension equal to 3M (M is the number of used RFIDreaders), in which 2M dimensionalities for the coordinates ofreader positions (i.e., a possible network layout), and 1Mdimensionalities for radiated powers of each reader (i.e., apossible network parameter setting). The ith particle of thekth swarm is defined as follows:

Xik ¼ ðx1ik,x2ik,. . .,x

3mik Þ, x

jikAR ð17Þ

For example, a real-number particle [12.55, 28.33, 16.78,11.21, 25.11, 13.56, 0.50, 0.62, 1.73] is a possible planningsolution of a RFID network containing 3 readers. The 1 to 6bit means that the three readers are located in the two-dimension working area at (12.55, 28.33), (16.78, 11.21), and

(25.11, 13.56) respectively. The 7 to 9 bits mean that theradiated power allocated to each reader is 0.50, 0.62, and1.73(watt) respectively.

(2)
Optimization phaseAt the end of the initialization phase, all the informationneeded for the optimization phase is obtained for generatingthe optimal RFID network planning solution. The basicbuilding blocks of this phase are:a. Fitness evaluation
For each particle in each swarm, evaluate its fitness on asuitable optimizing objective function. Various optimizingobjective functions are possible. That is, each function of theoptimizing objective function set (Eqs. (1)–(5)) for RFIDnetwork planning that presents in Section 2 can be appliedto meet different system requirements. For example, whendeploys a RFID network in a supermarket, the most importantthing is to find a planning solution that obtains the maximumtag reading rate. Then the objective function should be themaximum tag coverage objective function, namely Eq. (1).

b. Population evolutionCompare the evaluated fitness values and select pbest,sbest, and cbest for each particle in each swarm. Thenupdate the velocity and position of each particle accordingto Eqs. (9) and (10), respectively.

c. Termination conditionThe computation is repeated until the maximum numberof iteration is met or the whole population is converged.

5.2. Illustrative example

The readers used here are mobile and the tags are passive. weconsider an idea example shown in Table 5. that is, the proposedalgorithm is evaluated against an ideal square working area: a30 m�30 m working space with 100 tags that distributed

Table 4Performance of all algorithms. In bold are the best results.

Func. (dim. 30) PS2O CPSO FIPS UPSO FDR-PSO MCPSO

f1Best 0 2.4787e-116 1.1874e-030 3.4459e-185 6.9665e-190 7.5976e-039Worst 0 1.3486e-113 9.7762e-029 1.9929e-182 7.4365e-168 3.8011e-032Mean 0 2.4205e-114 1.7391e-029 3.7072e-183 2.4789e-169 4.5248e-033Std 0 3.3966e-114 2.2995e-029 0 0 1.1842e-032

f2Best 1.5203e-015 5.8889 17.4217 0.7070 0.0012 0.1394Worst 5.1336e-014 7.4375 23.4450 4.0368 4.0879 13.7541Mean 1.0412e-014 6.6172 22.5407 2.0983 0.2797 4.7876Std 9.7087e-015 0.4028 1.2748 0.7696 1.0224 4.6528

f3Best 0 0 0 0 0 0Worst 0 0.1152 0.0123 0.0388 0.0737 0.0346Mean 0 0.0183 0.0016 0.0347 0.0179 0.0059Std 0 0.0266 0.0038 0.0478 0.0182 0.0111

f4Best 0 2.8242e-005 0 0 0 4.2985Worst 0 3.7591 0.2856 8.1054 1.5086 13.2105Mean 0 1.3510 0.0201 4.4244 0.1581 7.6180Std 0 1.1606 0.0558 2.6022 0.4569 2.3769

f5Best 0 0.0051 0 0 0 1.4462e-030Worst 0 100.0071 45.5672 0.0467 300.00 3.7882e-028Mean 0 50.0061 33.7051 0.0136 100.00 2.3917e-028Std 0 70.7121 25.8645 0.0143 141.42 2.6684e-028

0 2000 4000 6000 8000 10000-350

-300

-250

-200

-150

-100

-50

0

50

Iterations

Fitn

ess

(log)

PS2O

CPSO

FDR-PSO

UPSO

FIPS

MCPSO

Fig. 6. The median convergence results of Sphere function.


uniformly (shown in Fig. 11). Ten RFID readers, whose radiatedpower is adjustable in the range from 0.1 to 2 watt, are consideredto serve this area. Here the interrogation range according to thereader radiated power is computed as in (Dobkin, 2004). The realnumber solution representation is used for solving the RNPproblem. When ten RFID readers are employed to serve the wholeworking area, the RNP problem can be considered as a continuousoptimization problem with 30 dimensions. Then each solutionvector is characterized by 30 genes, 10+10 genes for (x, y)coordinates of reader positions, and 10 genes for radiated powersof each reader.

In this experiment, the performance on RNP of three PSOvariants, namely the CPSO, MCPSO and the proposed PS2O, arecompared with two successful evolutionary algorithms, namelythe self-adaptive evolution strategies (SA-ES) and the geneticalgorithm with elitism (EGA). The maximum generation for each

algorithm is 1000. The initialized population size of 50 individualsis the same for all tested algorithms. However, the wholepopulation is divided into 5 swarms (each possesses 10individuals) for PS2O and MCPSO in the initialization step. Theother parameters of three PSO variants were set to the samevalues as in Section 4.2. The EGA we executed is a real-codedgenetic algorithm with crossover, mutation and elite units. ForEGA, intermediate crossover rate of 0.8, Gaussian mutation rate of0.01, and the global elite operation with a rate of 0.06 wasadopted (Sumathi et al., 2008). For SA-ES, the parent populationwas set to be 15 and the offspring number was 50, and all theother control parameters were set to be default (Hansen andOstermeier, 2001).

First, all algorithms are tested on the four objective functions(Eqs. (1)–(4)) separately. The results are an optimal solution for asingle objective that does not take account of the others. In this

0 2000 4000 6000 8000 10000-4

-3

-2

-1

0

1

2

Iterations

Fitn

ess

(log)

PS2O

CPSO

FDR-PSO

UPSO

FIPS

MCPSO

Fig. 7. The median convergence results of Rosenbrock function.

0 2000 4000 6000 8000 10000-5

-4.5

-4

-3.5

-3

-2.5

-2

-1.5

-1

-0.5

0

Iterations

Fitn

ess

(log)

PS2O

CPSO

FDR-PSO

UPSO

FIPS

MCPSO

Fig. 8. The median convergence results of Griewank function.

0 2000 4000 6000 8000 10000-3

-2.5

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

Iterations

Fitn

ess

(log)

PS2O

CPSO

FDR-PSO

UPSO

FIPS

MCPSO

Fig. 9. The median convergence results of Weierstrass function.


0 2000 4000 6000 8000 10000-30

-25

-20

-15

-10

-5

0

5

Iterations

Fitn

ess

(log)

PS2OCPSO

FDR-PSO

UPSO

FIPS

MCPSO

Fig. 10. The median convergence results Composition function 1.

Table 5The example.

Reader specification Topology specification

Reader number 10 Dimension 30 m�30 mRadiated power 0.1–2 w Tag number 100

Interrogation range 3–4 m Tag distribution Uniform

Interference range 3.5–4.5 m tag power threshold �10 dBm

0 5 10 15 20 25 300

5

10

15

20

25

30

X-coordinate

Y-co

ordi

nate

Tag

Fig. 11. The ideal square working area.


way, it is possible to obtain the optimal solution when thenetwork planning only needs to care one objective. Then all testedalgorithms are applied to the combined function (Eq. (5)). Itshould be noted that the optimal weight values of the combinedfunction can be varied according to different system require-ments. This experiment set w1¼0.4, w2¼0.2, w3¼0.2 andw4¼0.2. The results of the best, worst, mean and standarddeviation of the fitness values over 30 runs based on PS2O andother algorithms are tabulated in Table 6.

This experiment also accommodated a visual perspective ofthe RFID network optimization. Fig. 12 illustrates the resultobtained when only the coverage of readers is considered.Fig. 12(a) shows the search progress of the average values foundby PS2O, CPSO, MCPSO, SA-ES, and GA over 30 runs for theobjective function 1. The corresponding location of the RFIDreaders optimized by PS2O is shown in Fig. 12(b), which alsoreports the received power levels according to the path loss. Inthis case, the algorithms increase the powers and concentrate the

Table 6Performance of all algorithms.

Objective func. PS2O CPSO EGA SA-ES MCPSO

C

Best 520.0115 538.9030 546.4006 618.1919 631.7577Worst 733.5187 894.8067 940.1571 1.7747e+003 864.5434Mean 665.9523 756.7850 753.1622 1.1474e+003 723.2234Std 63.0878 114.7312 122.6468 331.7691 73.5192

I

Best 324.4194 323.5060 374.0674 293.9330 327.1281Worst 333.8609 369.0065 451.1235 356.6901 339.1420Mean 331.0160 335.2516 399.4469 332.0981 334.5781Std 2.9651 13.1858 23.6057 15.6403 3.4705

E

Best 1.1691e-010 1.1594e-010 1.5906e-010 1.4433e-010 1.5202e-010Worst 1.9290e-010 3.1124e-010 2.3239e-010 2.2596e-009 2.5476e-010Mean 1.6131e-010 1.8593e-010 1.9425e-010 4.2330e-010 1.9679e-010Std 2.2053e-011 4.9145e-011 2.5148e-011 4.9474e-010 3.6410e-011

L

Best 0 5.2401e-009 2.2319 2.7861e-014 1.7135e-004Worst 0 3.2467e-008 16.2483 15.5788 0.0039Mean 0 1.3935e-008 8.7723 2.9694 0.0011Std 0 6.6902e-009 3.8279 4.1988 0.0012

A

Best 0.1798 0.1994 0.2123 0.2351 0.1864Worst 0.2254 0.2411 0.2758 0.3566 0.2429Mean 0.1980 0.2118 0.2394 0.2871 0.2105Std 0.1314 0.1441 0.2173 0.3720 0.1556

In bold are the best results.

0 200 400 600 800 1000500

600

700

800

900

1000

1100

1200

1300

1400

1500

Iterations

Fitn

ess

PS2OSA-ESCPSOEGAMCPSO

X-coordinate

Y-co

ordi

nate

0 5 10 15 20 25 300

5

10

15

20

25

30

-15

-10

-5

0

5

10

15

20

25

30ReaderTag

Fig. 12. Results only consider tag coverage. (a) Convergence process. (b) Reader location and received power distribution.


position of readers in the regions of the working area wherehigher coverage is required. From Fig. 12(a), the PS2O convergedgreatly faster and obtained better results than the other threealgorithms. From Fig. 12(b), we can obviously see that the PS2Ocan generate an optimal reader network layout with high tagcoverage rate.

Fig. 13 illustrates the result obtained by all algorithms whenonly the interference between readers is considered. Fig. 13(a)shows the convergence progress of the average values forobjective function 2 by all algorithms over 30 runs. Thecorresponding location of the RFID readers optimized by PS2O isshown in Fig. 13(b). In this case, the algorithms endeavor tomaintain sufficient distances between RFID readers. FromFig. 13(a), it is clear to see that for the tested objective function2, the PS2O algorithm markedly outperformed the other threealgorithms. We can observe from Fig. 13 that the objectivefunction 2 is perfectly optimized but the reader network exhibitsa worse working condition: the tag coverage rate is strongly

reduced because the readers moves away from each other andthus located far from high traffic areas.

Fig. 14 illustrates the result obtained when the distancesbetween readers and cluster centers of tag dense areas areminimized. In this case, the algorithms locate the readers close tothe barycenter of the tag dense areas so that this also takes care ofcoverage. From Fig. 14(a), we can observe that for the objectivefunction 3, the PS2O algorithm also obtain superior results. FromFig. 14(b), it can be observed that the reader network alsoprovides a satisfactory economic efficiency because the best-server areas are increased. As well, it provides sufficiently limiteddistances between each reader and traffic barycenter.

Fig. 15 illustrates the result obtained when only the loadbalance of the RFID network is optimized. In this case, thealgorithms configure the network in a load balance scheme sothat each reader in the network serves the optimal amount of tagsaccording to its capacity. From Fig. 15, PS2O outperforms all theother algorithms in this case.

0 200 400 600 800 1000

330

340

350

360

370

380

390

400

410

Iterations

Fitn

ess


X-coordinate

Y-co

ordi

nate

0 5 10 15 20 25 300

5

10

15

20

25

30

-20-15-10-5051015202530

ReaderTag

Fig. 13. Results only consider interference. (a) convergence process. (b) Reader location and received power distribution.

0 200 400 600 800 1000-23

-22.5

-22

-21.5

-21

-20.5

-20

Iterations

Fitn

ess

(log)


X-coordinate

Y-co

ordi

nate

0 5 10 15 20 25 300

5

10

15

20

25

30

-15-10-5051015202530ReaderTag

Fig. 14. Results only consider economic efficiency. (a) Convergence process. (b) Reader location and received power distribution.

0 200 400 600 800 1000-40

-35

-30

-25

-20

-15

-10

-5

0

5

Iterations

Fitn

ess

(log)


X-coordinate

Y-co

ordi

nate

0 5 10 15 20 25 300

5

10

15

20

25

30

ReaderTag

-15

-10

-5

0

5

10

15

20

25

30

Fig. 15. Results only consider load balance. (a) Convergence process. (b) Reader location and received power distribution.


Fig. 16 reports the optimized results for the combined functionobtained by all algorithms when all the requirements are considered.We can observe from Fig. 16 that PS2O still finds the superior result,which is a reasonable compromised between each requirement.

6. Conclusions

In this paper, we develop an optimization model for planningthe positions of readers in the RFID network. We consider theRFID network planning problem as a multi-objective optimization

problem, where coverage, interference, economic efficiency andnetwork load balance are particularly considered as the primaryrequirements of the RFID system. Four objective functions aredefined according to these four system requirements and thecombined measure is also given so that the multiple objectivesare optimized simultaneously. We should note that the proposedRFID network planning architecture is genetic and extendible: anyother objectives can be added to this model according to othersystem requirements without affecting the general philosophy ofthis model; the model does not depend on the optimizationalgorithm used and other techniques could be equally well

0 200 400 600 800 1000

0.2

0.25

0.3

0.35

0.4

Iterations

Fitn

ess

PS2OSA-ESCPSOGAMCPSO

X-coordinate

Y-co

ordi

nate

0 5 10 15 20 25 300

5

10

15

20

25

30ReaderTag

-20-15-10-5051015202530

Fig. 16. Results of the combined measurement. (a) Convergence process. (b) Reader location and received power distribution.


adopted, which enable a comparison of various planning optionsfor the same application scenario.

Inspired by the biological coevolution phenomenon betweendifferent species, a multi-swarm coevolution optimizer calledPS2O has been proposed to improve the performance of originalPSO. PS2O extends the single population PSO to interacting multi-swarms model by constructing hierarchical interaction topologiesand enhanced dynamical update equations. With the hierarchicalinteraction topology, a suitable diversity in the whole populationcan be maintained. At the same time, the enhanced dynamicalupdate rule significantly speeds up the multi-swarm to convergeto the global optimum. This is proven by the comparison with fourversions of PSO on five benchmark functions. The experimentresults show that, for all the test functions, the PS2O reachesremarkably better results than the other five successful PSOvariants. PS2O is then employed to solve the real-world RNPproblem. The simulation studies, which compared to CPSO,MCPSO, SA-ES, and EGA, show that the PS2O obtains superiorsolutions for RNP problem than all the other methods in terms ofoptimization accuracy and computation robustness.

There are ways to improve our proposed PS2O algorithm.Further research efforts should focus on:

(1)
The tuning of the user-defined parameters, including swarmnumber, swarm size, and hierarchical topologies, for PS2Oalgorithm based on extensive evaluation on many benchmarkfunctions.
(2)
In PS2O model, only mutualistic relations between differentspecies are mimicked. More symbiosis types could be studiedand to be incorporate in PS2O, such as commensalisms, andparasitism.
(3)
Moreover, it remains to be see how practically useful the PS2Oalgorithm are for other complex engineering optimizationproblems.
Acknowledgements

This work is supported by the National 863 plans projects ofChina under grant 2006AA04119-5 and the National 863 plansprojects of China under grant 2008AA04A105. The first authorwould like to thank Prof. John Paddison for reviewing themanuscript.

References

Bäck T, Schwefel HP. Evolution strategies I: variants and their computationalimplementation. Genetic algorithms in engineering and computer science.Chichester: Wiley; 1995 pp. 111–126.

Cerrum G, De Leo R, Micheli D, Russo P. Base-station network planning includingenvironmental impact control. IEEE Proceedings-Communications2004;151:197–203.

Chen HN, Zhu YL, Hu KY. Cooperative bacterial foraging optimization. DiscreteDynamic in Nature and Society 2009;2009.

Chen HN, Zhu YL, Hu KY. Multi-colony bacteria foraging optimization with cell-to-cell communication for RFID network planning. Applied Soft Computing2010;10(2):539–47.

Clerc M, Kennedy J. The particle swarm-explosion, stability, and convergence in amultidimensional complex space. IEEE Transactions on Evolutionary Compu-tation 2002;6(1):58–73.

Dobkin DM. The RF in RFID: passive UHF RFID in practice.. Elsevier; 2004pp. 68–77.

Dong Qf, Shukla A, Shrivastava V, Agrawal D, Banerjee S, Kar K. Load balancing inlarge-scale RFID systems. Computer Networks 2008;52(9):1782–96.

Dorigo M, Maniezzo V, Colorni A. The ant system: optimization by a colony ofcooperating agents. IEEE Transactions on Systems, Man, and Cybernetics PartB: Cybernetics 1996;vol.26(1):29–41.

Dorigo M, Stützle T. Ant colony optimization. Cambridge, MA: MIT Press; 2004.Eric N, Fred R. RFID: technology, applications, and impact on business operations.

International Journal of Production Economics 2008;112(2):507–9.Eberchart, R.C., Kennedy, J., 1995. A new optimizer using particle swarm theory. In:

Proceeding of the 6th International Symposium on Micromachine and HumanScience, Nagoya, Japan, p. 39–43.

Elkamchouchi, H.M., Elragal, H.M., Makar, M.A., 2007. Cellular radio networkplanning using particle swarm optimization. In: Proceeding of the 2007 RadioScience Conference, p. 1–8.

Engels, D.W., Sarma S.E., 2002. The reader collision problem. In: Proceeding of the2002 IEEE International Conference on Systems, Man and Cybernetics,pp. 370–76.

Guan, Q., Liu, Y., Yang, Y.P., Yu, W.S., 2006. Genetic approach for network planningin the RFID systems. In: Proceeding of the Sixth International Conference onIntelligent Systems Design and Applications, p. 567–72.

Holland JH. Adaptation in natural and artificial systems. The University ofMichigan Press; 1975.

Hansen N, Ostermeier A. Completely derandomized self-adaptation in evolutionstrategies. Evolutionary Computation 2001;9(1):159–95.

Jourdan, D.B., de Weck, O.L., 2004. Layout optimization for a wireless sensornetwork using a multi-objective genetic algorithm. In: Proceeding of the IEEESemiannual Vehicular Technology Conference. Milan, Italy, p. 2466–70.

Koza JR. Genetic programming: on the programming of computers by means ofnatural selection. Cambridge, MA: MIT Press; 1992.

Karaboga, D., 2005. An idea based on honeybee swarm for numerical optimization.Technical report TR06, Erciyes University, Engineering Faculty, ComputerEngineering Department.

Kennedy J, Eberhart RC. Swarm intelligence. San Francisco: Morgan Kaufmann;2001.

Kennedy, J., Mendes, R., 2002. Population structure and particle swarmperformance. In: Proceeding of the Congress on Evolutionary Computation,Honolulu, HI, p. 1671–6.

Liang JJ, Suganthan PN, Chan CC, Huang VL. Wavelength detection in FBG sensornetwork using tree search DMS-PSO. IEEE Photonics Technology Letters2006a;18(12):1305–7.

Liang JJ, Suganthan PN, Qin AK, Baskar S. Comprehensive learning particle swarmoptimizer for global optimization of multimodal functions. IEEE Transactionson Evolutionary Computation 2006b;10(3):281–95.

Liang, J.J., Suganthan, P.N., Deb, K., 2005. Novel composition test functions fornumerical global optimization. In: Proceeding of the 2005 Swarm IntelligenceSymposium, p. 68–75.

Müeller S, Marchetto J, Airaghi S, Koumoutsakos P. Optimization based on bacterialchemotaxis. IEEE Transactions on Evolutionary Computation 2002;6(1):16–29.

Mendes R, Kennedy J, Neves J. The fully informed particle swarm: simpler, maybebetter. IEEE Transactions on Evolutionary Computation 2004;8(3):204–10.


Niu B, Zhu YL, He XX, Zeng XP, Wu H. MCPSO: a multi-swarm cooperative particleswarm optimizer. Applied Mathematics and Computation 2007;.185(2):1050–62.

Park, D., Choi, Y.B., and Nam, K.C., 2006. RFID-based RTLS for improvement ofoperation system in container terminals. In: Proceeding of the Asia-PacificConference on Communications, p. 1–5.

Passino KM. Biomimicry of bacterial foraging for distributed optimization andcontrol. IEEE Control System Magazine 2002;22:52–67.

Parsopoulos KE, Vrahatis MN. UPSO: a unified particle swarm optimizationscheme. LNCS 2004:868–73.

Sandalidis, H.G., Stavroulakis, P.P., Rodriguez-Tellez, J., 1998. Combinatorialevolution strategy-based implementation of dynamic channel assignment incellular communications. In: Proceeding of the 6th IEE Conference onTelecommunications, p. 170–4.

Shi, Y., Eberhart, R.C., 1998. A modified particle swarm optimizer. In: Proceeding ofthe IEEE International Conference on Evolutionary Computation. Piscataway,NJ, p. 69–73.

Shi, Y., Eberhart, R.C., 1999. Empirical study of particle swarm optimization. In:Proceeding of the 1999 IEEE Congress on Evolutionary Computation, Piscat-away, NJ, p. 1945–50.

Sumathi S, Hamsapriya T, Surekha P. Evolutionary intelligence: an introduction totheory and applications with Matlab. Springer; 2008.

Tomassini M. Spatially structured evolutionary algorithms: artificial evolution inspace and time. Springer; 2005.

Veeramachaneni, K., Peram, T., Mohan, C.K., and Osadciw, L.A. 2003. Optimizationusing particle swarms with near neighbor interactions. In: Proceeding ofthe Genetic and Evolutionary Computation Conference, Chicago, IL, USA.p. 110–21.

Wan, D., 1999. Magic medicine cabinet: a situated portal for consumer healthcare.In: Proceeding of the International Symposium on Handheld and UbiquitousComputing. Karlsruhe, Germany, p. 352–5.

Yao X, Liu Y, Lin G. Evolutionary programming made faster. IEEE Transactions onEvolutionary Computation 1999;3(2):82–102.

RFID network planning using a multi-swarm optimizerIntroductionRFID network planning problem formulationOptimal tag coverageReader collision avoidanceEconomic efficiencyLoad balanceCombined measure

PS2O algorithmCanonical particle swarm optimizationThe proposed multi-swarm optimizer

Benchmark testThe benchmark functionSettings for the algorithmsSimulation results

RFID network planning based on PS2ORFID network planning procedureIllustrative example

ConclusionsAcknowledgementsReferences

journal of network and computer applications · 2019. 5. 12. · 2001), bacterial foraging...

Documents