Traffic Control Strategy for a Surface Street on an Expressway-Arterial Corridor

TSINGHUA SCIENCE AND TECHNOLOGY ISSNll1007-0214ll15/16llpp776-781 Volume 14, Number 6, December 2009

Traffic Control Strategy for a Surface Street on an Expressway-Arterial Corridor* JIN Sheng (ࠡ ಙ), WANG Dianhai (ฆԯ‫**)ں‬, WANG Liming (ฆोੜ) Transportation College, Jilin University, Changchun 130022, China Abstract: Off-ramp area congestion and off-ramp queue spill over are wide-spread reasons for expressway traffic jams in China. A control strategy was developed to reduce the number of collisions between expressway and surface street vehicles and to reduce spill over by controlling the surface street vehicles to improve the traffic conditions near expressway off-ramps. This control algorithm monitors the traffic conditions at the off-ramp using occupancy rates and a performance index for the surface street vehicles, and then optimizes the control signal split and cycle length for the surface street vehicles assuming that the off-ramp queue is shorter than the minimum allowable length. Simulations indicate improvements in the traffic flow with the average vehicle travel time nearly 10% less and the off-ramp queue reduced significantly. Key words: off-ramp spill over; surface street metering; simulation

Introduction Urban expressways are widely used in China. However, urban economic development has created significant traffic congestion with very serious congestion on urban expressways. In Beijing, for example, the average speed on urban expressways is below 20 km/h during the rush hour. The traffic congestion on the expressways near the off-ramps is even more serious because of the bottlenecks formed by off-ramp queues. Due to historical reasons, off-ramps on expressways in Beijing are shorter than that on freeways in the United States and are always directly connected to the surface streets. The vehicles on the surface street and the off-ramp then interfere with each other around the off-ramp, which greatly reduces the capacity of the expressway. Received: 2008-10-15; revised: 2009-05-25

* Supported by the National Key Basic Research (973) Program of China (No. 2006CB705500) and the Specialized Research Fund for the

Doctoral

Program

of

Higher

Education,

China

20060183065)

** To whom correspondence should be addressed. E-mail: [email protected]; Tel: 86-431-85095857

(No.

The off-ramp queues then spill over onto the expressway and occupy the exit lane and even the mainline. Then non-exiting vehicles reduce their speeds upon seeing these queues which slows the flow in all lanes. Therefore, this is one of the most important reasons for expressway congestion in China. Intelligent transportation systems (ITS) can significantly reduce urban congestion. The management and control theories for urban expressways in China are based on freeway traffic research with most freeway control research focused on on-ramp metering which is a positive freeway control strategy. Ramp metering can use local ramp metering, such as in Chen et al.[1], Zhang and Ritchie[2], and Taylor et al.[3], and system ramp metering, such as in Papageorgiou[4], Yang and Yagar[5], Zhang and Rishnan[6], and Berg et al.[7] However, there is less research on off-ramp metering. Cassidy et al.[8] studied the traffic bottleneck characteristics near an off-ramp. Integrated control strategies have been proposed by Tian et al.[9] to adjust the signal operations at the surface street intersections and on-ramps to reduce the ramp queues with similar research by Neudorff et al.[10] and Urbanik et al.[11] However, there has been less research on off-ramp

JIN Sheng (ࠡ ಙ) et al.ġTraffic Control Strategy for a Surface Street …

traffic control. One off-ramp control mode is closure of the off-ramp which greatly increases the travel time of off-ramp vehicles. Therefore, this control method is not widely used. A new control method for surface street traffic near off-ramps called “surface street metering” was put into practice in Beijing in 2000. This control of the surface street traffic reduces congestion around off-ramp area and, thus, improves traffic flow on the expressway. There are more than 40 sites using this control method to ease traffic congestion in off-ramp areas. However, the control systems use a fixed timing so they lack the ability to adapt to temporary traffic variations. This paper presents an adaptive control strategy for surface street traffic with simulations to evaluate its effectiveness.

1

Control Strategy

A signal control method for surface street traffic was developed based on traffic conditions around typical off-ramps in Beijing. This control method can be called “surface street metering control (SSMC)”. The purpose of this method is to reduce the conflicts between off-ramp vehicles and surface street vehicles through control signals on the surface street upstream of the off-ramp. This control mode gives off-ramp vehicles priority but also considers the flow of surface street vehicles. The dynamic adaptive control algorithm is illustrated in Fig. 1.

777

step optimizes the timing parameters based on the off-ramp queue length while the second step adjusts for the queue on the surface street. The occupancy is used to describe the traffic conditions on the off-ramp. The control method has two thresholds for the off-ramp occupancy, a lower threshold oc, and a higher threshold os. If the detected off-ramp occupancy ooff is below the lower threshold oc, then the off-ramp is not congested and control is not needed. When ooff is higher than oc, the surface street metering rate, Rs, is calculated by comparing ooff with the desired saturated occupancy, os, at the off-ramp detector and then by adjusting Rs based on the off-ramp queue length and the surface street queue length. Finally, the cycle length is optimized to minimize disturbance of the surface street vehicles. The two thresholds in the control method are quite significant. Different occupancy thresholds will give different surface street metering strategies. For example, higher values of oc and os will result in more aggressive metering strategy where the high metering rate for the surface street will allow more surface street traffic flow even though the off-ramp occupancy is high. Similarly, a conservative metering strategy can be achieved by setting lower occupancy thresholds. The thresholds were currently selected to be 0.1 for oc and 0.35 for os. Future work will focus on optimizing these values.

2

Timing Parameter Optimization

The green signal time and the cycle length are two important control parameters in the control strategy. 2.1

Fig. 1 Control algorithm

The control method uses two control strategies for the two levels of traffic demand with light traffic loads having no control and heavy traffic loads having control optimized to reduce the off-ramp queue. The first

Green time

As with on-ramp metering, the objective of the surface street metering near an off-ramp is to determine the surface street metering rate. One popular local on-ramp metering method is ALINEA[12,13], which uses closedloop feedback control to periodically update the metering rate. The freeway occupancy in ALINEA is related here to the off-ramp occupancy while the onramp metering rate in ALINEA is related to the surface street metering rate. The control principle illustrated in Fig. 2 is simple and flexible. The controller adjusts the surface street metering rate based on feedback from the off-ramp occupancy.

Tsinghua Science and Technology, December 2009, 14(6): 776-781

778

Fig. 2

Feedback control to adjust the surface street traffic flow

The surface street metering rate is then based on the following algorithm: Rs (i  1) qs (i )  M (os  ooff (i)) (1)

where Rs(i+1) is the surface street metering rate to be applied during the next time period (veh/h), qs(i) is the actual surface street metering rate during the previous time period (veh/h), os is the desired saturated occupancy at the off-ramp detector, ooff(i) is the measured occupancy at the off-ramp detector during the previous time period, and ij is a parameter based on the off-ramp characteristics. After calculating the metering rate, the split on the surface street is calculated as g (i  1) Rs (i  1) (2) O (i  1) c(i  1) Ss where Ȝ(i +1) is the timing split for the surface street light during the next time period, g(i+1) is the green time for the surface street traffic during the next time period, c(i+1) is the cycle length for the surface street light during the next time period, and Ss is the saturated flow rate for all of the surface street lanes. Thus, the off-ramp occupancy is used to optimize the surface street metering rate. When the off-ramp occupancy is below the desired saturated occupancy at the off-ramp detector, the metering rate will increase. However, the metering rate will be reduced to keep the off-ramp occupancy from exceeding the desired offramp occupancy. The off-ramp vehicles move best with saturated occupancy through continuous feedback. 2.2

Cycle length

Most on-ramp metering methods allow one car per cycle. However, the surface street metering cannot use this mode because the flow characteristics of surface street vehicles are very different from that of on-ramp vehicles. Shorter cycle times will require frequent stops for the surface street vehicles so the traffic flow fluctuates more due to each vehicles’ repeated acceleration and deceleration. Thus, the cycle length optimization seeks to minimize the delay and the number of stops. Since the off-ramp vehicles have priority in

the split optimization, there should be no delays for the expressway vehicles and the performance index should only be calculated for the surface street vehicles. The performance index includes the delay and the number of stops as follows: PI Ds  KH s (3) where PI is the performance index for the surface street flow, Ds is the total delay for the vehicles, Hs is the total number of stops, and K is the stop weighting factor (default value 0.2) where K = 0 for the minimum delay, K = 0.2 for the minimum operating expense, and K = 0.4 for the minimum fuel consumption. PI provides a useful, generalized measure of the surface street performance and is widely used in signal design and analysis as a useful overall measure of traffic performance and mobility. The delay and the number of stops for both under-saturated and over-saturated traffic conditions are based on a time-dependent model which gives a transformation between a stochastic steady state delay and a deterministic over-saturation delay[14], 2 g· § 0.5qc ¨1  ¸ c¹ ©  Ds gº ª 1  « min(1, X ) » c¼ ¬ ª 8kIX º (4) 900qT «( X  1)  ( X  1)2  » QT ¼ ¬ g 1 c H s qf  gº ª 1  « min(1, X ) » c¼ ¬ 900T ª 8kIX º 2 (5) qf «( X  1)  ( X  1)  » c ¬ QT ¼ where q is the arrival flow rate (veh/s), g is the effective green time for a lane group (s), c is the cycle length (s), X is the lane group degree of saturation, T is the duration of the analysis period (h), k is the incremental delay factor that is dependent on the controller settings, I is the upstream metering adjustment factor, Q is the lane group capacity (veh/h), and f is the

779

JIN Sheng (ࠡ ಙ) et al.ġTraffic Control Strategy for a Surface Street …

adjustment factor for full stops. Figure 3 shows that changes of the cycle length result in changes of PI, Ds, and KHs. With increasing cycle length, the total delay increases while the total number of stops decreases. PI can then be minimized to match one threshold cycle length (also called the optimized cycle length). This will optimize the surface street vehicle flow.

(a)

Fig. 3 Optimization of the performance index based on the cycle length

The split Ȝ based on the off-ramp occupancy is independent of the cycle length c. The optimized cycle length can then be found by taking the derivative of PI with respect to c, noting that g/c is constant. d(PI) 0.5(1  O ) 2  dc 1  [min(1, X )O ] 900TfK ª 8kIX º 2 (6) «( X  1)  ( X  1)  » 2 c QT ¼ ¬ Setting d(PI) /dc 0 , the optimized cycle length is

then the positive real root: c 1800TfK [1  min(1, X )O ] ˜ ª 8kIX º 2 «( X  1)  ( X  1)  » (1  O ) QT ¼ ¬

(7)

The arrival flow rate and the split for surface street vehicles are two important variables when optimizing the cycle length in Eq. (7). Figure 4 shows how the optimized cycle length increases with the flow rate and decreases as the split increases.

(b) Fig. 4 Variation of the optimized cycle length with (a) the arrival flow rate and (b) the split

3.1

Test scenarios

Figure 5 shows the off-ramp area and detector locations used in this study. There were six detectors with off-ramp detector D1 located on the off-ramp to detect the off-ramp occupancy and detector D2 located at the beginning of the exit lane to detect off-ramp queue spillback. Detectors D3 and D4 were placed at the stop line of the surface street to detect the departure of surface street vehicles, while queue detectors D5 and D6 were placed 150 m before the stop line of the surface street to detect the queue and arrival rate. All levels of traffic demand in this area were considered in the simulations. The traffic demand was taken from expressway field data, divided into 12 intervals from

3 Simulations Simulations were conducted using VISSIM to compare the effectiveness of this optimized control method for various traffic conditions.

Fig. 5 Flow structure and detector locations (m)

Tsinghua Science and Technology, December 2009, 14(6): 776-781

780

light loads to heavy loads and back to light loads as shown in Fig. 6. The total simulation time was 6 h. The average vehicle travel time (AVTT) was used to measure the system effectiveness.

Fig. 6

Table 2 t-test results

Input traffic loads

The three control scenarios considered in the simulations were no control, fixed-time signal timing, and the current optimized control. The cycle length and green time were set as 60 s and 40 s in the fixed signal timing scenario. The timing parameters for the optimized control system were given by Eqs. (2) and (7) with an optimization interval of 120 s. The desired speed on the expressway was 80 km/h and on the surface street was 70 km/h. The amber time was 3 s, oc= 0.1, os= 0.35, ij = 1100, K = 20, and Ss = 1800 veh/(h Ž lane). 3.2

Simulation results

The simulated average vehicle travel times for the different control scenarios derived from simulation outputs with nine replications using different randomly generated seeds are summarized in Table 1. For each Table 1

Average vehicle travel time in each simulation

Random seed

simulation, the first 1800 s were used as the warm-up period with the data collection period from 1800 s to 21 600 s. The effectiveness measures were used to evaluate the control algorithm. The average vehicle travel time for these three control scenarios were compared based on a significance level of D = 0.05 (t0.05(16)=1.7459). The t-test results are shown in Table 2. The results show a significant statistical difference between the variable control and no control and between the variable control and fixed time control. Thus, the average time for the variable control 46.87f3.35 s was significantly lower than for the fixed time control 49.96f4.80 s, and the no control 51.93f9.46 s.

Control modes

t

Statistical significant at D = 0.05

No control vs. fixed time

1.5612

Variable vs. no control

4.2408

Ĝ

Variable vs. fixed time

3.2507

Ĝ

4

Discussion

The average vehicle travel time was 9.74% less with the variable control than with no control in the entire simulation. The data collection period was then divided into 11 sampling intervals with the average vehicle travel time for the three control modes in each period shown in Fig. 7. There was a less significant difference between the travel times for light traffic loads, but the reduction of average vehicle travel time by the variable control for heavy traffic loads was more significant because of the reduced off-ramp queue spill over. Also, the fixed signal timing did not improve the effectiveness for high traffic loads.

Travel time (s) No control Fixed Variable

12

57.59

52.82

47.08

22

49.14

48.45

46.53

32

52.75

52.96

49.39

42

54.77

50.89

48.42

52

51.46

50.43

44.57

62

52.62

46.47

47.08

72

51.53

50.85

44.91

82

47.03

48.55

44.82

92

50.46

48.24

49.00

Average

51.93

49.96

46.87

Fig. 7 Average vehicle travel time for various control methods

JIN Sheng (ࠡ ಙ) et al.ġTraffic Control Strategy for a Surface Street …

The average vehicle travel times for the expressway main flow (for short, MF), the surface street (for short, SS), and the off-ramp (for short, OR) are illustrated in Fig. 8. The improved effectiveness with the variable control is attributed to the reduced delays on the expressway and the off-ramp. Although the average vehicle travel time for the surface street was increased by about 10 s, on the whole, the conflicts between the expressway and surface street vehicles and the spill over of off-ramp vehicles onto the expressway were solved. Therefore, the control strategy significantly improves the expressway traffic flow.

781

[3] Taylor C, Meldrum D, Jacobson L. Fuzzy ramp metering: Design overview and simulation results. Transportation Research Record 1634, Transportation Research Board, National Research Council, Washington, D. C., USA, 1998: 10-18. [4] Papageorgiou M. Dynamic modeling, assignment, and route guidance in traffic networks. Transportation Research Part B, 1990, 24(6): 471-495. [5] Yang H, Yagar S. An algorithm for the inflow control problem on urban freeway networks with user-optimal flows. Transportation Research Part B, 1994, 28(2): 123-139. [6] Zhang H M, Rishnan J R. Coordinated traffic-responsive ramp control via nonlinear state feed-back. Transportation Research Part C, 2001, 9(5): 337-352. [7] Berg M, Bellemans T, Schutter B D, et al. Anticipative ramp metering control using dynamic traffic assignment. In: Conf. Rec. 2004 IEEE Int. Conf. Intelligent Transportation Systems. Washington, D. C., USA, 2004: 503-508. [8] Cassidy M J, Anani S B, Haigwood J M. Study of freeway traffic near an off-ramp. Transportation Research Part A, 2002, 36(6): 563-572. [9] Tian Z Z, Balke K, Rilett L. Integrated control strategies

Fig. 8 Average vehicle travel time for the different traffic flows

for surface, street and freeway systems. Transportation Research Record 1811. Transportation Research Board, National Research Council, Washington, D. C., USA, 2002:

5

Conclusions

92-99. [10] Neudorff L G, Randall J E, Reiss R, et al. Freeway man-

The paper presents a study of a variable timing control strategy of surface street traffic near expressway off-ramps with the control algorithm implemented in the VISSIM micro-simulation model. The effectiveness of the control strategies was evaluated based on performance measures for the simulation results. The results indicate that this control strategy gives improved traffic flow on both the expressway and the off-ramp. The simulations for different scenarios show that the control algorithm effectively reduces off-ramp congestion especially during peak traffic flows.

agement and operations handbook. US. DOT., FHWA-OP04-003, 2006. [11] Urbanik T, Humphreys D, Smith B, et al. Coordinated freeway and arterial operations handbook. US. DOT., FHWA-HRT-06-095, 2006. [12] Papageorgiou M, Salem H H, Blosseville J M. ALINEA: A local ramp feedback control law for on-ramp metering. Transportation Research Record 1320. Transportation Research Board, National Research Council, Washington, D. C., USA, 1991: 58-64. [13] Papageorgiou M, Middelham F. ALINEA local ramp me-

References

tering: Summary of field results. Transportation Research

[1] Chen C I, Cruz J B, Paquet J G. Entrance ramp control for

Research Council, Washington, D. C., USA, 1997: 90-98.

travel rate maximization in expressways. Transportation Research, 1974, 8: 503-508. [2] Zhang H M, Ritchie S G. Freeway ramp metering using artificial neural networks. Transportation Research Part C, 1997, 5(5): 273-286.

Record 1603. Transportation Research Board, National [14] Transportation Research Board. Highway Capacity Manual. National Research Council, Washington, D. C., USA, 2000.