A Data-Driven Robustness Algorithm for the Internet of Things in Smart Cities

Transcription

1 Emergng Trends, Issues, and Chaenges n Bg Data and Its Impementaton toward Future Smart Ctes A Data-Drven Robustness Agorthm for the Internet of Thngs n Smart Ctes Te Qu, Je Lu, Wesheng S, Mn Han, Huansheng Nng, and Mohammed Atquzzaman The authors propose an approach to mprove the robustness of networ topoogy based on a mut-popuaton genetc agorthm (MPGA). Frst, the geographc nformaton and neghbor st of nodes are extracted from a bg data server. Then a nove MPGA wth a crossover operator and a mutaton operator s proposed to optmze the robustness of topoogy. Abstract The Internet of Thngs has been apped n many feds, especay n smart ctes. The faure of nodes brngs a sgnfcant chaenge to the robustness of topooges. The IoT of smart ctes s ncreasngy producng a vast amount dfferent types of data, whch ncudes the node s geographc nformaton, neghbor st, sensng data, and so on. Thus, how to mprove the robustness of topoogy aganst macous attacs based on bg data of smart ctes becomes a crtca ssue. To tace ths probem, ths artce proposes an approach to mprove the robustness of networ topoogy based on a mut-popuaton genetc agorthm (MPGA). Frst, the geographc nformaton and neghbor st of nodes are extracted from a bg data server. Then a nove MPGA wth a crossover operator and a mutaton operator s proposed to optmze the robustness of topoogy. Our agorthm eeps the nta degree of each node unchanged such that the optmzed topoogy w not ncrease the energy cost of addng edges. The extensve experment resuts show that our agorthm can sgnfcanty mprove the robustness of topooges aganst macous attacs. Introducton The Internet of Thngs (IoT) [] s an ntegraton of mutpe dscpnes, ncudng ffth generaton (G) utra-dense ceuar networs [], heterogeneous ad hoc networs [], hybrd mobe networs [], wreess sensor networs [], and so on. The IoT has a broad range of appcatons n smart ctes. Typcay, t depoys a arge number of networng nodes wthn a certan area, and these nodes communcate wth each other to coect data and provde reference for smart ctes as shown n Fg., such as ndustry, agrcuture, securty, transportaton, smart home, and heathcare. Meanwhe, these are producng a vast amount and dfferent types of data. Therefore, how to mprove the robustness of IoT aganst node faure based on bg data servers of smart ctes has become an essenta ssue n recent years []. The scae-free mode s one of the cassc modes n compex networ theory. It s many used for modeng homogeneous networ topooges, n whch the node degree foows power-aw dstrbuton []. Scae-free topoogy has better performance n wthstandng random attacs than sma word topoogy, but t s frage under macous attacs []. Therefore, researchers and deveopers focus on how to construct a scae-free topoogy wth hgh robustness aganst macous attacs. In recent years, some researchers have empoyed evoutonary agorthms to enhance the robustness of topooges []. In ths artce, we are nterested n achevng ths by genetc agorthms (GAs) partcuary, whch s one cass of evoutonary agorthms. The popuaton of canddate soutons s used to evove toward the optma souton n GA. For the conventona GA, there s a typca mtaton caed premature convergence. However, one type of GA, the mut-popuaton genetc agorthm (MPGA), can effectvey overcome ths mtaton by usng mutpe popuatons to co-evove. In ths artce, the man contrbutons are as foows: Based on the topoogy characterstcs extracted from the bg data servers of smart ctes, we propose an MPGA wth nove crossover and mutaton operators to enhance the robustness of topooges. Our agorthm eeps the nodes degrees unchanged durng exchangng edges of topoogy. It w not ncrease the energy cost of addng edges. Inta Topoogy Constructon Due to the mtatons of energy and communcaton range, the scae-free IoT topoogy n smart ctes has the foowng two constrants:. The communcaton range of ts nodes cannot be arbtrary ong.. Its node degrees cannot be arbtrary arge. Because of these two constrants, the tradtona method of constructng scae-free topooges, such as the Barabás-Abert (BA) mode [0], cannot be drecty apped. In order to smuate the scae-free topoogy nformaton extracted from the bg data server n smart ctes [], we empoy the foowng method, whch adapts the BA mode, to construct the nta scae-free topooges for IoT n smart ctes. We add edges between nodes asynchronousy durng the process of constructng IoT topoogy after udgng whether they are n the communcaton range. Ths means that a par of nodes cannot generate a new edge at the same tme. The oca word of the newy oned node s composed of a the nodes wthn ts communcaton range. If a node has been connected wth the newy oned node or has reached the maxmum Dgta Obect Identfer: 0.0/MCOM.0.00 Te Qu, Je Lu, and Mn Han are wth Daan Unversty of Technoogy; Wesheng S s wth Western Sydney Unversty; Huansheng Nng s wth the Unversty of Scence and Technoogy Beng; Mohammed Atquzzaman s wth the Unversty of Oahoma. 0-0//$.00 0 IEEE IEEE Communcatons Magazne December 0

2 Agrcuture Industry Smart ctes coud Bg data server Node ID: Node 0 Locaton: ID: Node 0 E:. Locaton: ID: Node 0 N:.0 E:. Locaton: ID: 0 Neghbor N:.0 E:. Locaton: st Neghbor N:.0 E:. st Neghbor N:.0 st Neghbor st Securty Transportaton Topoogy Modeng Smart home Heathcare Fgure. The data-drven robustness topoogy of IoT n smart ctes. degree [], t w be removed from the oca area of the newy oned node. Furthermore, the newy oned node chooses ts neghbors to estabsh connecton by a rouette method accordng to the node s degree. The newy oned node prefers to connect wth hgher degree nodes n ts oca area. Premnary After the nta topooges for IoT nodes n smart ctes are constructed, we try to optmze the robustness of topoogy based on MPGA. Frst, we convert an adacency matrx of topoogy to a bnary-coded chromosome. To further ustrate ths operator, a topoogy wth four nodes s converted to a chromosome, as shown n Fg.. The topoogy conssts of node, node, node, and node, and the adacency matrx s a bnary matrx. It s feasbe to convert the adacency matrx nto a chromosome drecty. However, the storage space s wasted and the operatng compexty n GA s ncreased. The adacency matrx s a symmetrc matrx, and ts upper tranguar matrx s abe to represent the connecton reaton between nodes n networs. We convert the upper tranguar matrx to a chromosome as shown n Fg.. It can shorten the ength of the chromosome and mprove the effcency of the proposed agorthm. The probabty of seectng an ndvdua depends on ts ftness vaue. When the ftness vaue of an ndvdua s much hgher than other ndvduas n the current popuaton, ths ndvdua w be seected severa tmes n the operaton of constructng the next generaton. Fnay, the popuaton w be controed by ths ndvdua and become uncompettve, whch causes stagnant evouton n popuaton. The frequency of the crossover operator and mutaton operator s nfuenced by crossover probabty P c and mutaton probabty P m. The vaues of P c and P m drecty affect the baance between goba search and oca search n GA. The resut of evouton s qute senstve to the vaues of P c and P m. The sze of popuaton has a great nfuence Topoogy Fgure. The adacency matrx s converted to a chromosome. on the optmzaton performance. If the sze of the popuaton s sma, the dversty of popuaton w be decreased and the competton among ndvduas w be weaened. The popuaton w become a snge group soon, and the effect of the crossover operator w graduay dsappear. The update of popuaton ony rees on the mutaton operator. If the sze of the popuaton s arge, the cacuaton cost w be ncreased and the effcency w be affected. In order to sove these probems that exst n a conventona GA, the foowng mprovements are apped n an MPGA. The framewor of a GA, whch ony uses a snge popuaton to search for the optma souton, s broen by usng severa popuatons at the same tme. For dfferent popuatons, dfferent crossover probabty P c and mutaton probabty P m are seected. Athough the ranges of crossover probabty and mutaton probabty are suggested as P c (0. 0.) and P m ( ), respectvey, ther ranges are wde enough to aow penty of vaues to be seected. The optmzaton resuts w greaty dffer wth dfferent P c and P m. Gven the goba search and oca search at the same tme, we conduct experments wth varous P c and P m Adacency matrx Chromosome IEEE Communcatons Magazne December 0

3 (a) (b) (c) (d) (e) (f) (g) (h) Fgure. The process of the crossover operator: a) father; b) search n neghbors; c) swap edges; d) son; e) mother; f) search n neghbors; g) swap edges; h) daughter. to pc the best vaues to prevent the popuatons from fang nto the oca optmum. Each popuaton s ndependent and communcates wth other popuatons through the mmgraton operator. The mmgraton operator perodcay moves the optma ndvdua that appeared on each popuaton durng evouton to other popuatons (every certan number of generatons), whch acheves the gene exchange among popuatons. The best ndvdua that appears n each generaton s seected to compose the mmgraton popuaton. The operatons of crossover and mutaton are not executed n mmgraton popuaton, whch ensures that the best ndvdua of each popuaton w not be destroyed. The mmgraton popuaton s the foundaton of the mmgraton operator, whch ncreases the dversty of genetcs and guarantees the ftness functon to search for the optma souton n a wde range. To evauate the networ robustness, Schneder et a. [] proposed a new metrc, R. It consders the maxma connected subgraphs after removng the hghest degree node repeatedy to measure the robustness of networ topoogy, whch means that the mportant nodes n IoT of smart ctes are attaced. The vaue of R es n the range (0, 0.). We empoy metrc R to measure the robustness of the IoT topooges as the ftness functon of MPGA. Furthermore, Herrmann et a. [] have found that an onon-e structure s more stabe and robust aganst macous attacs. Thus, we mae the evouton of ndvdua topoogy toward the onon-e structure n the mutaton operator to mprove the robustness of topooges aganst macous attacs. Bascay, the connectons among nodes n an onon-e structure exhbt the foowng characterstcs: Nodes wth smar degrees connect to each other. Node degrees graduay decrease from nner nodes to outer nodes. The maortes of the nodes have sma degrees and are ocated n the outer ayers of the onon-e structure. Crossover Operator n MPGA The crossover operator has no fxed evouton drecton n a GA. Parent topooges generate new chdren topooges by the crossover operator, whch obtans a arger souton space. Thus, the ftness functon w search for the best souton n a arger space. Generay, the crossover operator retans a part of the father and mother genes, and eventuay generates new chdren topooges. The crossover operator n ths artce eeps the nta degree of each node unchanged, whch means that we cannot change the degree dstrbuton of parent topooges. The degree dstrbuton of new chdren topooges s the same as n the parent topooges. Tang nto account the mtaton of communcaton range n smart ctes, the crossover operator s desgned as foows. Frst, the parents are chosen by crossover probabty P c. We assume that the son topoogy nherts ts father topoogy, and the daughter topoogy nherts ts mother topoogy. Second, we get the sets of the father s excusve edges and the mother s excusve edges through the set of the father s edges and the set of the mother s edges. Here excusve means that an edge ony exsts n one parent s set but not the other. Fnay, the son topoogy dsconnects the exstng edges to bud every mother s excusve edges, durng whch the nta degree of each node s ept unchanged. The constructon process of the daughter topoogy s smar to the above operaton. Fgure ustrates the process of the crossover operator. 0 IEEE Communcatons Magazne December 0

4 Fgures a and e represent the connecton between nodes n the father topoogy and mother topoogy. It can be seen that the father has an excusve edge e between node and node n Fg. a, and the mother has an excusve edge e between node and node n Fg. e. Accordng to the crtera n the crossover operator, we bud the mother s excusve edge e n the son topoogy (Fg. d), and the father s excusve edge e n the daughter topoogy (Fg. h). Here s the detaed descrpton about how father topoogy (Fg. a) generates ts son topoogy (Fg. d). In order to generate a new edge e n Fg. a, we seect the canddate nodes that have no edge wth node from the neghbors of node. Then we cacuate the dstance of the canddate nodes to node. Fnay, we sort the dstances to generate a canddate st n ascendng order. As shown n Fg. b, node, whch s a neghbor of node and has no edge wth node, s the nearest node to node. Node searches each of ts neghbor nodes n Fg. c unt fnds a node that s n the communcaton range of node and has no edge wth node. As shown n Fg. c, node chooses ts neghbor node, and we dsconnect the edges e and e n Fg. d. After that we generate the edges e and e. Fnay, we successfuy generate a new edge, e, n the son topoogy (Fg. d). The degrees of node and node both equa before the crossover operator, and after the operator they reman unchanged. Therefore, t s consstent wth the crteron that eeps the nta degree of each node unchanged. The process that the mother generates ts daughter topoogy s smar to the above operatons, as shown n Fgs. e g. Fnay, we can see the father s excusve edge e n the daughter topoogy n Fg. h. Besdes, when the father topoogy generates ts son topoogy, f node cannot fnd an egbe node to match node, whch s the canddate neghbor of node, node w sequentay choose another canddate node n the canddate st. And node w search a of ts neghbors for each canddate node unt t fnds an egbe node to match the canddate node. If node st cannot fnd an egbe node after traversng the canddate st of node, we gve up generatng ths edge. Mutaton Operator n MPGA The mutaton operator s an mportant way to generate new ndvduas n GA. We choose the ndvdua by the mutaton probabty P m. The goa of the mutaton operator s to ncrease the robustness of the seected ndvdua through exchange edges, durng whch the nta degree of each node s unchanged. Metrc R s used to measure the robustness of topoogy. We search for the optma souton wthn the oca area by the mutaton operator. Nodes wth smar degrees connect to each other n an onon-e structure. If a node wth a arge degree fas, another node wth a arge degree w repace ts functon. Therefore, we can mnmze the adverse effects of faure nodes as much as possbe, and the networ topoogy w reman robust. In order to mae the evouton of ndvdua topoogy e the onon-e structure, we generate a new edge between two nodes (a) Fgure. The canddate for the topoogy connecton: a) nta topoogy; b) canddate ; c) canddate. that have smar degrees, durng whch the nta degree of each node s unchanged. We propose a crteron to sort degree and exchange edges as foows. We seect two edges n the ndvdua topoogy, and udge the four end nodes of these two edges as to whether they are n the communcaton range of each other to guarantee that we can generate a new edge n these four nodes. As shown n Fg. a, we seect e and e. Frst, we sort the degrees of node, node, node, and node n descendng order, and name them d, d, d, and d. Second, we add the absoute vaue of the dfference between d and d to the absoute vaue of the dfference between d and d as s. Then we add the absoute vaue of the dfference between d and d to the absoute vaue of the dfference between d and d as s. Fnay, we et s dvde s to get p and compare p wth the exchange threshod ϕ. If p s ess than ϕ, we exchange edges accordng to d, d, d, d. There are two canddate strateges n Fgs. b and c. Based on the crtera mentoned above, the nodes that have smar degrees w connect wth each other, thus enabng the evouton of ndvdua topoogy toward the onon-e structure. Besdes, the exchange threshod ϕ s defned n [0, ), and t cannot be because the two edges w not be exchanged n that case. We contro the effcency of the mutaton operator by changng the vaue of ϕ. The approprate exchange threshod ϕ can effectvey avod neffcent exchange edges operaton. Smuaton Resuts In order to extract vad nformaton of nodes n smart ctes, we smuate depoyment of IoT usng Matab. The nodes are depoyed randomy n a crcuar area wth dameter equa to 00 m. Consderng that each node must have suffcent neghbors durng the process of budng nta topoogy, the communcaton range s set to 00 m. Then the node s geographc nformaton and neghbor st are extracted for our agorthm. The parameters of our agorthm are obtaned by many experments. Fnay, we set the optma number of a popuaton to 0, and the optma number of ndvduas n a popuaton to 0. Because more teratons means more tme cost, we set teratons to 00 by consderng varous factors. The threshod of exchange edges pchange sdes from 0. to, and the sdng nterva s 0.. A the resuts of the experment are taen from the average vaue of the ( 0) tmes ndependent smuaton experments. The vertca axs (b) (c) IEEE Communcatons Magazne December 0

5 Roptmzed/Rnta pchange (a) R Iteratons (b) R MPGA-based Smuated Anneang H Cmbng Inta-BA N (c) Fgure. Performance of our agorthm: a) the threshod of exchange edges; b) the robustness evauaton of our agorthm; c) the compacton wth other agorthms. represents the rato of the r vaue after optmzng to the nta r vaue. As shown n Fg. a, the vaue of the vertca axs changes wth the dfferent vaues of PChange. The curve shows a growng trend wth the ncrease of PChange. As the curve grows, some oca peas can be seen, but the maxma vaue s gotten at 0.. Thus, the vaue of PChange s set to 0. n the foowng smuaton experments. Fgure b ustrates that the metrc r ncreases wth the number of teratons of MPGA. At the begnnng, the metrc r of the nta topoogy s ow; the vaue of r s ncreased obvousy from the st to the 0th generaton. After the 0th generaton, the optmzaton resut ncreases sowy due to the vaue of r havng ncreased to a hgh eve. Based on the estabshed nta IoT topoogy, we compare our agorthm to two exstng agorthms, namey the H Cmbng agorthm [] and the Smuated Anneang agorthm []. Both of these agorthms eep the nta degree of every node unchanged. Fgure c shows that the optmzaton resuts of the H Cmbng agorthm, the Smuated Anneang agorthm, and our MPGA-based agorthm n dfferent szes of IoT topoogy. The sze of topooges are set to 00, 0, 00, 0, and 00 nodes. The resuts are the average of ( > 0) ndependent experments, and each IoT topoogy remans connected after optmzaton. As shown n Fg. b, the vaue of r presents a downward trend wth the ncrease of networ szes, and our agorthm has better performance than the other two agorthms. Concuson Based on the bg data of smart ctes, nodes geographc nformaton and neghbor st are extracted. Then we construct the nta topooges of IoT. A nove MPGA-based agorthm s proposed to optmze the robustness of the networ topoogy aganst macous attacs. We have desgned two nove operators, the crossover operator and mutaton operator. The nta degree of each node s unchanged durng the process of these operators. Thus, the energy cost of addng edges w not be ncreased. Fnay, we have smuated our agorthm and two exstng agorthms. Ther performance n mprovng the robustness of topooges s compared under dfferent networ szes. The experment resuts show that the robustness of IoT topoogy aganst macous attacs can be mproved sgnfcanty by our agorthm. The vaues of R n the two exstng agorthms reduce qucy wth the sze of networs ncreasng, but our agorthm st mantans the vaues of R at a hgh eve. Thus, our agorthm can sgnfcanty mprove the robustness of IoT n smart ctes, especay aganst macous attacs. Acnowedgment Ths wor s supported by the Natona Natura Scence Foundaton of Chna (Grant No. and ) and the Fundamenta Research Funds for the Centra Unverstes (DUTQY and DUTZD). References [] J. A and Stanovc, Research Drectons for the Internet of Thngs, IEEE Internet of Thngs J., vo., no., Feb. 0, pp.. [] C. W. Tsa et a., Metaheurstcs for the Depoyment of G, IEEE Wreess Commun., vo., no., Dec. 0, pp. 0. [] T. Qu et a., Heterogeneous Ad Hoc Networs: Archtectures, Advances and Chaenges, Ad Hoc Networs, vo., no., Feb. 0, pp.. [] G. Han et a., Hysense: A Hybrd Mobe Crowdsensng Framewor for Sensng Opportuntes Compensaton under Dynamc Coverage Constrant, IEEE Commun. Mag., vo., no., Mar. 0, pp.. [] G. Han et a., Green Routng Protocos for Wreess Mutmeda Sensor Networs, Sensors, vo., no., June 0, pp. 0. [] J. Wu et a., Spectra Measure of Structura Robustness n Compex Networs, IEEE Trans. Systems, Man, and Cybernetcs Part A: Systems and Humans, vo., no., Nov. 0, pp.. [] T. Qu et a., Rose: Robustness Strategy for Scae-Free Wreess Sensor Networs, IEEE/ACM Trans. Net., 0. DOI: 0.0/TNET.0.0. [] R. H. L et a., Measurng Robustness of Compex Networs Under MVC Attac, Proc. st ACM Int. Conf. Info. and Knowedge Management, Oct. 0, pp.. [] M. Zhou and J. Lu, A Memetc Agorthm for Enhancng the Robustness of Scae-Free Networs Aganst Macous Attacs, Physca A: Statstca Mechancs and Its Appcatons, vo. 0, no., Sept. 0, pp.. [0] A. L. Barabás and R. Abert, Emergence of Scang n Random Networs, Scence, vo., no., Nov., pp. 0. [] W. Zhao, Performance Optmzaton for State Machne Repcaton Based on Appcaton Semantcs: A Revew, J. Systems and Software, vo., Nov. 0, pp. 0. [] W. Xao, L. Ln, and G. Chen, Vertex-Degree Sequences n Compex Networs: New Characterstcs and Appcatons, Physca A: Statstca Mechancs and Its Appcatons, vo., no., Nov. 0, pp.. IEEE Communcatons Magazne December 0

6 [] C. Schneder et a., Mtgaton of Macous Attacs on Networs, Proc. Nat. Academy of Scences, vo. 0, no. 0, Juy 0, pp.. [] H. J. Herrmann et a., Onon-Le Networ Topoogy Enhances Robustness Aganst Macous Attacs, J. Statstca Mechancs: Theory and Experment, vo. 0, no., Jan. 0, pp.. [] B. Perre, D. Fabo, and T. Marco, Optmzng the Robustness of Scae-Free Networs wth Smuated Anneang, Int. Conf. Adaptve and Natura Computng Agorthms, Apr. 0, pp.. Bographes Te Qu SM ] receved hs Ph.D. degree n computer scence from Daan Unversty of Technoogy (DUT), Chna, n 0. He s currenty an assocate professor n the Schoo of Software, DUT. He s the foundng drector of the Smart Cyber-Physca Systems Laboratory (SmartCPS Lab). Hs research nterests ncude the areas of embedded systems, cyber-physca systems, the Internet of Thngs, and mobe soca networs. Je Lu (ue.dut@gma.com) receved hs B.E. from DUT n 0. He s Master s student n the Schoo of Software, DUT. He obtaned an nventon patent n 0. He s a member of SmartCPS Lab. Hs research nterests cover robustness optmzaton of the Internet of Thngs. Wesheng S (w.s@westernsydney.edu.au) s currenty a ecturer n the Schoo of Computng, Engneerng and Mathematcs, Western Sydney Unversty. Pror to ths, he was a postdoctora researcher at Natona ICT Austraa. He receved hs Ph.D., M.S., and B.S. degrees n computer scence from the Unversty of Sydney, the Unversty of Vrgna, and Peng Unversty, respectvey. Hs research nterests ncude routng and topoogy contro n wreess networs, graph theory, and data center networs. Mn Han (mnhan@dut.edu.cn) [M, A 0, SM 0] receved her M.S. and Ph.D. degrees from Kyushu Unversty, Fuuoa, Japan, n and, respectvey. Snce 00, she has been a professor n the Facuty of Eectronc Informaton and Eectrca Engneerng, DUT. She s the author of four boos and more than 00 artces. Her current research nterests are neura networs, chaos, and ther appcatons to contro and dentfcaton. Huansheng Nng (nnghuansheng@ustb.edu.cn) receved hs B.S. degree from Anhu Unversty n and hs Ph.D. degree from Behang Unversty n 00. Now, he s a professor and vce dean of the Schoo of Computer and Communcaton Engneerng, Unversty of Scence and Technoogy Beng, Chna. Hs current research focuses on the Internet of Thngs and cyber-physca modeng. He s the founder of the Cyberspace and Cybermatcs Internatona Scence and Technoogy Cooperaton Base. Mohammed Atquzzaman (atq@ou.edu) s a professor of computer scence at the Unversty of Oahoma. Hs research nterests and pubcatons are n next generaton computer networs, wreess and mobe networs, swtchng and routng, optca communcatons, and mutmeda over networs. He serves as the Edtor-n-Chef of the Journa of Networ and Computer Appcatons and the Vehcuar Communcatons ourna, and an Assocate Edtor of IEEE Communcatons Magazne, among others. IEEE Communcatons Magazne December 0