Chpater 3 System Design and Algorithms
3.8 Our System design and Algorithms
In this section, we conclude the algorithm of our system by the flow chart in the Figure 3-7.
Figure 3-7 The flow chart of our system design and algorithms
0 25 50 75 100
0 200 400 600 800 1000
Speed(km/hr)
N
VD 05 2
If our system doesn’t receive the
c
i (the average speed of CFVD from cycles (i-4)-i), it will use the linear regression or power law to infer TMS fromn
i (the N in the segment at cycle i is estimated by the call arrivals and call completions in cycles (i-4)-i), else our system will execute the state determination algorithms to check the state of traffic condition at current cycle i. Then according to the state of traffic condition, the system will perform linear regression or power law training from the historical data of free flow or congested flow to correct the SMS to TMS.Chpater 4 Simulation Results and Performance
In this chapter, we describe how we perform the simulation experiment in the Section 4.1. Then we will show our simulation results and evaluate the accuracy of our speed estimation algorithm and state determination algorithm.
4.1 Simulation Environment
In this section, we design trace-driven experiments to investigate the traffic information estimations from cellular network data. As shown in Figure 4-1, this approach consists of the vehicle movement trace generation, MS communication trace generation, and the combined trace generation of the two behaviors described below.
Figure 4-1 The concept of our simulation
In this paper, the vehicle movement trace file is obtained from the traffic simulator (e.g., VISSIM) as well as real measurements of a highway in Taiwan. The inputs of trace generator include the road conditions (e.g., the length of the road, the number of lanes, handover locations, and traffic flow) and the vehicle movement behaviors (e.g., the desired speeds, the car following model and lane-changing model).
Moreover, we assume that an MS on each vehicle moving along the road. The MS communication behaviors (e.g., the call holding time and call inter-arrival time) are obtained from the random number generator (e.g., Microsoft Excel) for each MS with
a vehicle’s ID. Finally, the vehicle movement and MS communication trace files are combined with vehicle’s ID to output a trace file which records the vehicle’s ID, speed, locations, call arrival time, and call departure time. This trace file is then used to drive the mobility management simulator to estimate the real-time traffic information which includes speed, traffic density, and traffic flow. In Table 4-1, we show the simulation set up and parameters of our simulation enviornments.
Table 4-1 The simulation set up and parameters of our simulation Highway traffic condition with 3 lanes
Simulation tool VisSim
Length of road segment 10000 (m)
Distances of handover points (HOs) 2000, 1500, 750, 1000, 3000
Simulation time 7200 (s)
Distance between VDs 250 (m)
Cycle time of VDs 60 (s)
Traffic flow 5000 (car/hr)
Vehicle desired speed 85-120 (km/hr)
Accident location 8400 (m)
Accident time Simulation time 1200-2700 (s)
Accident range 100 (m) and impact 2 lanes
Vehicle desired speed in the accident 4-6 (km/hr)
Call inter-arrival time Poisson process with rate = 1 (call/hr) Call holding time Exponentially distributed with mean 1/
= 60 (sec)
We use the traffic simulator to generate the vehicle movement behaviors and the
random number generator to generate the communication behavior of MS accroding to our assumptions:
(1) The call arrival process is Poisson process with rate
(2) The call holding time is exponentially distributed with mean 1/
In simulation environment, we set the vehicle speed range from 85 km/hr to 120 km/hr, and the length of a 3-lane highway is 10 km. There are 6 handover points and 7 cells distributed on the road shown in Fig. 4-2, and the lengths of cells are 2000 m (Cell2), 1500 m (Cell3), , 750 m (Cell4) , 1000 m (Cell5) , and 3000 m (Cell6), respectively. The MS in the vehicle always performs and completes the handover procedure when the vehicle is driven though the handover point. We can record these handover events and use CFVD approach described in Session 3 to generate the speed report according to twice handover events from cellular network. For congestion situation, the accident occurs at simulated location 8150 m at simulated time 1500th second and it is eliminated at simulated time 3300th second. In each simulation run, up to 5,000 vehicles are injected in the road during 2 simulated hours, where the desired speed of a vehicle is uniformly randomly selected between 85-120 km/hr. For MS communication traces, the expected value (1/) of call inter-arrival time is 1 hr/call and the expected value (1/) of call holding time is 1 min/call.
Figure 4-2 The diagram of HOs and accident on the road segment
4.2 Simulation Results
The overall error ratio of our speed estimation algorithm is showed in Table 4-2 .We calculate the error ratio by the formula which compares the estimated TMS
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
HOs
1 2 3 4 5 6
Accident
with the VD speed obtained from our simulation:
i i
i y y
v
Ratio Error
Table 4-2 The overall error ratio of our speed estimation algorithm
Cell
2(2000m) Cell
3(1500m) Cell
4(750m) Cell
5(1000m) Cell
6(3000m)
Lv
8.23% 9.26% 14.10% 12.96% 8.06%Lv + Lc
5.64% 7.41% 10.59% 9.89% 7.46%Lv + Pc
5.28% 6.84% 11.02% 11.11% 7.48%Pv
8.96% 9.55% 15.74% 14.16% 7.77%Pv + Lc
5.48% 7.11%9.29% 8.99% 7.19%
Pv + Pc 5.12% 6.54%
9.72% 10.21% 7.22%Note that L and P represent the function of linear regression and the power law respectively, the under c and v indicate that L or P function is used to correct the SMS to TMS (c) or estimte the speed from N (v) . Hence , Lv means that we estimate the speed from N by linear regression without correction, and Lv+Pc means that we esimate the speed from N by linear regression and correct the SMS to TMS by power law, others can be infered by the same reason.
The results show that the correction from SMS to TMS will improve the accuracy of our system obviously, and the power law performs better than linear regression in estimation of speed from N but it is uncertain for correction from SMS to TMS. We also observe that the larger coverage range of cell and distance between cell and location of accident will increase the accuracy of the estimation.
Then we show the hit ratio of our state determination algorithm Table 4-3. The estimated state is compared to the real state for evaluating the hit ratio, and the real state is determined by the following algorithm:
jam
Table 4-3 The hit ratio of our state determination algorithm
Cell
2(2000m) Cell
3(1500m) Cell
4(750m) Cell
5(1000m) Cell
6(3000m)
Hit ratio 93.38%
90.44% 88.34% 86.03% 89.29%The results of hit ratio are similar to the speed estimation algorithm. For larger coverage range of cell and distance between cell and location of accident, the hit ratio will perform better.
Then we combine the state determination algorithm and present the separate evaluation of speed estimation algorithm in the free and jam state. The results are showed in Table 4-4.
Table 4-4 The error ratio of speed estimation in the free and jam state
Cell
2(2000m) Cell
3(1500m) Cell
4(750m) Cell
5(1000m) Cell
6(3000m)
First we can observe that the linear regression and power law perfoms better in free state and jam state respectively, and the evaluation of error ratios in free state are exactly good. Howerver, there are few speed reports of CFVD in the jam state, so we
should estimate the speed from N almost at every cycle in the quickly variation of traffic environment. Hence, the evaluation results of jam state are not bad and they are also good enough to depict the speed variation of traffic. We will show the best, worst and normal results of speed estimation and its corresponding information of our simulation in the following figures. We can easily find the two key factors that impact the accuracy of speed estimationa are the cell coverage range and distance between cell and accident.
(1) The best result of our algorithm:
Cell2 with coverage range of 2000 m
Distance between cell and accident: 6400 m
Figure 4-3 The information of the best result in our simulation
Figure 4-4 The best result of speed estimation in our simulation
0
900 1500 2100 2700 3300 3900 4500 5100 5700 6300 6900 7500 8100
Counts
900 1500 2100 2700 3300 3900 4500 5100 5700 6300 6900 7500 8100
State
(2) The worst result of our algorithm:
Cell4 with coverage range of 750 m
Distance between cell and accident: 4150 m
Figure 4-5 The information of the worst result in our simulation
Figure 4-6 The worst result of speed estimation in our simulation
(3) The normal result of our algorithm:
Cell6 with coverage range of 3000 m
Distance between cell and accident: 150 m
0
900 1500 2100 2700 3300 3900 4500 5100 5700 6300 6900 7500 8100
Counts
900 1500 2100 2700 3300 3900 4500 5100 5700 6300 6900 7500 8100
State
Figure 4-7 The information of the average result in our simulation
Figure 4-8 The average result of speed estimation in our simulation
0
900 1500 2100 2700 3300 3900 4500 5100 5700 6300 6900 7500 8100
Counts
900 1500 2100 2700 3300 3900 4500 5100 5700 6300 6900 7500 8100
State
Chpater 5 Conclusion and Future Work
In this thesis, we propose a novel traffic estimation algorithm using the CFVD.
Our algorithm consists of two parts:
State determination algorithm
Speed estimation algorithm using traffic density (N) and cellular network data The speed estimation algorithm is used to assist the CFVD system especially in the condition of traffic congestion, and the state determination algorithm is used to correct the SMS of CFVD reports to the TMS for accuracy evaluation with the speed of VD. The results indicate that the accuracies of our state determination algorithm and speed estimation algorithm can reach to 93.38% and 94.88%, respectively. We also show that our system is sensitive for the speed variations from the simulation results. Hence, our system can solve the problem that there are few effective speed reports of CFVD in the condition of traffic congestion.
However, our solution is based on the following assumptions:
Only moving vehicles on the road is considered
The call arrival process is Poisson process with rate
The call holding time is exponentially distributed with mean 1/
The variations of handover locations are not considered
For the usage of our system in the real traffic environment, we need to deal with these assumptions on the real cellular network data of telecommunication companies.
It will be more effective and practical to analyze the real data of cellular network to extend our system for real environment. Furthermore, we can also develop mechanisms to improve the effective CFVD reports even in the congested traffic condition in the future.
References
[1] Martin, P. T., Feng, Y., and Wang, X., 2003. Detector technology evaluation.
Department of CivilEnvironmental Engineering, University of Utah-Traffic Lab, Salt Lake City, UT.
[2] Leo, G.D., Pietrosanto, A., Sommella, P., 2009. Metrological performance of traffic detection systems. IEEE Transactions on Instrumentation and Measurement, Vol. 58, No. 9, pp. 3199-3206.
[3] Logghe, S., Maerivoet, S., 2007. Validation of travel times based on cellular floating vehicle data. Proceedings of the 6th European Congress and Exhibition on Intelligent Transportation Systems, Aalborg, Denmark.
[4] Bar-Gera, H., 2007, Evaluation of a cellular phone-based system for measurements of traffic speeds and travel times: A case study from Israel.
Transportation Research Part C, No. 15, pp. 380-391.
[5] Caceres, N., Wideberg, J.P., Benitez, F.G., 2008. Review of traffic data estimations extracted from cellular networks. IET Intelligent Transport Systems, Vol. 2, No. 3, pp. 179-192.
[6] Fontaine, M.D., Smith, B.L., 2005. Probe-based traffic monitoring systems with wireless location technology: an investigation of the relationship
between system design and effectiveness. Transportation Research Record:
Journal of the Transportation Research Board, No. 1925, pp. 3-11.
[7] Gundlegard, D., Karlsson, J.M., 2009, Handover location accuracy for travel time estimation in GSM and UMTS. IET Intelligent Transport Systems, Vol.
3, No. 1, pp. 87-94.
[8] Ygnace, J., Drane, C., Yim, Y.B. de Lacvivier, R., 2000. Travel time estimation on the San-Francisco bay area network using cellular phones as probes. University of California, Berkeley, PATH Working Paper UCB-ITS-PWP-2000-18.
[9] Thiessenhusen K.U., Schafer R.P., Lang T., 2003. Traffic data from cell phones: a comparison with loops and probe vehicle data. Institute of Transport Research German Aerospace Center, Germany.
[10] Danilo Valerio, “Road Traffic Information from Cellular Network Signaling”, Technical Report, FTW-TR-2009-003, 2009.
[11] Hesham Rakha and Wang Zhang, Estimating Traffic Stream Space-Mean Speed And Reliability From Dual And Single Loop Detectors, TRB Paper:
05-0850
[12] Soriguera, F. ,Robuste, F., Estimation of traffic stream space mean speed from time aggregations of double loop detector data, Center for Innovation in Transport, Technical University of Catalonia, Jordi Girona 29, 2-A, 08034
Barcelona, Spain
[13] Kerner, B. S., Three-phase Traffic Theory and Highway Capacity[J].
Physica A：Statistical Mechanics and Its Applications, 2004,333:379-400