Some Issues in Clustering Algorithms for Wireless Sensor Networks

Transcription

1 Sme Issues in Clustering Algrithms fr Wireless Sensr Netwrks Prakashgud Patil B. V. Bhmaraddi Cllege f Engineering. & Technlgy, Vidyanagar, Hubli Karnataka, India Umakant Kulkarni S. D. M. Cllege f Engineering. & Technlgy, Dharwad Karnataka, India N. H. Ayachit B. V. Bhmaraddi Cllege f Engineering. & Technlgy, Vidyanagar, Hubli Karnataka, India ABSTRACT Wireless Sensr Netwrks (WSNs) present new generatin f real time embedded systems with limited cmputatin, energy and memry resurces that are being used in wide variety f applicatins where traditinal netwrking infrastructure is practically infeasible. In recent years many appraches and techniques have been prpsed fr ptimizatin f energy usage in Wireless Sensr Netwrks. In rder t gather infrmatin mre efficiently, wireless sensr netwrks are partitined int clusters. Hwever, these methds are nt withut prblems. The mst f the prpsed clustering algrithms d nt cnsider the lcatin f the base statin. This situatin causes ht spts prblem in multi-hp wireless sensr netwrks. Unequal clustering mechanisms, which are designed by cnsidering the base statin lcatin, t sme extent slve this prblem. In this paper, we present issues related t these appraches. Keywrds Wireless Sensr Netwrks, Aggregatin, Fuzzy Lgic, Fuzzy Clustering, Prbabilistic Clustering, Unequal Clustering. 1. INTRODUCTION There have been recent advances in micr-electr-mechanical systems (MEMS) technlgy, wireless cmmunicatins, and digital electrnics. These advances have enabled the develpment f lw-cst, lw-pwer, multifunctinal sensr ndes that are small in size and cmmunicate with each ther using radi frequencies [1]. A single sensr nde has limited capability in sensing and is nt sufficient fr gathering useful infrmatin frm a specific dmain. This data gathering prcess can be accmplished by the cllective wrk f a number f sensr ndes. In many applicatins the number f sensr ndes culd be hundreds r thusands. These cllabratively wrking sensr ndes frm a netwrk which is called a wireless sensr netwrk (WSN). Wireless sensr netwrks have plenty f advantages. The deplyment f WSNs is easier and faster than the wired sensr netwrks r any ther wireless netwrks [10], because they d nt need any fixed infrastructure [22]. Since sensr ndes are densely deplyed in mst f the cases, they are able t tlerate the netwrk failures. Wireless sensr netwrks d nt require a central rganizatin and they are self-cnfiguring [10]. There are several types f wireless sensrs such as seismic, lw sampling rate magnetic, thermal, visual, infrared, and acustic and radar sensrs [1]. These sensr ndes can mnitr varius envirnmental cnditins. Sme f these cnditins are temperature, pressure, humidity, sil makeup, vehicular mvement, nise levels, lighting cnditins, the presence r absence f certain kinds f bjects and mechanical stress levels n attached bjects [3]. Wireless Sensr Netwrks prvide unfreseen applicatins in this new field f design [1]. Frm military applicatins such as battlefield mapping and target surveillance, t creating cntext aware hmes [2] where sensrs can mnitr safety and prvide autmated services tailred t the individual user; the number f applicatins are endless. In wireless sensr netwrks, each sensr nde receives signal frm a limited regin. This signal is prcessed in that sensr nde and sensed infrmatin is generally transmitted t the bservers (e.g. base statins) [21]. Sensr ndes cnsume energy while receiving infrmatin, prcessing infrmatin and transmitting infrmatin. In mst f the cases, these sensr ndes are equipped with batteries which are nt rechargeable. Therefre, energy efficiency is a majr design gal in wireless sensr netwrks [21]. Ndes can be partitined int a number f small grups, called clusters, fr aggregating data thrugh efficient netwrk rganizatin [21]. In general, each cluster has a cluster-head which crdinates the data gathering and aggregatin prcess in a particular cluster. Each cluster member frwards its data packets t the cluster-head. Clustering in wireless sensr netwrks guarantees basic perfrmance achievement with a large number f sensr ndes [17] [2]. In ther wrds, clustering imprves the scalability f wireless sensr netwrks [14]. This is because clustering minimizes the need fr central rganizatin and prmtes lcal decisins. Benefits f Clustering: The majr benefits f clustering in wireless sensr netwrks are listed [22] belw: 1) Clustering prvides the spatial reuse f resurces t increase system capacity. Fr example, if the clusters are nt neighbrs, they can use the same frequency fr wireless cmmunicatin. 2) Ruting infrmatin f a cluster is shared with nly ther cluster-heads r cluster gateways. This restrictin reduces the number f transmissins perfrmed fr distributing ruting infrmatin. By using this advantage f clustering, mre energy efficient ruting prtcls have been implemented. When cluster structure is used in a WSN, the lcal changes need nt be reflected t entire netwrk. This reduces the infrmatin prcessed by sensr ndes and data stred in sensr ndes. 18

2 2. CLASSIFICATION OF CLUSTERING ALGORITHMS There have been substantial amunt f research n clustering prtcls fr WSNs. These clustering prtcls are classified accrding t different criteria. The classificatin f clustering prtcls accrding t their bjectives is given [22] belw: Dminating-set-based clustering: This type f clustering prtcls try t find a weakly cnnected dminating set which is respnsible fr searching rute and maintaining ruting table. Thus, table-driven ruting and n-demand ruting can be applied easily. Lw-maintenance clustering: These types f clustering prtcls aim t prvide a stable cluster structure t upper layer prtcls. T achieve this gal, they try t limit reclustering situatins r reducing the cntrl messages fr clustering. Mbility-aware clustering: Mbility-aware clustering prtcls take the mbility f sensr ndes int cnsideratin. They try t grup the mbile ndes that mve with similar speed. The clusters that cnsist f mbile ndes mving with similar speed build a mre stable cluster structure fr wireless sensr netwrks. Energy-efficient clustering: Energy-efficient clustering prtcls try t use the battery energy f the sensr ndes mre wisely, because sensr ndes have limited battery energy, and they are generally nt rechargeable. Energy cnsumptin f sensr ndes can be reduced by eliminating redundant energy cnsumptin and balancing the energy usage f sensr ndes ver the netwrk. The main gal f this type f clustering prtcls is prlnging the netwrk lifetime. Lad-balancing clustering: This type f clustering prtcls tries t limit the number f sensr ndes in each cluster. This apprach prduces clusters with similar sizes. If the clusters are similar in size, lads can be mre evenly distributed within each cluster. Cmbined-metrics-based clustering: As the name implies, this type f clustering prtcls cnsider different metrics tgether. These metrics can be nde degree, battery energy, cluster size, mbility speed, etc. These types f metrics are generally used in cluster-head electin phase f clustering prtcls. 3. OVERVIEW OF CLUSTERING ALGORITHMS There are several prpsed clustering algrithms fr WSNs in recent years. In this sectin, we review Prbabilistic Clustering Algrithms, Fuzzy Clustering Algrithms and Unequal Clustering Algrithms. Prbabilistic Clustering Algrithms: In prbabilistic clustering appraches, each nde in the wireless sensr netwrk decides its rle by itself. This type f clustering algrithms aims t minimize the cmmunicatin between sensr ndes. Prbabilistic clustering algrithms guarantee rapid cnvergence and prvide balanced cluster sizes [21]. Basically, each nde assigns itself a prbability which is a number between 0 and 1. If this prbability is less than a predefined threshld, then that nde becmes a cluster-head. Based n this principle, varius prbabilistic clustering algrithms are prpsed. Here we verview LEACH (Lw Energy Adptive Clustering Hierarchy) [9], HEED (Hybrid Energy-Efficient Distributed Clustering) [20] and the algrithm prpsed by Kuhn et al. [12]. The bjective f LEACH prtcl is t minimize energy dissipatin in sensr netwrks. LEACH has distributed crdinatin and cntrl mechanisms fr cluster set-up and peratin prcesses [9]. Static clustering algrithms select cluster-heads fr WSNs nly nce, and these cluster-heads perate as cluster-head until they die. Since cluster-heads cnsume much mre energy than rdinary sensr ndes, energy cnsumptin ver the netwrk cannt be distributed evenly by using static clustering. Therefre, WSN can quickly mve t a useless state, because the number f cluster-heads decreases drastically. In LEACH prtcl, cluster heads are rtated in randmized manner, and cluster-head electin is dne peridically. The interval between tw cnsecutive cluster frmatin prcesses is called as rund. A single rund cnsists f tw phases which are set-up and steady-state phases [7]. The cluster-head electin and cluster frmatin are dne during setup phase. In steady-state phase, the data, which is gathered frm cluster member ndes, is aggregated at lcal cluster-head and transmitted t the base statin. In HEED prtcl, residual energy f each sensr nde is the primary parameter fr prbabilistic electin f cluster-heads [21]. As stated in [20], there are fur primary gals f HEED. These are listed belw: Prlnging the lifetime f the wireless sensr netwrk by evenly distributing energy cnsumptin Selecting cluster-heads in a cnstant number f iteratins Minimizatin f cntrl verhead Frmatin f well-distributed cluster-heads and cmpact clusters In case f a tie in cluster-head electin, nde degree r average distance t neighbrs parameters are used t determine the cluster-head. HEED prtcl is implemented in TinyOS, which is an perating system develped fr Berkeley mtes. Experimentatins that are emplyed fr evaluating HEED prtcl shw that clustering and data aggregatin at least duble the lifetime f the wireless sensr netwrk [21]. Kuhn et al: studied initializing newly deplyed adhc and sensr netwrks, and prpsed a prbabilistic cluster-head electin algrithm. In this apprach, the prbability f each nde depends n the nde degree [21]. Kuhn et al: shwed that their prpsed clustering apprach cmputes an asympttically ptimal clustering in ply-algarithmic time [12]. This algrithm tries t find a dminating set f ndes which will be assigned as cluster-heads. Sensr ndes cmpete t becme dminatrs by expnentially incrementing their sending prbability n a specified channel. Three different channels are used in this algrithm. Remaining tw channels are used t keep the number f dminatrs small in a vicinity f an emerging dminatr [12]. 19

3 Pure prbabilistic clustering algrithms (e.g. LEACH) have sme disadvantages, which are listed [11] belw: Since pure prbabilistic clustering algrithms nly depend n prbability, they can prduce cluster-heads clser t each ther. They d nt cnsider the residual energy f the sensr ndes. Therefre, the ndes that have lwer energy levels may becme cluster-heads. These algrithms may randmly elect cluster-heads in vicinities that have lw nde density. Pure prbabilistic clustering appraches are useful fr clusterhead electin, but they are nt sufficient. In rder t make a mre accurate cluster-head electin, sme additinal parameters such as nde degree, residual energy and lcal distance shuld be taken int cnsideratin. Fuzzy Clustering Algrithms: Fuzzy lgic is useful fr making real time decisins withut needing cmplete infrmatin abut the envirnment. On the ther hand, cnventinal cntrl mechanisms generally need accurate and cmplete infrmatin abut the envirnment [5]. Fuzzy lgic can als be utilized fr making a decisin based n different envirnmental parameters by blending them accrding t predefined rules. Sme f the clustering algrithms emply fuzzy lgic t handle uncertainties in the wireless sensr netwrks. Basically, Fuzzy Clustering algrithms use fuzzy lgic fr blending different clustering parameters t elect cluster-heads. They assign chances t tentative cluster-heads accrding t the defuzzified utput f fuzzy if-then rules. The tentative cluster-head becmes a clusterhead if it has the greatest chance in its vicinity. There are distributed and centralized fuzzy lgic clustering appraches. Here we are ging t verview the centralized apprach f Gupta et al: [5] and the distributed apprach f Kim et al: [11] which are abbreviated as CHEF. In the fuzzy clustering apprach prpsed by Gupta et al. the cluster-heads are elected at the base statin. In every rund, each sensr nde frwards its clustering infrmatin t the base statin. There are three fuzzy descriptrs which are cnsidered by the base statin during cluster-head electin. These fuzzy descriptrs are nde cncentratin, residual energy in each nde and nde centrality [5]. The definitins f these fuzzy descriptrs are given [5] belw: Nde Cncentratin: Number f the ndes in the vicinity. Residual Energy: Remaining battery energy f each sensr nde. Nde Centrality: A parameter that indicates hw central the nde is t the cluster There are 27 fuzzy if-then rules which are defined at the base statin. The base statin elects the cluster-heads accrding t these fuzzy rules. After the base statin elects the cluster-head, it frwards the electin results t entire netwrk. This algrithm is a centralized clustering algrithm, because all clustering decisins are made at the base statin. Gupta et al: claims that a centralized clustering apprach will prduce mre accurate cluster-heads, because the base statin has all clustering infrmatin abut the netwrk and base statins are mre pwerful than rdinary ndes [5]. Hwever, this centralized apprach has sme disadvantages [11]: The base statin must cllect all clustering infrmatin frm the netwrk. Repeating this prcess in every rund brings a high verhead t sensr ndes. Thus, the battery levels f the sensr ndes may run lw quickly. In this apprach the simulatin is dne fr electing nly ne cluster-head per rund. Therefre, this simulatin is nt a realistic ne. CHEF is a similar apprach t that f Gupta et al: [5], but it perfrms cluster-head electin in a distributed manner. Clusterhead electin is dne lcally. Thus, the base statin des nt need t cllect clustering infrmatin frm all sensr ndes [11]. In every rund, each nde generates a randm number between 0 and 1. If the randm number is smaller than the predefined threshld, then that nde becmes a tentative cluster-head. There are tw fuzzy descriptrs that are used in cluster-head electin. These are residual energy f each nde and lcal distance. Lcal distance is the ttal distance between the tentative cluster-head and the ndes within predefined cnstant radius r. There are 9 fuzzy if-then rules that are defined in all sensr ndes. Tentative cluster-heads calculate their chances t be a cluster-head using these fuzzy rules. If the chance f a tentative cluster-head is greater than the ther tentative cluster heads chances in radius r, then that tentative cluster-head becmes an actual cluster-head. Afterwards, it sends a cluster-head advertisement message t the ndes in the vicinity. The ndes that are nt elected as clusterhead jin t the clsest cluster by sending a message t that cluster-head. CHEF guarantees that any tw cluster-heads cannt exist within r distance [11]. Unequal Clustering Algrithms: The sensr ndes clser t the base statin cnsume mre energy, because the netwrk traffic increases as we get clse t the base statin [21]. Therefre, the ndes clser t the base statin quickly run ut f battery. In rder t balance energy cnsumptin ver the netwrk, unequal clustering apprach is intrduced. This apprach is based n the idea f decreasing the cluster sizes as we get clse t the base statin. If a cluster-head clser t the base statin has less intracluster wrk, then it can cntribute t inter-cluster data frwarding mre. Unequal clustering is meaningful even in the cases where each cluster-head frwards its aggregated data t the base statin directly. Here, we verview tw unequal clustering appraches. These are the appraches that are prpsed by Shu et al: [18] and Li et al: [11] which are abbreviated as EEUC. If a cluster-head is clser t the base statin, it has t relay mre data frwarding traffic than the sensr ndes which are far frm the base statin [18]. Each sensr nde in the netwrk tries t send its data t the base statin. Therefre, as we get clse t the base statin, the data frwarding traffic increases. Shu et al: prpsed an apprach that aims t achieve ptimal pwer allcatin ver the sensr netwrk. This apprach assigns larger cluster sizes t cluster-heads that take fewer rles in data frwarding prcess. This apprach is illustrated in Figure 1. The prpsed netwrk mdel in this apprach assumes a circular sensing regin. Hwever, generally sensr ndes are deplyed randmly by thrwing them t the target regin. Therefre, this apprach is nt a practical ne fr real envirnments in mst f the cases. This mdel shuld be imprved t handle nn-circular regins. 20

4 Figure 1: Cluster size distributin f Shu et al: apprach EEUC (Energy-Efficient Unequal Clustering) is a distributed cmpetitive unequal clustering algrithm where cluster-heads are elected by lcal cmpetitin [11]. Every nde has a pre assigned cmpetitive range. This range gets smaller as we get clse t the base statin. This makes EEUC an unequal clustering algrithm. EEUC algrithm is als a prbabilistic clustering algrithm, because in each cluster frmatin rund, each nde generates a randm number between 0 and 1 t decide whether it is ging t participate t the cluster-head electin cmpetitin r nt. If a sensr nde has decided t participate t the cmpetitin, then it becmes a tentative cluster-head. Tentative cluster-heads in lcal regins cmpete in rder t becme an actual cluster-head. This cmpetitin is based n the residual energy f each tentative cluster-head. After cluster-head electin is cmpleted, the remaining sensr ndes jin t the clsest cluster. 4. OBSERVATIONS Mst f the clustering algrithms utilize tw techniques which are selecting cluster-heads with mre residual energy and rtating cluster-heads peridically t balance energy cnsumptin f the sensr ndes ver the netwrk [13]. These clustering algrithms d nt take the lcatin f the base statin int cnsideratin. This lack f cnsideratin causes the ht spts prblem in multi-hp wireless sensr netwrks. The cluster-heads near the base statin die earlier, because they will be in heavier relay traffic than the cluster heads which are relatively far frm the base statin. In rder t avid this prblem, sme unequal clustering algrithms are prpsed in literature [18] [13]. In unequal clustering, the netwrk is partitined int clusters with different sizes. The clusters clse t the base statin are smaller than the clusters that are far frm the base statin. EEUC mechanism fr peridical data gathering partitins the sensr ndes int clusters f unequal size, and clusters clser t the base statin have smaller size. The different clustering algrithms, types and classificatins and methds are studied. The Table 1 belw summarizes the different types f clustering algrithms alng with advantages and drawbacks. 5. CONCLUSION AND FUTURE WORK In this paper we have examined the current state f prpsed clustering prtcls. Several clustering prtcls have been prpsed fr wireless sensr netwrks. Reviewing the prtcls presented in this paper is pssible t bserve that all f them are cncerned n hw t prlng the WSN life time and hw t make a mre efficient use f the critical resurces lcated at the sensr ndes. The mst f the prpsed clustering algrithms d nt cnsider the lcatin f the base statin. This situatin causes ht spts prblem in multi-hp wireless sensr netwrks. Unequal clustering mechanisms, which are designed by cnsidering the base statin lcatin, slve this prblem. The Unequal Clustering Algrithm prpsed by Shu et al: apprach assumes a circular sensing regin. Hwever, generally sensr ndes are deplyed randmly by thrwing them t the target regin. Therefre, this apprach is nt a practical ne fr real envirnments in mst f the cases. Sme f the clustering algrithms emply uncertainties in the Wireless Sensr Netwrk. Basically, a fuzzy clustering algrithms use a fuzzy lgic fr blending a different clustering parameters t elect cluster heads accrding t the defuzzified utput f fuzzy if-then rules. Clustering algrithm presented in this paper ffer a prmising imprvement ver a cnventinal clustering; hwever, there is a still much wrk t be dne. Optimal clustering in terms f energy efficiency shuld eliminate all verhead assciated nt nly with the cluster head selectin prcess, but als with nde assciatin t their respective cluster heads. 6. REFERENCES [1] I. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci. Wireless sensr netwrks: a survey. Cmput. Netw., 38(4): , [2] E.M. Belding-Ryer. Hierarchical ruting in ad hc mbile netwrks. Wireless Cmmunicatins and Mbile Cmputing, 2(5): , [3] D. Estrin, R. Gvindan, J. Heidemann, and S. Kumar. Next century challenges: Scalable crdinatin in sensr netwrks. In Prceedings f the 5th annual ACM/IEEE internatinal cnference n Mbile cmputing and netwrking, page 270. ACM, [4] I. Gupta. Cluster head electin using fuzzy lgic fr wireless sensr netwrks. Master s thesis, Dalhusie University, Halifax, Nva Sctia, March [5] I. Gupta, D. Rirdan, and S. Sampalli. Cluster-head electin using fuzzy lgic fr wireless sensr netwrks. In Cmmunicatin Netwrks and Services Research Cnference, Prceedings f the 3rd annual, pages , [6] Y. Han, S. Park, J. Em, and T. Chung. Energy-E_cient Distance Based Clustering Ruting Scheme fr Wireless Sensr Netwrks. LECTURE NOTES IN COMPUTER SCIENCE, 4706:195, [7] MJ Handy, M. Haase, and D. Timmermann. Lw energy adaptive clustering hierarchy with deterministic clusterhead selectin. In IEEE MWCN. Citeseer,

5 Prbabilistic Clustering Algrithm IJCA Special Issue n 2nd Natinal Cnference- Cmputing, Cmmunicatin and Sensr Netwrk [8] W. Heinzelman, A. Chandrakasan, and H. Balakrishnan. An applicatin-specific prtcl architecture fr wireless micrsensr netwrks. IEEE Transactins n Wireless Cmmunicatins, 1(4): , [9] W.R. Heinzelman, A. Chandrakasan, and H. Balakrishnan. Energy-e_cient cmmunicatin prtcl fr wireless micrsensr netwrks. In Prceedings f the 33rd Hawaii Internatinal Cnference n System Sciences, vlume 8, page Citeseer, [10] J. Ibriq and I. Mahgub. Cluster-based ruting in wireless sensr netwrks: issues and challenges. In Prceedings f the 2004 Sympsium n Perfrmance Evaluatin f Cmputer Telecmmunicatin Systems (SPECTS), [11] J.M. Kim, S.H. Park, Y.J. Han, and T.M. Chung. CHEF: cluster head electin mechanism using fuzzy lgic in wireless sensr netwrks. In Prceedings f the ICACT, pages , [12] F. Kuhn, T. Mscibrda, and R. Wattenhfer. Initializing newly deplyed ad hc and sensr netwrks. In Prceedings f the 10th annual internatinal cnference n Mbile cmputing and netwrking, pages ACM New Yrk, NY, USA, [13] C. Li, M. Ye, G. Chen, and J. Wu. An energy-e_cient unequal clustering mechanism fr wireless sensr netwrks. In Prceedings f the 2nd IEEE Internatinal Cnference n Mbile Ad-Hc and Sensr Systems, page 8. Citeseer. [14] M. Ltfinezhad and B. Liang. E_ect f partially crrelated data n clustering in wireless sensr netwrks. In Prc. f the IEEE Internatinal Cnference n Sensr and Ad hc Cmmunicatins and Netwrks (SECON). Citeseer, [15] V. Mhatre and C. Rsenberg. Design guidelines fr wireless sensr netwrks: cmmunicatin, clustering and aggregatin. Ad Hc Netwrks, 2(1):45 63, [16] M. Negnevitsky. Artificial intelligence: A guide t intelligent systems. Addisn-Wesley, Reading, MA, [17] C.E. Perkins and E.M. Ryer. The ad hc n-demand distance-vectr prtcl [18] T. Shu, M. Krunz, and S. Vrudhula. Pwer balanced cverage-time ptimizatin fr clustered wireless sensr netwrks. In Prceedings f the 6th ACM internatinal sympsium n Mbile ad hc netwrking and cmputing, page 120. ACM, [19] T. Vigt, A. Dunkels, J. Alns, H. Ritter, and J. Schiller. Slar-aware clustering in wireless sensr netwrks. In Prceedings f the Ninth IEEE Sympsium n Cmputers and Cmmunicatins, [20] O. Yunis and S. Fahmy. HEED: a hybrid, energy-e_cient, distributed clustering apprach fr ad hc sensr netwrks. IEEE Transactins n Mbile Cmputing, pages , [21] O. Yunis, M. Krunz, and S. Ramasubramanian. Nde clustering in wireless sensr netwrks: Recent develpments and deplyment challenges. IEEE Netwrk, 20(3):20 25, [22] JY Yu and PHJ Chng. A survey f clustering schemes fr mbile ad hc netwrks. IEEE Cmmunicatins Surveys & Tutrials, 7(1):32 48, [23] H.J. Zimmermann. Fuzzy set thery and its applicatins. Kluwer Academic Pub, Table 1 Different types f clustering algrithms alng with advantages and drawbacks. Categry Prtcl Methds / Advantages Drawbacks LEACH HEED Cluster Head Electin is dne peridically Cluster head rtated randmized manner Residual energy f sensr nde is the primary parameter fr prbabilistic electin f cluster-heads Selecting cluster-heads in a cnstant number f iteratins. Minimizatin f cntrl verhead Frmatin f well-distributed cluster-heads and cmpact clusters Prlnging the lifetime f the wireless sensr netwrk by evenly distributing energy cnsumptin Des nt guarantee gd cluster head distributin. Assumes unifrm energy cnsumptin fr CH. Cst f the verhead t frm the clusters is expensive. Since pure prbabilistic clustering algrithms nly depend n prbability, they can prduce cluster-heads clser t each ther. These algrithms may randmly elect clusterheads in vicinities that have lw nde density. Lcalized cmmunicatin verhead and energy cnsumptin during the data transmissin fr far away cluster heads is significant, especially in large scale netwrks. 22

6 Unequal Clustering Algrithms Fuzzy Clustering Algrithms IJCA Special Issue n 2nd Natinal Cnference- Cmputing, Cmmunicatin and Sensr Netwrk Table 1 Different types f clustering algrithms alng with advantages and drawbacks. Categry Prtcl Methds / Advantages Drawbacks CHEF EEUC Centralized clustering algrithm -all clustering decisins are made at the base statin. These fuzzy descriptrs are nde cncentratin, residual energy in each nde and nde centrality. There are 27 fuzzy if-then rules which are defined at the base statin. The base statin elects the cluster-heads accrding t these fuzzy rules. Prbabilistic clustering algrithm, Based n the idea f decreasing the cluster sizes as we get clse t the base statin cluster-heads are elected by lcal cmpetitin The base statin must cllect all clustering infrmatin frm the netwrk. Repeating this prcess in every rund brings a high verhead t sensr ndes. Thus, the battery levels f the sensr ndes may run lw quickly. Assumes circular sensing regin fr clustering frmatin. Hwever, generally sensr ndes are deplyed randmly by thrwing them t the target regin. Therefre the apprach is nt a practical ne fr real envirnments in mst f the cases 23