Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks

COMCOM 4842

No. of Pages 11, Model 5G

23 September 2013 Computer Communications xxx (2013) xxx–xxx 1

Contents lists available at ScienceDirect

Computer Communications journal homepage: www.elsevier.com/locate/comcom 5 6

Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks

3 4 7 8 9 10

Q1

a

Department of Computer Science and Engineering, Thapar University, Patiala, Punjab, India Department of Computer Science and Computer Engineering, LaTrobe University, Melbourne, Australia c Instituto de Telecomunicações, University of Beira Interior, Rua Marques D’Avila e Bolama, Covilhã, Portugal b

Q3

11

a r t i c l e

1 2 3 2 14 15 16 17 18 19 20 21

Neeraj Kumar a, Naveen Chilamkurti b,⇑, Joel J.P.C. Rodrigues c

i n f o

Article history: Available online xxxx

Q4

Keywords: Vehicular Ad Hoc Networks Opportunistic forwarding Data aggregation Learning automata

a b s t r a c t Due to the highly mobile and continuously changing topology, the major problem in Vehicular Ad Hoc Networks (VANETs) is how and where the collected information is to be transmitted. An intelligent approach can adaptively select the next hop for data forwarding and aggregation from the other nodes in the networks. To address these issues, we propose a Learning Automata-based Opportunistic Data Aggregation and Forwarding (LAODAF) scheme for alert generation in VANETs. Learning automata (LA) operate separately which are deployed to the nearest Road Side Units (RSUs) to collect and forward the data from respective regions along with alert generation. Once data is aggregated, LA adaptively selects the destination for data transfer, based on the newly defined metric known as Opportunistic Aggregation and Forwarding (OAF). LA predicts the mobility of the vehicle and adaptively selects the path for forwarding, based on the value of OAF. Moreover, it updates its action probability vector and learning rate based on the values of OAF. This will reduce network congestion and the load on the network as it is aggregated and forwarded only when required. An algorithm for opportunistic data aggregation and forwarding is also proposed. The proposed strategy is evaluated using various metrics, a number of successful transmissions, connectivity, link breakage rate, traffic density, packet reception ratio, and delay. Ó 2013 Published by Elsevier B.V.

23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39

40 41

1. Introduction

42

The advancements in wireless technology and embedded systems have led to the development of next generation intelligent transport systems (ITS). The most common networks used in ITS are Vehicular Ad Hoc Networks (VANETs). VANETs have many application domains to enhance the safety of the driver and passengers. Vehicles/nodes share information with each other that can be helpful in making an adaptive decision. Vehicles can communicate with each other, which is called vehicle-to-vehicle (V2V), or they can communicate with the infrastructure which is called vehicle-to-infrastructure (V2I). Most modern vehicles have information on board about the outside parameters, such as the density of the traffic, a map of the places to be visited/has visited, pollution quantity, duration of traffic lights on the next intersection etc., which can be used by drivers to make adaptive decisions about the selection of an appropriate route and to reduce the pollution on the road, while the passengers on board can use

43 44 45 46 47 48 49 50 51 52 53 54 55 56 57

Q2

⇑ Corresponding author. E-mail addresses: [email protected] (N. Kumar), n.chilamkurti@latrobe. edu.au (N. Chilamkurti), [email protected] (J.J.P.C. Rodrigues).

various types of resources provided by the infrastructure via the Internet, which is known as Internet Vehicular Ad Hoc networks (IVANETs). VANETs are traditionally different from mobile Ad Hoc networks (MANETs) due to the provision of valuable features such as well-defined routes, mobility patterns and information on the environment in which they are operating [1–3]. Although VANETs can be used in many applications, there are still many challenges facing these networks which need to be addressed. The biggest challenge in VANETs is the continuous topological changes due to the high mobility of the nodes. Moreover, interference from neighboring vehicles, disconnection of the nodes in some regions of the network and limited communication range are other factors which severely affect the performance of wireless networks. As discussed above, data is collected by vehicles with sophisticated devices from different locations across dense urban or sparse regions. But due to the high mobility of the nodes, data aggregation is a challenging task in VANETs. Moreover, the aggregated data needs to be transferred to other nodes so that it can reach the final destination. But each collected piece of data does not need to be sent at once to all the nodes as this will create an overload on the network, causing performance degradation. Hence, there is a

0140-3664/$ - see front matter Ó 2013 Published by Elsevier B.V. http://dx.doi.org/10.1016/j.comcom.2013.09.005

Please cite this article in press as: N. Kumar et al., Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks, Comput. Commun. (2013), http://dx.doi.org/10.1016/j.comcom.2013.09.005

58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79

COMCOM 4842

No. of Pages 11, Model 5G

23 September 2013 2 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130

N. Kumar et al. / Computer Communications xxx (2013) xxx–xxx

need for an intelligent approach which takes care of incoming traffic requests and transfers them according to the network conditions such as load on the nodes, resource availability etc. For data aggregation and forwarding in VANETs, many proposals have used broadcasting and multicasting mechanisms from the source node [4–7]. In these proposals, authors have constructed the minimum spanning tree to broadcast the data. Most recently, Ruiz et al. [8] proposed an information dissemination mechanism based upon tree topology, proposing a decentralized broadcasting algorithm for continuous topological changes in VANETs. Moreover, a tree topology construction algorithm is also proposed by the authors. In addition to this approach, researchers have also used multipath routing schemes to mitigate the frequent disconnection problem in highly dynamic networks. Ad Hoc on demand Multipath Distance Vector (AoMDV) [9] is one such protocol which uses this technique. In this scheme, if one of the paths fails for some reason, then another path may be taken as an alternative to complete the route. From the above discussion, it is clear that there is a need for an efficient mechanism for opportunistic data aggregation and forwarding for highly dynamic VANETs. The vehicles in VANETs should have built-in intelligence mechanisms to adapt to the network conditions. They should have the capability to make the best decision about when and where the data aggregation and forwarding technique should be applied [10]. This is necessary as during high mobility scenarios, vehicles need to communicate with each other to share information. Vehicles which are equipped with intelligence mechanisms can be used in a wide range of applications in both dense urban and sparse regions, e.g., to generate alarms for safety operations, military applications etc. Hence, to address these issues, we propose a new Learning Automata-based Opportunistic Data Aggregation and Forwarding (LAODAF) scheme for VANETs. Learning automaton (LA) operates separately in each vehicle in which it is deployed and collects data in its respective region as the vehicle moves from one place to another. LA adaptively selects the destination for data transfer, based on the newly defined metric by considering various factor such as load, available resources etc. The moving vehicles share information with each other using LAs. Each action taken by an LA is either rewarded or penalized by the environment in which it is operating. Based on the input provided by the environment corresponding to the action taken by the LA, the future course of action is decided, i.e., in which direction the LA will take its action. The main contributions of the paper are as follows: 1. A new LA-based opportunistic data aggregation and forwarding scheme has been proposed 2. A metric OAF has been proposed which is used by the LA for forwarding the collected data at respective nodes in the network 3. Mobility of the nodes is predicted, based on their current position which is used by the automaton to make the decision to forward the data or buffer the data opportunistically

131

139

The rest of the paper is organized as follows. Section 2 describes the most relevant related work in this area. Section 3 provides an overview of the LA approach. Section 4 describes the problem statement. Section 5 describes the proposed approach. In this section, the description about the proposed algorithm is also provided. The simulation environment with results and discussions are described in Section 6. Section 7 provides the conclusions and future work.

140

2. Related work

141

Over the years, VANETs have emerged as a new technology which is used in a wide area of applications with the ultimate goal

132 133 134 135 136 137 138

142

to provide maximum safety to the user and the passengers on board. However, VANETs present a number of challenges to researchers in this domain to provide new ideas and solutions for maximum benefits. To address these challenges, Ruiz et al. [8] proposed a new mechanism for information dissemination in VANETs, using tree topology, that is, a decentralized algorithm for the construction of tree topology in VANETs. Li et al. [11] used a directional antenna approach for broadcasting in VANETs. Tonguz et al. [12,13] proposed a broadcasting algorithm for topology maintenance with respect to the traffic patterns generated both from urban and sparse regions. Bai et al. [14] presented a multi-hop broadcasting mechanism for various types of safety message dissemination in VANETs. Slavik et al. [15] proposed a stochastic scheme for broadcasting in VANETs. These authors also proposed security architecture for secure data dissemination using the proposed algorithm. Mylonas et al. [16] proposed a speed adaptive probabilistic forwarding scheme and considered various parameters such as speed and delivery delay for the vehicles in dense urban regions. Huang et al. [17] proposed an opportunistic scheme for generating a collision warning in VANETs. As VANETs can be used in a wide range of applications in the future, researchers are working to design new protocols which are capable of providing various services such as mobile marketing, mobile multimedia and social networking. For all such applications, there is a need to upgrade on-board hardware [18]. Moreover, there is a need for powerful sensor nodes capable of sensing the environment in which they are operating so that they can collect data from the environment and disseminate this as per the requirements [19,20]. Most recently, Pau et al. [21] outlined the various challenges and opportunities in vehicular sensing networks. The authors proposed an architecture for VANETs which can be used for pollution control in dense urban regions and can also be used in designing the next generation ITS. Recent studies have shown that pollution control is a serious problem in dense urban regions, which can be controlled by using intelligent computational techniques [22,23]. Giordano et al. [24] proposed a realistic urban propagation model for data dissemination in VANETs. Schwartz et al. [25] presented a directional dissemination protocol for vehicular networks. The authors presented a novel mechanism to overcome the broadcast storm problem in dense regions using the store and forward technique. Borsetti et al. [26] presented an application level framework for information dissemination in VANETs where vehicles can share the application level role with each other by using the mobility patterns of vehicles. Li et al. [27] presented an opportunistic event-driven broadcasting scheme for loss links in VANETs. An opportunistic protocol was proposed by the authors, which is capable of maintaining a high packet reception ratio and multi-hop message transmissions. Moreno et al. [28] presented contention-based opportunistic data dissemination in VANETs. Moreno et al. [29] presented vehicle-to-vehicle communication for safety critical applications. Moreno et al. [30] proposed a contention-based packet forwarding scheme for VANETs. In these schemes, the contention period is considered and the distant node is opportunistically selected for transmission. Vinel et al. [31] provide a detailed analysis of trustworthy broadcasting in vehicular networks. Campolo et al. [32] described the broadcast packet loss in vehicular networks. Computation intelligence techniques can also be applied to solve many real time problems in wireless networks. In this regard, researchers have found that LAs can be very useful in many applications [33–39] as it is an optimization approach in which an action is taken by LA by using its own knowledge and then using this knowledge to adapt to the situation [31–37].

Please cite this article in press as: N. Kumar et al., Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks, Comput. Commun. (2013), http://dx.doi.org/10.1016/j.comcom.2013.09.005

143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208

COMCOM 4842

No. of Pages 11, Model 5G

23 September 2013 3

N. Kumar et al. / Computer Communications xxx (2013) xxx–xxx

Fig. 1. Learning automata.

209

3. Overview of learning automata

210

Fig. 1 describes the working principle of the proposed LA. An automaton takes some input parameters and then takes an action to produce an output. It is an optimization technique that can be applied in various domains to solve a given problem. Moreover, the machine has the capability of learning from its environment so that it can choose the desired action from a finite set of allowed actions through repeated steps [33–39]. The initial action chosen is random in nature, but by taking input from the environment in terms of reinforcement signals, after a finite number of steps, the solution converges and produces the desired output. Mathematically, an LA is defined by a 4-tuple as hK; V; P; Ti, where K is the input provided by the environment in discrete form, V denotes the set of actions performed by the automaton, P is the state probability distribution, and T represents the reinforcement updates [33–39]. The environment in which the automaton operates can be defined as a triplet hx; y; zi, where x ¼ ðx1 ; x2 ; . . . ; xn Þ are finite number of inputs, y ¼ ðy1 ; y2 ; . . . ; yn Þ are values of reinforcement signal, and q ¼ ðq1 ; q2 ; . . . ; qn Þ are penalty probability associated with each xi ; 1 6 i 6 n. The automaton performs a finite number of actions, and based upon their actions, the response of the environment can be either a reward or a penalty. According to the response received from the environment, LA decides its action by taking the reinforcement signal to which stage it has to move. For each action selected by the LA, the environment gives a reinforcement signal [40,41]. Corresponding to each input parameter and action, LA updates its action probability vector by using the learning algorithm. The action probability matrix is updated based upon the input that LA receives from the environment in which it is operating. In the proposed scheme, we have considered the Linear reward–inaction scheme, in which if the automaton receives a reward from the environment, then the action probability is updated, else the probability remains the same [37,42–44]. The formula for the probability update is as follows:

211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242

243

pj ðn þ 1Þ ¼ ð1  aÞpj ðnÞ; j–i; y ¼ 0 pj ðn þ 1Þ ¼ apj ðnÞ; j ¼ i; y ¼ 0 245

pj ðn þ 1Þ ¼ pj ; y ¼ 1

246

where a is the parameter.

247

4. Problem statement

248 249 250 251 252 253 254 255 256

ð1Þ

4.1. System model and assumptions Fig. 2 shows the system model used in the proposed scheme for VANETs. When the vehicles are mobile, all the resources are provided to them by the Road Side Units (RSUs). The RSUs are connected to the Internet so that users have full access to all the resources of the network. In the proposed system model, we assume that LAs are deployed in each vehicle which will provide built-in intelligence to each vehicle. Vehicles communicate with each other using the on-board units (OBUs) and are equipped with

GPS devices to find each other’s location. If the GPS is not available in certain places, then vehicles communicate with each other by estimating the relative with distance from the other vehicles [27]. As the vehicles are moving at high speed from source to destination, their transmission range also varies with time. We take this range as Dmax as the measure of how long vehicles can communicate with other vehicles to perform the action of data aggregation and forwarding. LAs operate in a collaborative manner and pass the aggregated data to other vehicles depending on the available resources at that time, i.e., network load status, mobility of the vehicles or connectivity of the nodes etc. By considering these parameters, opportunistic forwarding of data is performed by the vehicles. The automaton is located at the nearest RSUs/Access points. The automaton may communicate with each other using the existing infrastructure as shown below.

257

4.2. Objectives

272

Due to the highly dynamic nature of VANETs, the topology of the network changes continuously. In this context, the alert messages generated from the source may not reach the final destination due to link breakage at certain points which causes performance degradation in the network. Hence, we use an LAbased approach to mitigate the link breakage problem in VANETs. In the proposed scheme, LA adaptively switches from the dense urban regions to the sparse regions to maintain the connectivity of the nodes. So, in view of this, the objective of the current proposal is to have highest Packet Delivery Ratio (PDR) without much delay from the source within the high values of connectivity of the nodes. G = (V, E) is the graph which represents the vertices V ¼ fV 1 ; V 2 ; . . . ; V n g and edges E; V ¼ fV 1 ; V 2 ; . . . ; V n g are the nodes/ vehicles in the network; L ¼ fLA1 ; LA2 ; . . . ; LAn g are the LAs operating in different regions of the network under consideration; a ¼ fa1 ; a2 ; . . . ; an g are the learning rates of each of these LAs defined above; and E ¼ fl1 ; l2 ; . . . ; ln g describes the number of links available. The strategy consists of providing some training to LA in the environment where they are operating. LAs use opportunistic forwarding in some parts of the network depending on the parameters such as load, connectivity and mobility of the nodes. Corresponding to each input parameter and action, LA updates its action probability vector by using the learning algorithm as defined in Eq. (1) above. The connectivity of the nodes is measured in terms of link breakage rate during a particular time interval due to the mobility of the nodes. The mobility of the nodes is measured in terms of relative velocity of the nodes and load is measured in terms of the number of requests satisfied in relation to the total number of requests generated. These factors are calculated separately as follows: Let pv be the total number of packets received by vehicle v in the transmission range. Dmax and n are the total number of vehicles in a particular region and the number of packets transmitted or received in the transmission range of the vehicle. Moreover, let ½t1 ; t2 ; ½t3 ; t4 ; ½t5 ; t6     ½tn1 ; tn  be the different time intervals for various events on the road. For the sake of simplicity, we assume that the transmission range of the vehicles does not change within Dmax . We define a new metric, called Load Factor Value (LFV), to take care of the number of packets transmitted or received to the total number of packets transmitted as follows

273

LFV ¼

Z 0

Dmax

Z ti

tj

 Pn  i¼1 pv P dtdR; n ni¼1 pv

where 0 6 R 6 Dmax

ð2Þ

In the above equation, LFV is the measure of the number of packets correctly received between source and destination within the transmission range of each other. The integration of different values of

Please cite this article in press as: N. Kumar et al., Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks, Comput. Commun. (2013), http://dx.doi.org/10.1016/j.comcom.2013.09.005

258 259 260 261 262 263 264 265 266 267 268 269 270 271

274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313

314 316 317 318 319

COMCOM 4842

No. of Pages 11, Model 5G

23 September 2013 4 320 321 322 323 324 325 326 327 328 329 330

331

N. Kumar et al. / Computer Communications xxx (2013) xxx–xxx

successful packets transmission during different time intervals for all the vehicles within the transmission range of each other is computed. The values of LFV is useful for LA to take adaptive decisions about the selection of particular path compared to other paths. As vehicles are continuously moving on the road, so their location and relative speed to each other also varies. Let Lnew ; Lold be the new and old locations of the vehicles and V speed be the speed of the vehicle in any time interval as defined above. Each vehicle shares its location information, time, identity and direction of motion with the other vehicles. The direction of motion of the vehicle (h) is calculated as follows:

ð3Þ 333 334 335 336 337

338

As shown in Eq. (3), h has two values +1 if the vehicles are moving towards the source from which alert messages are generated and 1 if the vehicles are moving away. The combined values for Probabilistic Mobility Prediction ðPMPÞ are summarized as follows: new

340 341 342 343

344 346 347 348 349 350

351

L

¼L

old

þ

Z

tj

hV speed dt

ð4Þ

ti

The value of PMP is the ratio between the change in the position of an individual vehicle to the total number of changes of position of all the vehicles and is given as follows

Lnew PMP ¼ Pn new v ¼1 L

ð5Þ

Based on Eqs. (4) and (5), the connectivity ratio of the links is calculated. This is the measure of the number of links which are broken due to poor connectivity in some regions compared to the total number of links available in set E. kt

353 354 355 356 357 358 359 360 361

362 364

365 366 367 368 369

370 372 373 374 375

e CR ¼ Pn

i1 li

ð6Þ

where k is the link breakage rate. The reason for link breakage is the mobility of the nodes or poor connectivity due to the change of the position of the vehicles as defined in Eq. (4). The values of k has a greater impact on the performance of the network as they measure the stability of the nodes in a particular region. Combining all these values, a new metric called Opportunistic Aggregator and Forwarding (OAF), is introduced in the current proposal, calculated as follows:

OAF ¼ minðLFV þ PMP þ CRÞ

ð7Þ

OAF is evaluated at each intermediate node before making a decision about aggregation and opportunistic forwarding. In this case, LA makes the decision about opportunistic forwarding and aggregation. The following function is defined for the operation of LA

g : LAi ! OAF for 1 6 i 6 n

ð8Þ

Eq. (7) illustrates that each LA matches the OAF metric for making the decision about the packet forwarding. LA selects the node for forwarding the packet, which has the minimum value of OAF.

376

5. Proposed solution

377

The proposed solution consists of mobility prediction with opportunistic data aggregation and forwarding, and updating the action probability vector. Each of these activities is maintained by LA. The sequence of activities performed by the LA is as follows. The basic principle on which LA works is that it selects one of the solutions at the initial stage and then, as per the response received from the environment in which it is operating, decides its future

378 379 380 381 382 383

action. For each input received from the environment, it updates its action probability vector or buffers its value to constant. In the current proposal, we have not proposed any new technique for vehicles clustering but we have used the exiting vehicular technique defined in [45].

384

5.1. Initialization and activation

389

This phase is the source of alert generation, i.e., any activity happening at this stage should be intimated to all the nodes in the network so that they can take action accordingly. The event is generated at the source and the LA forwards or stores the same as per the available resources in the network. Each node has the LA operating on it, performing all the activities on that node such as receiving the message, processing the message or forwarding the message. Messages generated from the initial node are processed by the LA at that node and forward the alert to the next hop. After receiving the message from the source node, each intermediate node activates the LA and updates its information with the new message.

390

5.2. Mobility prediction

402

As the LA is activated at each intermediate node, the direction of each vehicle’s motion is estimated. Two values are considered in the proposed scheme: one is the vehicles which are moving towards the source and the other is the vehicles which are moving away from the same or different lanes. The position of the vehicle is measured in terms of ðx; yÞ coordinates of the vehicle. These coordinates can be obtained by the GPS system. As the vehicle moves from one location to another, its position and coordinates also vary with time. The corresponding LA operates on the node and calculates the new and old position of the vehicle using Eq. (4). After which it calculates the PMP of each vehicle using Eq. (5). Once the mobility pattern is analyzed by the LA, it uses PMP for opportunistic forwarding. The following conditions are checked to make a decision as to whether to forward the packet and update the action probability vector or buffer the data and not update the action probability vector.

403

Poppor ij

 8  PMP > !k < D > > n  < X PMP  k ¼ D ¼ 1  > > V¼1 k > D > :  0

D ¼ Dmax þ



Rq

1e



387 388

391 392 393 394 395 396 397 398 399 400 401

404 405 406 407 408 409 410 411 412 413 414 415 416 417 418

419

421

ð10Þ

v ¼1

where R is the transmission range on the road segment and q is the traffic density in a particular road segment.

N Now; q ¼ Pn

386

ð9Þ

where Poppor is the probability of opportunistic forwarding of the ij packet and D is the delay in sending the packet. As can be seen from Eq. (9), three conditions are checked by the LA for performing opportunistic forwarding or aggregation. If the value of k is greater than D, then is does not update its action probability vector and only performs aggregation of the message, i.e., messages are buffered in the current node in which it is operating, else it is forwarded with the probability of PMP. When it is equal to D, then PMP is subtracted from 1. The value of D is calculated as follows: n X

385

422 423 424 425 426 427 428 429 430 431

432

434 435 436

437

Si

i¼1 Si

ð11Þ

where S ¼ fS1 ; S2 ; . . . ; Sn g and N Si are the average number of vehicles in a particular segment.

Please cite this article in press as: N. Kumar et al., Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks, Comput. Commun. (2013), http://dx.doi.org/10.1016/j.comcom.2013.09.005

439 440 441

COMCOM 4842

No. of Pages 11, Model 5G

23 September 2013 5

N. Kumar et al. / Computer Communications xxx (2013) xxx–xxx

Internet

RSU

RSU

RSU

RSU

Dmax Dmax Dmax Dmax Dmax Dmax

Fig. 2. System model in the proposed scheme.

442 443 444 445 446 447

Each action taken by the LA is either rewarded or penalized as defined in Eq. (1). If the action of the LA is rewarded by the environment, then it updates its action probability vector, else the value of the action probability vector remains the same. This process continues until the alert generated from the source does not reach the final destination.

448

5.3. Reward and penalty functions

449

For each action taken by the automaton, its action may be rewarded or penalized by the environment where it is operating. The environment may give a positive or negative feedback with some constant values for each action taken by the automaton and accordingly its action probability vector is updated by the Eq. (1) as defined above. As we have considered the LRI policy [39,43,46,47] for reward and punishment, so corresponding to each reward from the environment, the action probability value is increased by some constant value else it remains unchanged. Find the probability of CR for the moving vehicles on the road as P r ðCRÞ ¼ CR . The reward and penalty functions are defined as n follows:

450 451 452 453 454 455 456 457 458 459 460

Reward function If (current node==destination) then OAF = OAF + w n Path = OAF Else if ðOAF 6 thrÞ then OAF = OAF + maxððpathÞn1 P r ðCRÞÞ n Path = OAF Else n Path = OAF 471 472 473 474 475 476

Where w is a constant having values [0.2, 0.7]. In the above reward function, if the current node is the destination then value of OAF is incremented by w and the value of OAF is assigned to the associated path used for selection of data aggregation and selection. In the current proposal, we have assumed this path as the

n

Path else if the values of OAF are less than the predefined threshold then the value of OAF is get incremented by the constant value chosen from maximum of all previous paths and their associated CR. The penalty function is described as follows:

477 478 479 480

Penalty function If (current node==destination) then OAF = OAF/w Pathn = OAF Else OAF = OAF + minððpathÞn1 P r ðCRÞÞ Pathn = OAF

490

The penalty function is same as that of the reward function except that in this case we have divided the OAF by w. So as soon as some penalty is applied to a path for opportunistic forwarding then it becomes less favorable as compared to the other paths. Hence the chances of selection of this path becomes very less. For selection of the paths other than the case when source and destination are equal, we have selected only the case when probability of CR among all the paths is minimum in comparison to the reward function. Hence the chances of selection of best path is increased which ultimately improves the overall performance of the system.

491

5.4. LAODAF algorithm

501

The algorithm consists of calculation of PMP, OAF and CR based upon which opportunistic forwarding is performed in a particular geographical location. The algorithm is executed by the automaton and sequences of operations are performed iteratively based upon the penalty and reward obtained from the environment. Each time a reward is obtained the probability of section of particular route is increased while the probability is decreased if corresponding to the action taken by the automaton is penalized. Based upon these parameters, automaton selects its next action to be performed. The detailed algorithm is described as follows:

502

Please cite this article in press as: N. Kumar et al., Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks, Comput. Commun. (2013), http://dx.doi.org/10.1016/j.comcom.2013.09.005

492 493 494 495 496 497 498 499 500

503 504 505 506 507 508 509 510 511

COMCOM 4842

No. of Pages 11, Model 5G

23 September 2013 N. Kumar et al. / Computer Communications xxx (2013) xxx–xxx

Learning Automata-based Opportunistic Data Aggregation and Forwarding (LAODAF)Inputs: V Speed , PMP, OAF, ai Output: Opportunistic forwardingParameters:

LA action is penalized as follows Ai ðn þ 1Þ ¼ maxðOAFÞ Jþþ Update the action probability vector as defined in Eq. (1) K ¼ LJ Select the maximum value of the variable K End Procedure

53. 54. 55.

The algorithm starts with number of iterations and number of nodes and as soon as some event is generated from the source node, the LAs operating on the intermediate nodes are activated (lines 1–5). The LA calculates the direction of the motion of the vehicle (moving away or towards the source). Using these values, the position, PMP and CR of the vehicles are calculated (lines 6–14). Using the intermediate distance between the vehicles and density of the vehicles in the particular regions, the probability of opportunistic forwarding is computed (lines 16–18). Based on the link breakage rate and distance between the vehicles, the OAF values are updated and replaced with the new values if the newly computed values are better than the existing ones (lines 19–36). Each action taken by the LA is either rewarded or penalized. Moreover, the values of the action probability vector are also updated and the number of times the action of the LA is rewarded or penalized is also computed (lines 42–52). The ratio of the number of times the action of the LA is rewarded compared to the number of times the action of the LA is penalized is calculated and the maximum value of this ratio is selected for the next iteration (lines

Table 1 Parameters and their values. Parameters

Values

Number of vehicles Velocity of vehicles Communication range Vehicles arrival rate

500 50–60 km/hr 100 m 20/min

500

nsmissions Number of tra

400

300

200 0.9

100

fo rw ar di ng

L: The number of times an action i is rewarded J: The number of times an action i is penalized V: Total number of times action i has been taken Ai: Action performed by LA 1. forði ¼ 1; i 6 number iteration; i þ þÞ 2. forðV ¼ 1; V 6 n; V þ þÞ 3. Start form the root node from where event is generated 4. forðLA ¼ 1; LA 6 n; LA þ þÞ 5. Activate the LA at the node 6. Calculate the direction of movement of the vehicles 7. if ðh ¼ þ1Þ 8. Vehicles are moving towards the source node 9. Else 10. Vehicles are moving away from the source 11. For each time interval t 12. Calculate the position of the vehicles using Eq. (4) 13. Calculate the PMP using Eq. (5). 14. Calculate the CR of the vehicles using Eq. (6) 15. repeat 16. Calculate the distance D using Eq. (10) 17. Calculate the density of the region using Eq. (11) 18. Calculate the probability of opportunistic forwarding as follows 19. If ðk < DÞ then 20. Poppor PMP ij 21. Exchange the information with the other LAs 22. DOAF ¼ OAF  OAFnew 23. if ðDAF < Current OAFÞ then 24. Set Current OAF DAF 25. Else 26. Continue with the current OAF value 27. Else if ðk ¼ DÞ P 28. Poppor 1  nV¼1 PMP ij 29. Exchange the information with the other LAs 30. DOAF ¼ OAF  OAFnew 31. if ðDAF < Current OAFÞ then 32. Set Current OAF DAF 33. Else 34. Continue with the current OAF value 35. Else 36. Poppor 0 ij 37. Call ProcedureðLA;ai ; L; J; V; Ai Þ 38. untilðR 6 Dmax Þ 39. End for 40. End for 41. End for 42. ProcedureðLA; ai ; L; J; V; Ai Þ 43. if ðk 6 DÞ then 44. LA action is rewarded as 45. Ai ðn þ 1Þ ¼ minðOAFÞ 46. Update the action probability vector as defined in Eq. (1) 47. Lþþ 48. Else 49. LA action is penalized as follows 50. Ai ðn þ 1Þ ¼ maxðOAFÞ 51. Jþþ

49. 50. 51. 52.

Pr ob ab ilit yo fo pp or tu ni tic

6

0.8

0

0.7

90

Traff ic

80

0.6

70

60 dens ity (n 50 umb er of vehic les

40

/kam )

0.5

Fig. 3. Opportunistic forwarding with number of transmissions.

Please cite this article in press as: N. Kumar et al., Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks, Comput. Commun. (2013), http://dx.doi.org/10.1016/j.comcom.2013.09.005

512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530

COMCOM 4842

No. of Pages 11, Model 5G

23 September 2013 7

N. Kumar et al. / Computer Communications xxx (2013) xxx–xxx 531 532 533 534

53–54). The LA learns from its previous actions in relation to whether the action was rewarded or penalized. Initially, the process of learning is slow but with the passage of time, it adapts to the situation quickly and finally converges to the solution.

535

6. Simulation results and discussion

536

The proposed scheme is evaluated using Vanet Mobi Sim [48] with respect to various metrics such as the number of successful transmissions, connectivity, link breakage rate, traffic density, packet reception ratio and delay. The values of the various parameters are summarized in Table 1.

1.2

1.0

0.6 60

vehic les

50

40

/km)

0.5

ilt iy o

70 dens ity (N umbe r of

ab

Traffic

80

ob

90

fo

0.7

pp o

0.2

r tu nis

0.8

tic f

0.9

0.4

or

0.6

wa rd ing

0.8

Pr

540

Fig. 4. Opportunistic forwarding with probability of connectivity.

541

6.1.1. Impact of probability of opportunistic forwarding on number of transmissions Opportunistic forwarding consists of data aggregation and transfer of the same as per the requirements; hence this mechanism improves the use of the number of resources in the network. Fig. 3 shows the impact of the probability of opportunistic forwarding and traffic density on the number of transmissions. As shown in Fig. 3, with an increase in probability of opportunistic forwarding and traffic density, the total number of transmissions increased linearly. This is due to the fact that with an increase in probability of opportunistic forwarding, there are lots of chances for the vehicles to forward messages to other vehicles. Moreover, as the automatons are assumed to be located on vehicles, these vehicles can communicate share information with each other in a collaborative manner. This mechanism allows more and more data to be sent towards the destination in the event of an emergency situation such as an alert generation in the case of fire or a road accident. Moreover, the automaton learns from the environment in which it is operating and, based on the feedback provided by the environment, decides its future action. Hence, the proposed approach is adaptive and the respective automaton is capable of making independent decisions for forwarding the message to its final destination, hence there is an increase in the number of transmissions in the proposed scheme.

542

6.1.2. Impact of opportunistic forwarding probability on connectivity As nodes in VANETs are highly mobile in nature, connectivity is required for efficient operations such as data aggregation and forwarding in a particular time interval. Fig. 4 shows the impact of the probability of opportunistic forwarding on the probability of connectivity. As shown in Fig. 4, with an increase in probability of opportunistic forwarding and traffic density, the connectivity probability of the nodes is also increased. The reason for this increase in connectivity is due to the fact that with an increase in opportunistic forwarding probability, the LA operating on the

566

1 0.9

Link breakage rate=25 links/ms Link breakage rate=20 links/ms

0.8

Link breakage rate=15 links/ms Link breakage rate=10 links/ms

0.7

Link breakage rate=5 links/ms

Connectivity ratio

539

nnectivity

538

Probabiltiy of co

537

6.1. Discussion

0.6 0.5 0.4 0.3 0.2 0.1 0 0

20

40

60

80

100

120

Traffic density (Number of vehicles/Km) Fig. 5. Connectivity ratio with different link breakage rates.

Please cite this article in press as: N. Kumar et al., Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks, Comput. Commun. (2013), http://dx.doi.org/10.1016/j.comcom.2013.09.005

543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565

567 568 569 570 571 572 573 574 575

COMCOM 4842

No. of Pages 11, Model 5G

23 September 2013 8

578 579 580 581 582

vehicles has more chances to route the packets to their destination as it does not store the data, which enhances the connectivity of the nodes in the network. Moreover, as the LA has more chances for opportunistic forwarding due to the nature of making independent decisions, the connectivity of the nodes is also increased. Moreover, during the communications among the respective automaton, major decisions about opportunistic forwarding are

made in collaboration with the respective automaton. Also, according to the feedback provided by the environment, the automaton decides its action and performs the respective operation of sending the information from the source to destination, which makes the approach more adaptive. Hence, the proposed approach has a higher probability of connectivity with an increase in probability of opportunistic forwarding.

1.1 Link breakage ratio=25 links/ms Link breakage ratio=20 links/ms

1 Link breakage ratio=15 links/ms Link breakage ratio=10 links/ms

Packet reception ratio

577

Link breakage ratio=5 links/ms

0.9

0.8

0.7

0.6

0.5 0

20

40

60

80

100

120

Traffic density (Number of vehicles/Km) Fig. 6. Link breakage rate with packet reception ratio.

Link breakage rate=25 links/ms

40

Link breakage rate=20 links/ms Link breakage rate=15 links/ms Link breakage rate=10 links/ms

30

Link breakage rate=5 links/ms

Broadcast delay (sec.)

576

N. Kumar et al. / Computer Communications xxx (2013) xxx–xxx

20

10

0

0

20

40

60

80

100

120

Traffic density (Number of vehicles/Km) Fig. 7. Impact of vehicle density on broadcast delay with link breakage rate.

Please cite this article in press as: N. Kumar et al., Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks, Comput. Commun. (2013), http://dx.doi.org/10.1016/j.comcom.2013.09.005

583 584 585 586 587 588 589

COMCOM 4842

No. of Pages 11, Model 5G

23 September 2013 9

N. Kumar et al. / Computer Communications xxx (2013) xxx–xxx

593 594 595 596 597 598 599 600 601 602

(a)

As the link breakage rate decreases from 25 to 10 links/ms, there is an 85% increase in the connectivity ratio with an increase in traffic density. The reason for the increase in connectivity ratio, even with an increase in traffic density, is due to the presence of the LA at each node which controls all the operations such as connectivity, link breakage etc., so that the alert generated from the source node can be transmitted to its final destination without any delay. During the initial phase of the operation of automaton, sometime delay was observed, but once it is activated and starts functioning on respective nodes, then as per the feedback provided to it by the environment, it makes adaptive decisions and finds the best suitable destination for each of the messages generated from the source, i.e., the learning of the automaton is slow in the begin-

1

0.8 Oppurtunistic Forwarding probability

592

6.1.3. Impact of varying the link breakage rate on connectivity ratio with traffic density As the density of the vehicles varies in different parts of the network, the link breakage rate in the network also changes according to the density of the vehicles. This may happen due to the high mobility of the nodes in the network. Fig. 5 shows the impact of the link breakage rate on connectivity ratio with an increase in traffic density. As shown in Fig. 5, with an increase in the link breakage rate, the connectivity ratio falls. This is due to the fact that at a high link breakage ratio, routes are not stabilized and the connectivity of the nodes is not maintained at a particular level. However, even with an increase in network density, the proposed LA-based approach maintains more than 80% connectivity.

0.6

Learning rate=0.2

0.4

Learning rate=0.4 Learning rate=0.6

0.2

Learning rate=0.8

0 0

100

200

300

400

500

600

Number of Vehicles /Km

(b)

1

0.8

0.6 Packet Reception ratio

590 591

Learning rate=0.2

0.4

Learning Rate=0.4 Learning Rate=0.6 0.2

Learning Rate=0.8

0 0

10

20

30

40

50

60

70

80

90

100

Number of Vehicles/Km Fig. 8. Impact of learning rate on opportunistic forwarding probability and packet reception ratio.

Please cite this article in press as: N. Kumar et al., Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks, Comput. Commun. (2013), http://dx.doi.org/10.1016/j.comcom.2013.09.005

603 604 605 606 607 608 609 610 611 612 613 614 615

COMCOM 4842

No. of Pages 11, Model 5G

23 September 2013 10 616 617 618

619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641

642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661

662 663 664 665 666 667 668 669 670 671 672 673 674 675

N. Kumar et al. / Computer Communications xxx (2013) xxx–xxx

ning but with the passage of time, it improves and makes better decisions. This is the reason why the behavior shown in Fig. 5 is observed in the proposed scheme. 6.1.4. Impact of link breakage rate on packet reception ratio with varying traffic density Packet reception ratio is the number of packets successfully received compared to the total number of packets sent from the source. With a variation in the density of the vehicles on the road, there is a high chance that the packet reception ratio will also vary with the variation in the link breakage rate. Fig. 6 shows the impact of the proposed scheme on the packet reception ratio with varying traffic density and link breakage rates. As the mobility of the nodes in VANETs is quite high, there is a high chance of link breakage in these networks, compared to other networks. With an increase in the link breakage rate, the packet reception ratio falls. The packet reception ratio is the number of packets successfully received at the destination compared to the total number of packets transmitted from the source. As the link breakage rate increases, it is quite difficult to aggregate the packets at the destination node due to high mobility. In this case, it is also more difficult to perform opportunistic forwarding of the packets to their final destination. When the link breakage rate is 5–25 links/ms with vehicle density is 100 vehicles/km, the packet reception ratio falls by 40%. Similarly, when vehicle density is 80 vehicles/km on the road, the packet reception ratio falls by about 19%. Similar behavior is observed for other cases with varying traffic density. 6.1.5. Impact of link breakage rate on delay As most of the time, information from source to destination is transferred by broadcasting; the link breakage rate is one factor which significantly affects different network conditions. Fig. 7 shows the impact of vehicle density on broadcast delay with varying link breakage rates. As shown in Fig. 7, with an increase in link breakage rates, there is an increase in broadcast delay. Broadcast delay is the total delay occurrence in the generation of a particular alert from the source to all the other nodes in the network. In particular, as the link breakage rate increases from 5 to 25 links/ms over traffic density of 100 vehicles/km on the road, the broadcast delay increases from 2 to 27 s. Similar behavior is observed for other cases in the proposed approach. Moreover, with an increase in traffic density, the broadcast delay decreases for the alerts generated from the source node. This is due to the use of the LA-based approach where even with an increase in traffic density, delay decreases due to the fact that the LA-based approach is self-adaptive and learns from the environment in which it is operating. Hence, it selects the best path from the available ones using the defined OAF metric, which reduces delay to large extent. 6.1.6. Impact of learning rate on probability of opportunistic forwarding The action performed by the automaton depends on how quickly it learns from the environment in which it is operating. Initially, the process of learning is slow but with the passage of time, the automaton adapts to the situation and performs its action according to the feedback provided by the environment. Fig. 8(a) and (b) shows the impact of the learning rate on opportunistic forwarding and packet reception ratio with an increase in traffic density. As shown in the figure, with an increase in vehicle density and the learning rate of the LA, the probability of opportunistic forwarding is increased. Moreover, the packet reception ratio is decreased. The reason for such behavior is due to the fact that with an increase in the learning rate of the LA, more opportunities are

provided for forwarding the packets as the learning of automaton is slow in the initial stage but with the passage of time, learning is increased and it performs better. Moreover, the probability of opportunistic forwarding is increased but there is only a marginal decrease in the packet reception ratio with an increase in density at the same learning rate. This is due to the fact that in realistic situations, a collision or some type of interference from neighboring vehicles may occur, causing a marginal decrease in the packet reception ratio. Similar behavior is observed in other scenarios.

676

7. Conclusions and future work

685

Vehicular Ad Hoc Networks (VANETs) have emerged as a key technology to be used in next generation wireless networks. They are used in many applications, especially for alert generation on the road in cases of emergency. But, due to the high mobility of the nodes in the networks, it is quite difficult to send packets to their final destination. To address this issue, in this paper, we propose Learning automata-based Opportunistic Data Aggregation and Forwarding (LAODAF) scheme. Data is aggregated by the LA which adaptively selects the destination for data transfer, based on the newly defined metric known as Opportunistic Aggregation and Forwarding (OAF). LA predicts the mobility of the vehicle and adaptively selects the path for forwarding, based on the value of OAF. Moreover, it updates its action probability vector and learning rate based on the values of OAF. This will reduce network congestion and the load on the network as it is aggregated and forwarded only when required. An algorithm for opportunistic data aggregation and forwarding is also proposed. The proposed strategy is evaluated using various metrics, a number of successful transmissions, connectivity, link breakage rate, traffic density, packet reception ratio and delay. As future work, we would like to apply LAs to the detection of intrusion in VANETs by using an event-based mechanism in which various events performed by nodes are detected and monitored statistically by using the Markov chain model, which can also be used for intrusion detection.

686

Acknowledgments

711

This work has been partially supported by the Instituto de Telecomunicações, Next Generation Networks and Applications Group (NetGNA), Portugal, and by National Funding from the FCT – Fundação para a Ciência e a Tecnologia through the PEst-OE/EEI/ LA0008/2011 Project.

712

References

717

[1] H. Fußler, M. Mauve, H. Hartenstein, M. Kasemann, D. Vollmer, Mobicom poster: location-based routing for vehicular ad-hoc networks, SIGMOBILE Mobile Computing and Communications Review, vol. 7, ACM, New York, NY, USA, 2003, pp. 47–49. [2] B. Selvaretnam, K. Wong, Handling the inter-vehicular communications challenge – a survey, in: The Ninth International Conference on Communication Systems, ICCS, 2004. pp. 86–90. [3] António Fonsecan, Teresa Vazao, Applicability of position-based routing for VANET in highways and urban environments, Journal of Network and Computer Applications (2012). [4] J.M. McQuillan, I. Richer, E.C. Rosen, The new routing algorithm for the arpanet, IEEE Transactions on Communications COM-28 (5) (1980) 711–719. [5] IEEE standard 802.1d-2004: IEEE Standard for Local and Metropolitan Area Networks: Media Access Control (MAC) Bridges, June 2004. [6] W. Zhao, M. Ammar, E. Zegura, Multicasting in delay tolerant networks: semantic models and routing algorithms, in: WDTN ’05: Proceedings of the 2005 ACM SIGCOMM Workshop on Delay Tolerant Networking, ACM, New York, NY, USA, 2005, pp. 268–275.

Please cite this article in press as: N. Kumar et al., Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks, Comput. Commun. (2013), http://dx.doi.org/10.1016/j.comcom.2013.09.005

677 678 679 680 681 682 683 684

687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710

713 714 715 716

Q5

718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735

COMCOM 4842

No. of Pages 11, Model 5G

23 September 2013 N. Kumar et al. / Computer Communications xxx (2013) xxx–xxx 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799

[7] Piyatumrong, P. Bouvry, F. Guinand, K. Lavangnananda, Trusted spanning trees for delay tolerant mobile ad hoc networks, in: IEEE Conference on Soft Computing in Industrial Applications, SMCia ’08, 2008, pp. 131–136. [8] Patricia Ruiz, Bernabé Dorronsoro, Pascal Bouvry, Lorenzo Tardón, Information dissemination in VANETs based upon tree topology, Ad Hoc Networks 10 (2012) 111–127. [9] M.K. Marina, S.R. Das, Ad hoc on-demand multipath distance vector routing, ACM SIGMOBILE Mobile Computing and Communications Review 6 (3) (2002) 969–988. [10] S.S. Manvi, M.S. Kakkasageri, Jaremy Pitt, Multiagent based information dissemination in vehicular ad hoc networks, Mobile Information System 5 (4) (2009) 363–389. [11] D. Li, H. Huang, X. Li, M. Li, F. Tang, A distance-based directional broadcast protocol for urban vehicular ad hoc network, in: International Conference on Wireless Communications, Networking and Mobile Computing, WiCom, 2007, pp. 1520–1523. [12] O. Tonguz, N. Wisitpongphan, F. Bai, P. Mudalige, V. Sadekar, Broadcasting in VANET, in: Mobile Networking for Vehicular, Environments, 2007, pp. 7–12. [13] O. Tonguz, N. Wisitpongphan, F. Bai, DV-CAST: a distributed vehicular broadcast protocol for vehicular ad hoc networks, IEEE Wireless Communications 17 (2) (2010) 47–57. [14] S. Bai, Z. Huang, D. Kwak, S. Lee, H. Oh, J. Jung, Vehicular multi-hop broadcasting protocol for safety message dissemination in VANETs, in: Proc. of IEEE 70th Vehicular Technology Conference Fall (VTC 2009-Fall), Anchorage, AK, 2009, pp. 1–5. [15] M. Slavik, I. Mahgoub, Stochastic broadcast for vanet, in: 7th IEEE Consumer Communications and Networking Conference (CCNC), Las Vegas, 2010, pp. 1– 5. [16] Y. Mylonas, M. Lestas, A. Pitsillides, Speed adaptive probabilistic flooding in cooperative emergency warning, in: WICON ’08: Proceedings of the 4th Annual International Conference on Wireless Internet, 2008, pp. 1–7. [17] C.-M. Huang, L. Tu, C.-H. Chou, Rewarn: an opportunistic relay scheme for cooperative collision warning in VANETs, in: IEEE 20th International Symposium on Personal, Indoor and Mobile Radio Communications, Tokyo, 2009, pp. 3030–3034. [18] Rowstron, G. Pau, Characteristics of a Vehicular Network, Technical Report 090017, University of California Los Angeles, Computer Science Department, 2009. [19] European-Community, Futureinternetinitiative http://www. futureinternet.eu/S, 2009. [20] BBN-Technologies, Exploringnetworksofthefuture – NSF award CNS-0714770 http://www.geni.net/S, 2009. [21] Giovanni Pau, Rita Tse, Challenges and opportunities in immersive vehicular sensing: Lessons from urban deployments, Signal Processing: Image Communications (2012). [22] S. Fruin, A. Winer, C. Rodes, Black carbon concentrations in California vehicles and estimation of in-vehicle diesel exhaust particulate matter exposures, Atmospheric Environment 38 (2004) 4123–4133. [23] D. Houston, P. Ong, J. Wu, A. Winer, Proximity of licensed child care facilities to near-roadway vehicle pollution, American Journal of Public Health 96 (2006) 1611. [24] E. Giordano, R. Frank, G. Pau, M. Gerla, Corner: a realistic urban propagation model for VANET, in: The Seventh IEEE International Conference on Wireless On-demand Network Systems and Services (WONS), IEEE Press, 2010, pp. 57– 60. [25] S.R. Schwartz, R.R.R. Barbosa, N. Mertina, G. Heijick, H. Scholten, A directional dissemination protocol for vehicular environment, Computer Communications 34 (17) (2011) 2057–2071. [26] D. Borsetti, M. Fiore, C. Casetti, C. Chiasserni, An application level framework for information dissemination and collection in vehicular networks, Performance Evaluation 68 (9) (2011) 876–896. [27] M. Li, Z. Kai, W. Lou, Opportunistic broadcast of event driven warning messages in vehicular ad hoc networks with lossy links, Computer Networks 55 (10) (2011) 2443–2464.

11

[28] M. T-Moreno, Inter-vehicle communications: assessing information dissemination under safety constraints, in: WONS, IEEE, 2007, pp. 59–64. [29] M. Torrent-Moreno, J. Mittag, P. Santi, H. Hartenstein, Vehicle-to-vehicle communication: fair transmit power control for safety-critical information, IEEE Transactions on Vehicular Technology 58 (7) (2009) 3684–3703. [30] M. Torrent-Moreno, F. Schmidt-Eisenlohr, H. Fussler, H. Hartenstein, Effects of a realistic channel model on packet forwarding in vehicular ad hoc networks, in: IEEE WCNC, 2006, pp. 385–391. [31] A Vinel, C. Campolo, J. Petit, Y. Koucheryavy, Trustworthy broadcasting in IEEE 802.11 p/WAVE vehicular networks: delay analysis, IEEE Communications Letters, 15 (9) 1010–1012. [32] C. Campolo, Y. Koucheryavy, A. Molinaro, A. Vinel, Characterizing broadcast packet losses in IEEE 802.11 p/wave vehicular networks, in: Personal Indoor and Mobile Radio Communications (PIMRC), 2011. [33] M. Esnaashari, M.R. Meybodi, A cellular learning automata based deployment strategy for mobile wireless sensor networks, Journal of Parallel and Distributed Computing 71 (7) (2011) 988–1001. [34] M. Esnaashari, M.R. Meybodi, A learning automata based scheduling solution to the dynamic point coverage problem in wireless sensor networks, Computer Networks 54 (14) (2010) 2410–2438. [35] J.A. Torkestani, M.R. Meybodi, Mobility-based multicast routing algorithm for wireless mobile Ad-hoc networks: a learning automata approach, Computer Communication 33 (6) (2010) 721–735. [36] J.A. Torkestani, M.R. Meybodi, Learning automata based algorithms for solving stochastic minimum spanning tree problem, Applied Soft Computing 11 (6) (2011) 4064–4077. [37] J.A. Torkestani, M.R. Meybodi, An intelligent backbone formation algorithm for wireless ad hoc networks based upon distributed learning automata, Computer Networks 54 (5) (2010) 826–843. [38] S. Chatzichristofis, A.M. Dimitris, A.S. Georgios, C. Boutalis, S. Yiannis, A novel cellular auomata based technique for visual multimedia content encryption, Optics Communications 283 (21) (2010) 4250–4260. [39] S. Misra, V. Tiwari, M. Obaidat, LACAS: learning automata-based congestion avoidance scheme for healthcare wireless sensor networks, IEEE Journal on Selected Area in Communication 27 (4) (2009) 466–479. [40] K.S. Narendra, M.A.L. Thathachar, On the behavior of a learning automaton in a changing environment with application to telephone traffic routing, IEEE Transactions on Systems, Man, and Cybernetics SMC-l0 (5) (1980) 262–269. [41] K. Najim, A.S. Poznyak, Multimodal searching technique based on learning automata with continuous input and changing number of actions, IEEE Transactions on Systems, Man, and Cybernetics-Part B: Cybernetics 26 (4) (1996) 666–673. [42] K.S. Narendra, M.A.L. Thathachar, Learning Automata-An Introduction, Prentice Hall, 1989. [43] M.S. Obaidat, G.I. Papadimitriou, A.S. Pomportsis, Learning automata: theory, paradigms and applications, IEEE Transactions on Systems, Man, Cybernetics, B 32 (6) (2002) 706–709. [44] M. Jerbi, S.M. Senouci, T. Rasheed, Y.G. Doudane, Towards efficient geographic routing in urban vehicular networks, IEEE Transaction on Vehicular Technology 58 (9) (2009) 5048–5059. [45] Ameneh Daeinabi, Akbar Ghaffar Pour Rahbar, Ahmad. Khademzadeh, VWCA: an efficient clustering algorithm in vehicular ad hoc networks, Journal of Network and Computer Applications 34 (1) (2011) 207–222. [46] S. Misra, P.V. Krishna, V. Saritha, LACAV: an energy efficient channel assignment mechanism for vehicular adhoc networks, Journal of Supercomputing 62 (2012) 1241–1262. [47] S. Misra, P.V. Krishna, A. Bhiwal, A.S. Chawla, B.E. Wolfinger, C. Lee, A learning automata based fault tolerant routing algorithm for mobile adhoc networks, Journal of Supercomputing 62 (2012) 4–23. [48] J. Härri, F. Filali, C. Bonnet, M. Fiore, VanetMobiSim: generating realistic mobility patterns Vanets, in: Proceedings of the 3rd International Workshop on Vehicular Ad hoc Networks, (VANET, ’06), Los Angles, California, USA, September 29 2006, pp. 96–97.

Please cite this article in press as: N. Kumar et al., Learning Automata-based Opportunistic Data Aggregation and Forwarding scheme for alert generation in Vehicular Ad Hoc Networks, Comput. Commun. (2013), http://dx.doi.org/10.1016/j.comcom.2013.09.005

Q6

800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863