40th Annual IEEE Conference on Local Computer Networks
LCN 2015, Clearwater Beach, Florida, USA
An Intelligent Vehicular Traffic Prediction (ITP) Protocol Maram Bani Younes
Azzedine Boukerche
Xiaoli Zhou
PARADISE Research Laboratory University of Ottawa, Canada Philadelphia University, Jordan Email:
[email protected]
PARADISE Research Laboratory DIVA Strategic Research Center University of Ottawa, Canada Email:
[email protected]
PARADISE Research Laboratory DIVA Strategic Research Center University of Ottawa, Canada Email:
[email protected]
Abstract—Several traffic evaluation and congestion detection protocols have been proposed in the literature for the vehicular networks. These protocols gather and analyze the basic traffic data of all traveling vehicles at each area of interest to obtain the vehicular traffic evaluations. All vehicles should be equipped by wireless transceiver in order to send and receive the packets of their traffic data individually. In general, the accuracy of typical traffic evaluation protocols depends mainly on the correctness and the quantity of the gathered data. In this paper, we propose an intelligent vehicular traffic prediction protocol (ITP) that estimates the traffic characteristics of each road segment in urban areas. ITP requires the basic traffic data of only some vehicles traveling on the investigated road segment. It then uses the gathered data to predict traffic characteristics of that road segment. The paper also investigates the minimum number vehicles that should participate broadcasting their data in order to obtain accurate traffic evaluation of the investigated road segment. From the experimental results, we can infer that the ITP protocol predicts an accurate level of the traffic evaluation when 25% of the traveling vehicles or more are equipped by wireless transceivers.
I. I NTRODUCTION Designing application protocols for vehicular networks requires evaluating the vehicular traffic characteristics of investigated areas of interest first. Several protocols have been proposed in the literature [1], [3], [6], [10], [9] aiming to evaluate the vehicular traffic characteristics of each area of interest. Previous protocols have required the basic traffic data of all vehicles on the area of interest. In general, considering the basic data of all traveling vehicles provides accurate evaluations. However, that introduces several challenges over the connecting network. These challenges include consuming the bandwidth and witnessing a high delay time to gather and process the basic data of all vehicles [6], [12]. To the best of our knowledge, none of the previous protocols have consider the existence of vehicles that are not equipped with wireless transceivers or vehicles that do not broadcast their basic data. This is a real limitation of previous protocols that prevent their practical usage. In this paper, we are first introducing an intelligent traffic evaluation protocol (ITP) that predicts the traffic characteristics of the investigated road segment. The ITP protocol does
978-1-4673-6773-8/15/$31.00 ©2015 IEEE
not require the basic data of all vehicles on the road segment. The minimum percentage of vehicles that should broadcast their basic data in order to predict correct and accurate vehicular traffic characteristics for each road segment are investigated deeply in this paper as well. The remaining of this paper is organized into four more sections: Section II overviews some previous researches in this field. Section III presents the details of our proposed protocol (i.e., ITP). Section IV illustrates the benefits of the ITP protocol, by comparing the performance of the ITP protocol to previous traffic evaluation protocols (i.e. ECODE [3]). Finally, Section V draws the conclusion of the paper. II. R ELATED W ORK The vehicular traffic distribution over urban areas is one of the most popular issues that has been studied recently using the vehicular network technology. Several protocols have been designed in the literature aimed at investigating the traffic characteristics of located road segments in urban areas [5], [3], [6], [10], [13]. In this section, we introduce a brief review of previous vehicular traffic evaluation protocols that have been proposed using the vehicular network technology. In the previous protocols, three general successive phases are designed to evaluate the traffic characteristics of each area of interest: (1) Vehicles broadcast their basic traffic data periodically or reactively. (2) Broadcasted data are gathered and analyzed regarding each area of interest. (3) The realtime traffic characteristics of each area are investigated, and then the highly congested areas are detected and predicted. All previous protocols have assumed that each vehicle over the investigated area of interest is equipped by a wireless transceiver. Thus, each vehicle transfer the basic data with its neighboring vehicles and infrastructures [2]. In the case of scalable area of interest or large number of traveling vehicles, the dissemination phase of the traffic evaluation protocols overload the network. This occurs by transmitting huge number of packets and lots of redundant data. The overloading network problem was handled in the previous protocols using several mechanisms and techniques [2], [10], [9], [13].
899
These solutions reduce the bandwidth consumption at the dissemination phase. In general, all of these previously introduced protocols have required the basic traffic data of all traveling vehicles over the area of interest. Thus, if a certain vehicle fails to broadcast its basic data properly, it will not be considered in the traffic evaluation of the targeted area of interest. To the best of our knowledge, none of the previous protocols have considered the existence of vehicles that are not equipped with wireless transceivers. Moreover, none of those protocols have considered vehicles that face difficulty to broadcast their basic traffic data. In real scenarios, many vehicles are not equipped by connection transceivers or they face failure problems that prevent the basic data broadcast messages. These scenarios produce misleading traffic characteristic reports which reflect inefficient control protocols. Due to the aforementioned problem, we aim to propose an intelligent traffic evaluation protocol (ITP). This protocol needs the basic traffic data of only some vehicles over the investigated road segment to estimate and predict the correct traffic characteristics. III. T HE P ROPOSED I NTELLIGENT V EHICULAR T RAFFIC P REDICTION (ITP) P ROTOCOL In this section, we introduce the details of our proposed protocol to predict the traffic characteristics of each investigated road segment in the urban area. The introduced protocol is named as Intelligent Traffic Prediction protocol (ITP). The remaining of this section illustrates the details of successive phases of the ITP protocol. A. Broadcasting Traffic Data In this phase, only vehicles equipped by a wireless transceiver (VW T i s), broadcast the basic traffic data including the vehicle’s identifier, position, direction, destination and its instant speed. The vehicle identifier is important to prove the identity of each vehicle, thus redundant messages can be easily detected. The position and direction data determine which road segment the vehicle is traveling on. The destination helps to predict the mobility of the vehicle and determine the next road segment. Finally, the speed information is essential to predict the traffic characteristics and traffic conditions of each road segment. This is due to the fact that the traffic speed is usually affected by the traffic density and the traffic conditions of each road segment. For instance, vehicles can move with the maximum allowed speed on low density road segments and/or on road segments in good conditions. However, high traffic density and bad road segment conditions force vehicles to decrease their speed and move slowly. B. Collecting Data Vehicles equipped by wireless transceivers (VW T i s) also can receive and collect the broadcasted messages of their
Algorithm 1: Broadcasting and Collecting Data 1
2
3 4
5 6 7 8
9
Each VW T i sends the basic traffic data periodic message; VW T i then receives the messages from it’s neighboring VW T s inside the configured zone Zi ; for k times do Vf forwards all the updated data gathered of each zone; if VW T i receives Fmes then Vi updates its records; end Each VW T i combine all received messages to keep only one record related to each vehicle and delete all redundant data; end
neighboring VW T s. In the case that the length of the investigated road segment is more than the transmission range of traveling vehicles, VW T s should retransmit the gathered data towards farther vehicles on the road segment. Dividing the road segment into a set of manageable zones enhance the process of collecting the traffic data of each road segment. The details of the road zoning mechanism is presented in Section III-B1. Furthermore, in order to forward the traffic data of each zone towards far zones, a few VW T s are selected. This should decrease the bandwidth consumption and to reduce the redundant packets over the network. The process of selecting the most suitable forwarding vehicle at each cluster is presented in details in Section III-B2. Each VW T i should collect the basic traffic data regarding to the entire road segment in order to predict the accurate traffic characteristics. This happens by accumulating the gathered data of all the configured zones of the road segment as explained in Section III-B3. Algorithm 1 illustrates a systematic presentation of the data dissemination and collecting phases. 1) Zoning The Road Segment : In order to facilitate the data collecting process, each road segment is divided into a set of manageable zones. Each configured zoned is referred by a certain identifier (Zid ). The length of each configured zone is set to be less than the transmission range of vehicles (i.e., Lz < VT R ). Thus, VW T s are able to receive any transmitted message inside the zone where it is located. Figure 1 illustrates an example of a certain investigated road segment and the configured zones there. As we can see from Figure 1 a small overlapped area is configured between each two adjacent zones. Inside each zone (Zi ), any VW T i can gather the basic traffic data of other vehicles if these latter vehicles broadcast their basic traffic data. 2) Selecting The Forwarding Vehicle of Each Zone: A certain vehicle is selected at each configured zone to efficiently forward the gathered data towards the neighboring zones. Since we investigate the traffic characteristics of road
900
Fig. 1.
An Example of a Zoned Road Segment and a Selected Forwarding Vehicle.
segments with two directions, the closest vehicle to the center of each configured cluster is the most recommended vehicle to forward (Vf ) the gathered traffic data. Decreasing the number of vehicles that forward the gathered data of each zone will decrease the bandwidth consumption especially in the scenarios that large number of vehicles are equipped by the wireless transceivers. Figure 1 illustrates an example of selecting the most suitable vehicle to forward the gathered data at a certain cluster. 3) Accumulating The Traffic Characteristics of Each Road Segment: The gathered traffic data at each configured zone are forwarded for several times towards neighboring zones using multi-hop transmission technique. This is what enables each VW T i to gather the basic data of all other VW T s on the road segment. Each VW T i keeps one record related to each VW T on the investigated road segment. Then, the traffic characteristics of the road segment are predicted using all of these gathered data. In the ITP protocol, the main factor to estimate the traffic characteristics of each road segment is the traveling speed of vehicles. C. Analyzing Gathered Data After gathering the entire traffic data of VW T s on a certain road segment. This data should be analyzed to estimate the traffic of each road segment and to predict the highly congested areas. First, the average traffic speed of VW T s is usually a good indicator for the average speed of all vehicles over the road segment. The fewer the number of VW T s, the higher the effect of the voluntarily slow vehicles on the accuracy of the computed traffic speed of the road segment. However, in
order to enhance the accuracy of the predicted traffic speed of each road segment, voluntarily slow vehicles are detected as negative outliers that should be trimmed before evaluating the traffic characteristics. On the other hand, fast vehicles, which are moving in a relative high speed compared to average traveling speed over the road segment, indicate an ability to increase the average traffic speed of the road segment. These vehicles are also detected as positive outliers that should be emphasized before evaluating the traffic characteristics. After that, from the evaluated average traffic speed, the traffic density and the estimated traveling time of each road segment can be predicted, as illustrated in Section III-D. These traffic characteristics are investigated for each direction of the road segment separately. 1) Detecting Voluntarily Slow Vehicles : The voluntarily slow vehicles are detected as outliers. First, the average speed (S) of all VW T s on each road segment direction is computed by Equation 1. P (Si ) S= (1) n
where Si is the speed of the vehicle Vi and n is the number of detected VW T s on the investigated road segment direction. Then, the standard deviation of the traffic speed (σ) is computed using Equation 2. The standard deviation represents the range of the speed distribution. s P (Si − S)2 (2) σ= n−1
After that, the negative outliers of the traffic speed are investigated. This is done by comparing the speed of each
901
the traffic characteristics of highly traveling vehicles (i.e., positive outliers) are emphasized. The traffic speed of each direction of the road segment is evaluated using Equation 5. P P SHi SN i SAcc = β × + where β > 1 (5) nH nN where the SHi is the speed of the highly traveling vehicle Hi; nH the number of highly traveling vehicles; SN i is the speed of the vehicle N i that is not detected as slowly or highly traveling vehicle; and, nN is the number of N vehicles on that road segment direction. The best value of β is set based on an empirical study of the investigated scenarios. Fig. 2.
Algorithm 2: Traffic Characteristics Prediction
An Example of Positive and Negative Outliers Vehicles’ Speed. 1 2
vehicle to the defined negative threshold (thdneg ) in Equation 3. thdneg = S − (α × σ) where α > 1 (3) where the best value of α is set based on the investigated scenario. From the comparison process, in the case that the value of (S − Si ) is more than thdneg the vehicle Vi is detected as voluntarily slow vehicle (i.e., negative outliers). Then, it should be trimmed while predicting the traffic characteristics of the road segment direction. In the case that most of VW T s are moving in slower speed than other vehicles on the road segment direction, the obtained information can be missleading. Thus, in this scenario wrong traffic characteristics are predicted. 2) Detecting Fast Vehicles : In order to enhance the accuracy of the predicted traffic characteristics we emphasis the effect of VW T s that are moving in a relative highly speed. If one or more vehicles move in a higher speed compared to the average traveling speed of vehicles, that means the traffic density does not force vehicles to move slowly. In order to detect these fact traveling vehicles (i.e., positive outliers) we define a positive threshold (thdpos ) that is computed in Equation 4. thdpos = S + (α × σ) where α > 1
(4)
In the case that (Si − S) is more than thdpos the vehicle Vi is detected as highly speed vehicle (i.e., positive outliers). Then, it should be highly considered while predicting the traffic characteristics of each road segment. Figure 2 illustrates an example of positive and negative outliers of the traffic speed of 20 traveling vehicles over a certain road segment. 3) Analyzing The Traffic Speed of Each Road Segment: Voluntarily slow vehicles that are detected as negative outliers on each road segment direction are trimmed to evaluate an accurate traffic speed of each road segment. However,
3 4 5 6 7
8 9 10
Each VW PT i computes the S and σ: S = q (Si )/n; P
2
(Si −S) σ= n−1 The thdpos and thdneg values are determined: thdneg = S − (α × σ) where α > 1; thdpos = S + (α × σ) where α > 1; The traffic characteristics SAcc , Di and ET are predicted for each P road segment: P SAcc = β × SHi /nH + SN i /nN ; Di = Rc / SAcc ; ET = Lrs / SAcc ;
D. Predicting The Traffic Characteristics Finally, the traffic characteristics of any road segment include the traffic density, traffic speed and the estimated traveling time of such a road segment are predicted. The traffic speed of each road segment is computed using Equation 5 as previously justified. In the ITP protocol, the traffic density and the estimated traveling time characteristics are predicted using the traffic speed of each road segment (SAcc ). The traffic density characteristic at each road segment direction is predicted following Greenshield model [11], in the ITP protocol. Based on the traffic speed of each road segment ITP predicts the traffic density on the road segment. The slower the traffic speed of each road segment the more the existed density. Equation 6 computes the traffic density (Di ) of the investigated road segment. Di = Rc / SAcc
(6)
where Rc is the capacity of the road segment that is previously determined. The estimated traveling time (ET ) of the road segment direction is predicted mainly based on the length of the road segment (Lrs ). The ITP protocol uses Equation 7 to compute the ET value of each investigated road segment. ET = Lrs / SAcc
902
(7)
Algorithm 2 illustrates a detail systematic presentation of data analyzing and traffic prediction phases. IV. P ERFORMANCE E VALUATION In this section we compare the performance of our proposed protocol (i.e., ITP) to ECODE, one of the most recent traffic evaluation protocol. ECODE has shown better performance to previous traffic evaluation and congestion detection protocol [3]. The investigated comparison in this section aims at illustrating the benefits and overheads of the ITP protocol. This is investigated in terms of bandwidth consumption, end-to-end delay and the accuracy of the obtained traffic characteristics of each road segment. These metrics are measured for several scenarios where different percentage of vehicles are equipped by wireless transceiver (W T ): 25%, 50%, 75%, 100%. As extensive set of experiments have been simulated using NS-2 [8] to evaluate the performance of these protocols. These experiments have used the following parameters illustrated in Table I.
Finally, the accuracy metric is measured by comparing the predicted traffic characteristics of each road segment to the real traffic. The accuracy of the traffic evaluation at each road segments depends only on the gathered traffic data in ECODE. Vehicles are considered on any road segment only in the case that they are equipped by wireless transceivers and can broadcast their basic data. However, in ITP vehicles could be predicted on a road segment from the gathered data and by deeply analyzing the gathered data. The ITP protocol predicts on average 70% more accurate traffic characteristics compared to ECODE as illustrated in Figure 3(c). In the scenarios that fewer vehicles are equipped by the wireless transceivers (i.e., 25%), the ITP protocol becomes relatively way more accurate that ECODE. Overall, the ITP protocol is an intelligent traffic evaluation protocol that predicts accurate traffic characteristics for each road segment on the urban area. It requires more time to predict accurate traffic characteristics of each road segment. However, no extra bandwidth consumption are required for ITP compared to ECODE [3]. V. C ONCLUSION
TABLE I S IMULATION PARAMETERS Parameter Transmission range (m) Vehicle’s Speed (m/s) Simulation time (s) Simulated area (m2 ) No. RSUs No. Vehicles Simulation map α β
Value 250 5 - 20 1000 25 X 200 - 25 X 1000 2 20 - 500 2 Sides Urban Road Segment 2 4
First, the bandwidth consumption is measured in these compared protocols by the total number of transmitted packets. Noticing that, the same message size is used in both protocols. The bandwidth consumptions of the ITP protocol is very close to the bandwidth consumption of the ECODE protocol as illustrated in Figure 3(a). This is because of the fact that only vehicles equipped by wireless transceivers transmit the basic traffic data in both protocols. In ECODE vehicles that are not equipped by a wireless transceiver cannot broadcast their basic data, however these vehicles are not considered or predicted on the road segment. The delay time of each protocol is measured at each vehicle starting from the time that vehicle broadcasts the basic traffic data until the traffic reports are generated on the located RSUs. Figure 3(b) illustrates the comparative results of ITP and ECODE protocols in terms of delay time. The ITP protocol requires on average 10% more time to process and evaluate the traffic characteristics of road segment compared to ECODE. This is due to the extra analyzing processes to predict the traffic characteristics based on the available data.
This paper proposed an intelligent traffic prediction protocol (ITP) that estimates the traffic characteristics of each road segment in the urban areas. In the ITP protocol partial traffic gathered data is used and analyzed to predict accurate traffic characteristics. Mainly the traffic speed of each road segment is used to predict accurate traffic characteristics in this protocol. From the experimental results, the ITP protocol requires more delay time to process the data and evaluate the characteristics. However, it predicts more accurate traffic evaluation even in the scenarios that low percentage of traveling vehicles on the road segment are equipped by wireless transceivers. The ITP protocol achieves very good performance in terms of predicting the traffic characteristics at each road segment in the urban area. It works efficiently accurate in the case that 25% or more vehicle on each road segment are equipped by wireless transceivers. ACKNOWLEDGMENT This work is partially supported by NSERC DIVA Strategic Research Network, Canada Research Chairs Program, and MRI/OIT Research funds. R EFERENCES [1] R. Bauza, J. Gozalvez, and J. Sanchez-Soriano, Road Traffic Congestion Detection through Cooperative Vehicle-to-Vehicle Communications, IEEE 35th Conference on Local Computer Networks (LCN), PP. 606-612, 2010. [2] T. Zhong, B . Xu, and O. Wolfson, Disseminating Real-Time Traffic Information in Vehicular Ad-Hoc Networks, in IEEE Intelligent Vehicles Symposium, 2008. [3] M. BaniYounes, A. Boukerche, A performance evaluation of an efficient traffic congestion detection protocol (ECODE) for intelligent transportation systems, Ad Hoc Networks, vol. 24, part. A, PP. 317-336, 2015. [4] B. Schunemann, J. Wedel, and I. Radusch, V2X-Based Traffic Congestion Recognition and Avoidance, Tamkang Journal of Science and Engineering, vol. 13, no. 1, PP. 63-709, 2010.
903
0.4
ITP ECODE
120 0.35
Delay (Sec)
Number of Sent Packets
ITP ECODE
100
80
0.3
0.25
60 0.2 40 25
50
75
Percent of Vehicles Equipped by WT
0.15
100
25
50
75
Percent of Vehicles Equipped by WT
(a)
100
(b) 1
ITP ECODE
0.9
Accuracy
0.8 0.7 0.6 0.5 0.4 0.3
25
50
75
Percent of Vehicles Equipped by WT
100
(c) Fig. 3. The performance evaluation of ITP compared to ECODE [3]: (a) Number of transmitted messages, (b) End-to-End communications delay and (c) The accuracy of the evaluated traffic congestion on each road segment.
[5] Y. Xu, Y. Wu, G. Wu, J. Xu, B. Liu, and L. Sun, Data Collection for the Detection of Urban Traffic Congestion by VANETs, IEEE Asia-Pacific Services Computing Conference (APSCC), PP. 405-410, 2010. [6] S. Dornbush, and A. Joshi, SmartStreet Traffic: Discovering and Disseminating Automobile Congestion Using VANET’s Vehicular Technology Conference, PP. 11-15, 2007. [7] L.D. Chou, D. Li, and H. Chao, Mitigation Traffic Congestion with Virtual Data Sink based Information Dissemination in Intelligent Transportation System, Third International Conference on Ubiquitous and Future Networks (ICUFN), PP. 37-42, 2011. [8] ”Network Simulator nss-2,” http://www.isi.edu/nsnam/ns/. [9] N. Shibata, T. Terauchi, T. Kitani, K. Yasumoto, M. Ito, T. Higashino , A Method for Sharing Traffic Jam Information using Inter-Vehicle Communication, Third Annual International Conference on Mobile and Ubiquitous Systems: Networking and Services, PP. 1-7, 2006. [10] J. Fukumoto, N. Sirokane, Y. Ishikawa, T. Wada, K. Ohtsuki, and H. Okada, Analytic method for real-time traffic problems by using Contents Oriented Communications in VANET, 7th International Conference on ITS Telecommunications(ITST ’07), PP. 1-6, 2007. [11] P. Kachroo, S.J. Al-nasur, S.A. Wadoo, A. Shende, Prdestrian Dynamics Feedback Control of Crowd Evacuation, Springer, 2008. [12] G. Korkmaz and E. Ekici, Urban Multi-Hop Broadcast Protocol for Inter-Vehicle Communication Systems, Proceedings of the 1st ACM international workshop on Vehicular ad hoc networks(VANET ’04), 2004. [13] S. Inoue, K. Shozaki, Y. Kakuda, An Automobile Control Method for
Alleviation of Traffic Congestions Using Inter-Vehicle Ad Hoc Communication in Lattice-Like Roads, Globecom Workshops, PP. 1-6, 2007
904
