Genetic Algorithm for Sensor Scheduling with Adjustable Sensing Range

Transcription

1 Genetc Algorthm for Sensor Schedulng wth Adjustable Sensng Range D.Arvudanamb #, G.Sreekanth *, S.Balaj # # Department of Mathematcs, Anna Unversty Chenna, Inda arvu@annaunv.edu skbalaj8@gmal.com * Department of Computer Scence Engneerng, JNN Insttute of Engneerng Chenna, Inda sreeranga00@yahoo.com Abstract Sensng, processng communcaton can consume energy n wreless sensor network. The lfetme of wreless sensor network can be extended by effcent power management. One effcent method to extend the network lfetme s to dvde the set of all deployed sensors nto non dsjont subsets or sensor covers so that each sensor cover can montor the entre targets. Adjustng the sensng range of each sensor can avod the redundantly covered targets by multple sensors. Ths process saves the energy of sensors maxmzes the number of sensor covers. By dentfyng maxmum number of sensor covers then actvatng these sensor covers n sequence wll mprove the network lfetme. Ths paper, proposes Genetc Algorthm (GA) as a better soluton to the Sensor Covers wth Adjustable Sensng Range (SC-ASR). The smulaton results of the varous problem nstances confrm that GA mproved the lfetme of sensor network compared wth fxed sensng range. Keyword- Genetc Algorthm, Sensor schedulng, Adjustable sensng range I. INTRODUCTION Wreless Sensor Network (WSN) conssts of large number of sensors a base staton. Sensors are used to montor the regon the base staton s used to collect the data from the sensors to facltate crucal acton. There are many successful applcatons n wreless sensor network lke battle feld montorng, forest fre detecton, tsunam detecton, nuclear power plant montorng so on. Each sensor contans actuator (to sense the envronment), processor (to perform certan tasks), transrecever (to send receve data from sensors) battery (to supply energy to perform certan tasks). Each sensor contans lmted battery energy t cannot be replaced or recharged frequently. The entre task has to be performed wth the lmted battery energy [,, ]. If the battery runs out of energy then the sensor cannot perform any task. The crucal queston s how to prolong the lfetme of the sensor network wth the lmted energy. Sensor sensng devce can be scheduled to go nto sleep mode whenever t does not requre to sense the regon [5, ]. The processor can elmnate certan redundant data [8] dentfy the shortest path from source to destnaton [, 9]. The transrecever can then send the aggregated data to destnaton va the shortest path dentfed by the processor. Another way to prolong the sensor energy would be to schedule the transrecever to go nto sleep mode whenever the sensor does not have a data to transmt to base staton []. All the concepts dscussed above wll conserve the sensor energy. Ths paper deals wth schedulng the sensor sensng devce n actve/sleep mode. Many sensors are romly deployed to montor the gven targets. In ths rom placement, more than one sensor can cover the same targets. However, only one sensor s enough to montor that partcular target. If all the sensors are actvated at the same tme then the redundant sensors energy wll be unnecessarly consumed. To avod ths, the set of sensors wth fxed sensng range can be parttoned nto non dsjont subset or sensor covers whch montor the entre targets. Then actvaton of the sensor covers n successon wll ncrease the sensor network lfetme. Many algorthms lke greedy [, ], genetc algorthm [6], memetc algorthm [7], learnng automata [6] are proposed to dentfy the maxmum number of sensor covers. Improvsaton would stll be requred to maxmze the lfetme of sensor network. Adjustng the sensng range of each sensor whch does not affect the coverage requrement wll ncrease the lfetme of sensor network. It means that, each actve sensor have a mnmum sensng range to montor the gven targets. In ths paper, two power savng technques are studed ) schedulng the sensors nto actve/sleep mode ) adjustng the sensng range of each sensor. These two technques are bound to mprove the overall lfetme of WSN. ISSN : Vol 6 No 5 Oct-Nov 0 8

2 II. RELATED WORKS Area coverage problem s to montor the gven regon by set of romly deployed sensors. Area coverage problem s modfed as target coverage problem by effectve parttonng of the gven area nto subfeld each subfeld s treated as targets [7]. Many algorthms were proposed to maxmze the lfetme of sensor network n target coverage problem wth fxed sensng range. MA fnally mproved the qualty compared wth the exstng work n target coverage problem wth fxed sensng range [5]. In the meanwhle, adjustable sensng range concept was ntroduced n area coverage problem ths aded maxmzng the lfetme of WSN compared wth fxed sensng range [0]. Mhaela carde et al. used the concept of adjustable sensng range n target coverage problem to mprove the lfetme of wreless sensor network [5]. Intally they modeled the target coverage problem n nteger programmng Lnear Programmng (LP) then desgned the LP based heurstc algorthm. Further, authors approached the centralzed dstrbuted greedy algorthm to maxmze the soluton qualty. Fnally, author dentfed that centralzed greedy algorthm has hgher lfetme than other methods. Target coverage problem wth adjustable sensng range can be a model n lnear program wth exponental number of varables [8]. Further, garg-konemam algorthm s approached to solve the lnear program formulaton dentfes four tmes better results compared to [5]. In [9], author modelled the target coverage problem wth adjustable sensng range as mxed nteger programmng. Moreover, orgnal greedy greedy romzed adaptve search procedure heurstc s approached to maxmze the lfetme of sensor network. Let S { S, S,, Sn} set of targets { } III. FORMULATION OF SENSOR COVERS WITH ADJUSTABLE SENSING RANGE = be the set of sensors wth the ntal energy E are romly deployed to montor the T T T T m =,,,. Each sensor S s placed n the regon (, ) regon ( x j, y j ). The sensng range of each sensor can be varyng from { r, r,, rp} energy consumpton s { e, e,, e p }. satsfes the followng condton ( ) ( ) j j k x y targets T j placed n the the correspondng The sensor S wth the sensng range r k s cover the target T j f t x x + y y r where n, j m, k p () In ths rom placement of sensors targets, a partcular target T j s covered by many sensors. But only one sensor wth mnmum sensng range s enough to montor the target. If all the sensors are actvated at the same tme then the redundant sensor energy wll be consumed unnecessarly. So the requred sensors are actvated to montor the entre targets. If the sensors are actvated wth the hgher sensng range then t wll also reduce the lfetme of sensor network. Therefore, there s a need to adjust the sensng range of each sensor wth mnmal sensng range then actvate the sensors to montor the entre target. Identfyng the maxmum number of sensor cover s modelled as sensor covers wth adjustable sensng range. It s as descrbed below ) Let C = { c, c,, ck} where c S, =,,, k be the cover set. Each covers c contanng some sensors wth dfferent sensng ranges, whch cover the entre target set k should be maxmzed. ) Sensor S can be used n any sensor cover c up to the energy E. ) Actvate the sensor cover c, =,,, k n successon where c s the number of sensor actvated n the th schedule. Fgure shows the wreless sensor network of fve sensors wth sensng range 00 to montor the four targets. Intal energy of each sensor s fve. Each sensor has an adjustable sensng range. The dfferent sensng range for the sensor s {00, 00, 00, 00, 500} the correspondng energy consumpton s {,,,, 5} for one tme unt. Frst actvate all the sensors wth sensng range 00. Therefore, the sensor coverage for sensng range 00 s S = S = S = φ, S = { T} S 5 = { T }. No cover can be dentfed wth the sensng range 00. Therefore, the lfetme of WSN wth sensng range 00 s zero. Actvate all the sensors wth sensng range 00. Therefore, the sensor coverage for sensng range 00 s S S φ, S T, T, S T S = T One cover = = = { } = { }, { } can be dentfed for ths network. The cover s c { S, S, S } 5 5. = the correspondng energy s two for all the sensors. But the ntal energy s fve. Therefore, the sensors can be n actve mode up to the sensor energy runs out. The lfetme of WSN wth sensng range 00 s.5 =.5. ISSN : Vol 6 No 5 Oct-Nov 0 8

3 Fgure : Rom placement of sensors wth sensng range 00 targets Actvate all the sensors wth sensng range 00. Therefore, the sensor coverage for sensng range 00 s S T, T, S T, T, S T, T, S T, T S = T, T, T. Two covers can be = { } = { } = { } = { } 5 { } dentfed for ths network. The covers are c = { S, S } c { S, S } = the correspondng energy s three 5 for all sensors. Therefore, the lfetme of WSN wth sensng range 00 s.667 =.. Actvate all the sensors wth sensng range 00. Therefore, the sensor coverage for sensng range 00 s S T, T, T, S T, T, S T, T, S T, T, T, T S = T, T, T, T. Three sensor = { } = { } = { } = { } 5 { } covers can be dentfed for ths network. The covers are c = { S, S }, c = { S } c { S }, = the 5 correspondng energy s four for all sensors. Therefore, the lfetme of WSN wth sensng range 00 s.5 =.75. Actvate all the sensors wth sensng range 500. Therefore, the sensor coverage for sensng range 500 s S T, T, T, T, S T, T, T, S T, T, T, T, S T, T, T, T S = T, T, T, T. = { } = { } = { } = { } 5 { } Four covers can be dentfed for ths network. The covers are c = { S }, c { } = S, c { } S c { S } 5 = = the correspondng energy s fve for all the sensors. Therefore, the lfetme of WSN wth sensng range 500 s =. For sensng range 00, only one sensor cover s dentfed sensor S S are not used. S, S S 5 wth sensng range 00 wll cover the entre target but S wth sensng range 00, S S 5 wth sensng range 00 wll cover the entre targets. Therefore, sensors can be adjustng the sensng range contnuously then the hgher lfetme can be attaned. Fve covers can be dentfed wth the adjustable sensng range. The fve c S, S, c S, S, c S, S, S, c S c = S In S, covers are = { } = { } = { } = { } { } represents the energy j represents the sensor number. The energy consumed for S, S, S, S 5 s fve S s four. Therefore, the lfetme of WSN wth adjustable sensng range s ( ) ( ) +.5 = 5.5. Hence, the sensor wth adjustable sensng range have a hgher lfetme than the sensor wth the fxed sensng range whch s shown n the Table I. In ths paper, Genetc Algorthm (GA) s proposed to mprove the lfetme of sensor network wth adjustable sensng range. TABLE I Lfetme of Sensor Network wth Fxed Adjustable Sensng Range Sensng Range Lfetme Adjustable sensng Range j ISSN : Vol 6 No 5 Oct-Nov 0 85

4 IV. IMPLEMENTATION OF GENETIC ALGORITHM FOR SC-ASR Genetc algorthm s a stochastc algorthm based on the prncples of natural genetcs t s successfully appled n many optmzaton problems. GA s proposed to adjust the sensng range of each sensor schedule the sensors n actve/sleep mode n order to maxmze the network lfetme. GA encodes the cdate soluton as chromosome, whch s represented n the form of a matrx. An optmal soluton can search through the ntal populaton, whch s generated wth the cdate soluton. The qualty of cdate soluton can be measured by the ftness functon. A hgher ftness value s the better soluton for maxmze the network lfetme. Selecton operator chooses a chromosome as a parent matrx from ntal populaton. Crossover mutaton s appled to the parent matrx to change some genes to produce the offsprng. Selecton, crossover mutaton are repeated to dentfy the complete offsprng. The fttest chromosome s selected for the next generaton. Ths process s contnued tll the maxmum number of generaton s reached. A. Representaton Sensors wth dfferent sensng range can be represents n the matrx form. Each column represents a sngle sensor wth dfferent level of energes the total energy of each sensor cannot exceed the ntal energy. Frst column shows that sensor one wth dfferent energy level. S romly assgns the energy for sensor one from 0 to ntal energy E. Sensor one utlzes the S energy from the ntal energy E. Therefore, the remanng energy for sensor one s E = E S. S romly assgns the energy for sensor one from 0 to remanng energy of sensor one E. So, the remanng energy for sensor one s E = E S. Ths process s repeated untl the energy of sensor one s fully utlzed. The same process can be repeated to all other sensors. R = S S S S S S S S S n n u u un B. Intal Populaton Intal populaton matrx s generated n the same way as s explaned n the representaton. The set of 00 populatons s generated by makng a rom assgnment of ntal energy. Therefore, the ntal populaton matrxes are pop = 0 pop = Pop pop are the two ntal populaton matrx for the network shown n the fgure. Intal energy for the entre sensor s fve. Frst column of pop s sensor one energy assgnment. The frst element of frst column s fve, whch s romly assgned one number from 0 to ntal energy 5. The total energy of sensor one s utlzed. Therefore, the remanng element n the frst column wll be zero. Second column of pop s second sensor energy assgnment. The frst element of second column s two, whch s romly assgned one number from 0 to 5. Second element of second column s one, whch s romly assgned from second sensor remanng energy 0 to. Thrd element of second column s two. Therefore, the total energy of second sensor s fve, whch s fully utlzed the remanng element s zero. In a smlar manner, remanng columns are fxed. C. Ftness Functon Lfetme of sensor network s assgned as the ftness value. If one sensor covers all targets or combnaton of varous sensor covers all targets wth dfferent sensng range then t forms one sensor cover. The total number of sensor cover for each matrx can be dentfed set as a lfetme of sensor network. The hgher ftness value wll move on to the next generaton. Unon operaton s used to dentfy the sensor cover. In pop, S s 5 t means that sensor one utlzng energy 5 the correspondng sensng range s 500. Therefore, frst sensor wth sensng range 500 can cover all targets. So, t forms one cover. The next element S s. It means that second sensor utlzng energy two the correspondng sensng range s 00. Therefore, sensor two wth sensng T, T. It does not cover all targets so next sensor s choosen to form the cover. The next range 00 can cover { } ISSN : Vol 6 No 5 Oct-Nov 0 86

5 element S s one. It means that sensor three usng energy one the correspondng sensng range s 00. Therefore, sensor three wth sensng range 00 can cover { } T. Unon of S S can cover {,, } T T T stll the cover was not found. Therefore next element was chosen to form the sensor cover. Fnally, S S S S5 covered all targets { T, T,T, T }. Therefore, another cover was dentfed. The same process can be repeated untl the last element of the matrx. The total cover for pop s two. Therefore, the lfetme of sensor network s two for pop matrx. The same process was used to dentfy the ftness value for the entre matrx. D. Selecton Parent selecton survval selecton are used n the GA. 50 populatons are selected from the ntal populaton to form the matng pool based on the hgher ftness value. Parent selecton can be done romly from the matng pool to reproduce the chld matrx. Survval selecton apples the prncple of survvor of the fttest. Chld matrx s compared wth the parent matrx the fttest one based on the network lfetme s selected. If the chld has maxmum lfetme compared wth parent matrx then parent matrx s replaced by chld matrx n the matng pool. E. Crossover Genetc algorthm used one pont crossover to produce the offsprng chromosome. Two chromosomes are chosen from the matng pool the crossover pont s romly generated. Then two chromosomes changng the genes to produce the offsprng. pop pop are chosen from the matng pool the crossover pont s two. Therefore thrd, fourth ffth column of pop pop are nterchanged to produce the offsprng offsprng. offsprng = offsprng = F. Mutaton Creep mutaton s used to modfy the offsprng value wth the mutaton rate. Genetc algorthm romly chooses one sensor from offsprng matrx re-generates the energy level. Fourth sensor s chosen from offsprng t re-generates the energy level of sensor four. Fnally ths offsprng matrx after the mutaton dentfed the maxmum number of sensor cover ts lfetme s 5.5. When the ftness value of offsprng s hgher than the populaton matrx then populaton matrx can be replaced by offsprng matrx. offsprng = offsprng = V. PERFORMANCE EVALUATION The proposed genetc algorthm has been mplemented to maxmze the lfetme of wreless sensor network wth adjustable sensng range results obtaned by GA are compared wth lfetme of fxed sensng range. Sensors targets are romly deployed n the regon. Sensng ranges of each sensor vared from 50 to 00 wth the ncremental of 50. Results generated for each fxed sensng ranges were compared wth contnuously adjustable sensng range. Populaton sze 00, matng pool 50, generaton lmt 000 mutaton rate 0.5 were consdered n ths smulaton. The seres of smulaton was mplemented n the matlab. Each result shown n the tables are averages of 0 runs. Smulaton results consst of two sectons. Frst secton conssts of smulaton results wth varyng dfferent number of targets. Second secton conssts of smulaton results wth dfferent number of sensors. ISSN : Vol 6 No 5 Oct-Nov 0 87

6 A. Smulaton Results for Dfferent Number of Targets Ths secton shows the smulaton results to valdate the lfetme of WSN wth dfferent number of targets. Table II shows the results for 50 sensors dfferent number of targets. Table III shows the results for 00 sensors dfferent number of targets. Thrd column shows the number of maxmum possble cover can be generated by all the sensors wth sensng range 50. Smlarly fourth, ffth sxth columns show the number of maxmum possble cover can be generated by all the sensors wth sensng range 00, respectvely. Seventh column shows the number of cover generated by GA wth each sensor havng dfferent sensng ranges. Adjustable Sensng Range (ASR) generated hgher lfetme compared wth fxed sensng range. TABLE II Results Generated for Fxed Adjustable Sensng Range wth 50 Sensors Dfferent Number of Targets Sensor Target SR-50 SR-00 SR-50 SR-00 ASR TABLE III Results Generated for Fxed Adjustable Sensng Range wth 00 Sensors Dfferent Number of Targets Sensor Target SR-50 SR-00 SR-50 SR-00 ASR B. Smulaton Results for Dfferent Number of Sensors Ths secton shows the smulaton results of WSN lfetme wth dfferent number of sensors. Table IV shows the results for 00 targets wth dfferent number of sensors. Table V shows the results for 500 targets wth dfferent number of sensors. Thrd column shows the results generated by sensors wth sensng range 50 generated results s zero. It means that some of the targets are not covered by any sensors wth sensng range 50. Adjustable sensng range attan maxmum lfetme compared wth the fxed sensng range. When the number of sensors s less, then the search space s reduced adjustable sensng range mproves the large amount of soluton qualty compared wth fxed sensng range. Ths result valdates the hypothess that the adjustable sensng range attan maxmum lfetme compared wth fxed sensng range. TABLE IV Results Generated for Fxed Adjustable Sensng Range wth 00 Targets Dfferent Number of Sensors Target Sensor SR-50 SR-00 SR-50 SR-00 ASR ISSN : Vol 6 No 5 Oct-Nov 0 88

7 TABLE V Results Generated for Fxed Adjustable Sensng Range wth 500 Targets Dfferent Number of Sensors Target Sensor SR-50 SR-00 SR-50 SR-00 ASR VI. CONCLUSION Energy effcency s an mportant ssue n a wreless sensor network. Genetc algorthm mproves the lfetme of sensor network by adoptng two technques. The technques nvolve adjustng the sensng range of each sensor schedulng the sensors n actve\sleep mode. Ths method mproves the wreless sensor network lfetme. Smulaton results shows that GA ncreases WSN lfetme wth adjustable sensng range to a large extent compared wth fxed sensng range. The results valdate the effectveness effcency of GA to gve a better soluton. ACKNOWLEDGMENT One of the authors S. Balaj gratefully acknowledges the fnancal support receved from Anna Unversty under Anna Centenary Research Fellowshp to carry out ths research work. REFERENCES [] Adamu Murtala Zungeru, L-Mnn Ang Kah Phoo Seng, Classcal swarm ntellgence based routng protocols for wreless sensor networks: A survey comparson, Journal of Network Computer Applcatons, vol. 5, pp , 0. [] Akyldz I.F., Su W., Sankarasubramanam Y. Cayrc E., Wreless sensor networks: a survey, Computer Networks, vol. 8, pp. 9-, 00. [] Akyldz I.F., Tommaso Meloda Chowdury K.R., Wreless multmeda sensor networks: a survey, IEEE Wreless Communcatons, vol. (6), pp. -9, 007. [] Anar A. Hady, Sherne M. Abd El-kader Hussen S. Essa, Intellgent sleepng mechansm for wreless sensor networks, Egyptan Informatcs Journal, vol., pp. 09-5, 0. [5] Arvudanamb D., Balaj S., Rekha D., Improved memetc algorthm for energy effcent target coverage n wreless sensor networks, The Eleventh IEEE Internatonal Conference on Networkng, Sensng Control, pp. 6-66, 0. [6] Chh-Chung La, Chuan-Kang Tng Ren-Song Ko, An effectve genetc algorthm to mprove wreless sensor network lfetme for large-scale survellance applcatons, n:proceedngs of the 007 Congress on Evolutonary Computaton, pp. 5-58, 007. [7] Chuan-Kang Tng Chen-Chh Lao, A memetc algorthm for extendng wreless sensor network lfetme, Informaton Scences, vol. 80 (), pp. 88-8, 00. [8] Dhawan A., Vu C.T., Zelkovsky A. L Y. Prasad S.K., Maxmum lfetme of sensor networks wth adjustable sensng range, Seventh ACIS Internatonal Conference on Software Engneerng, Artfcal Intellgence, Networkng, Parallel/ Dstrbuted Computng, pp , 006. [9] Fang Zhou, Energy effcent coverage usng sensors wth contnuously adjustable sensng ranges, Seventh Internatonal Conference on Natural Computaton, vol., pp. 09-, 0. [0] Je Wu Shuhu Yang, Coverage ssue n sensor networks wth adjustable ranges, Internatonal Conference on Parallel Processng Workshops, pp. 6-68, 00. [] Lug Alfredo Greco, Gennaro Bogga, Sabrna Scar Petro Colombo, Secure wreless multmeda sensor networks: a survey, Thrd Internatonal Conference on Moble Ubqutous Computng, Systems, Servces Technologes, pp. 9-0, 009. [] Mhaela Carde Dng-Zhu Du, Improvng wreless sensor network lfetme through power aware organzaton, Wreless Networks, vol. (), pp. -0, 005. [] Mhaela Carde, Tha M.T., Yngshu L Wel Wu, Energy effcent target coverage n wreless sensor networks, IEEE INFOCOM, vol., pp , 005. [] Mhaela Carde Je Wu, Energy effcent coverage problems n wreless ad-hoc sensor networks, Computer Communcatons, vol. 9 (), pp. -0, 006. [5] Mhaela Carde, Je Wu, Mngmng Lu Mohammad O. Pervaz, Maxmum network lfetme n wreless sensor networks wth adjustable sensng ranges, IEEE Internatonal Conference on Wreless Moble Computng, Networkng Communcatons, vol., pp. 8-5, 005. [6] Mostafae H. Meybod M.R., Maxmzng lfetme of target coverage n wreless sensor networks usng learnng automata, Wreless Personal Communcatons, Vol. 7 (), pp. 6-77, 0. [7] Sasa Sljepcevc Modrag Potkonjak, Power effcent organzaton of wreless sensor networks, IEEE Internatonal Conference on Wreless Communcatons, vol., pp. 7-76, 00. [8] Shoulng J, Jng He, Y Pan, Yngshu L, Contnuous data aggregaton capacty n probablstc wreless sensor networks, Journal of Parallel Dstrbuted Computng, vol. 7, pp , 0. [9] Wang Ke, Wang Qanpng, Jang Dong Xu Qn, A routng postonng algorthm based on a k-barrer for use n an underground wreless sensor network, Mnng Scence Technology, vol., pp , 0. ISSN : Vol 6 No 5 Oct-Nov 0 89