Based on the existence of traffic light and the number of connected lanes, the common roundabouts in Changchun can be classified into four types:

1) Unsignalized four-leg roundabouts (e.g. Weixing Square and Jingyang Square)

2) Signalized four-leg roundabouts (e.g. Xi’an Square)

3) Unsignalized five-leg roundabouts (e.g. Zhengyang Square and Zhanqian Square)

4) Signalized five-leg roundabouts (e.g. Nanhu Square and Xinmin Square)

An example of each type of roundabouts is used as case studies in this paper

Table 1 shows the details related to the data collection plan. The evening peak periods are chosen as higher flows are present.

The roundabout at Weixing Square is one of the more important ones in Changchun. Figure 3 shows the configuration of the roundabout.

The evaluation index of unsignalized multi-lane roundabout is the V/C ratio of weaving section [1]. This is computed using the geometric data from digital map and peak hour flow using the Index Calculator, as illustrated in Figure 4.

The corresponding levels of service based on the computed V/C ratios [2] are shown in Table 2.

As shown in Table 2, weaving sections 2 and 4 are at level of service F, indicating long queues and delays, implying that actions must be taken to improve the performance.

Zhengyang Square is an unsignalized roundabout as shownin Figure 5.

Using the Index Calculator, the V/C ratios and service levels are computed and shown in Table 3.

The roundabout is functioning acceptably though weaving sections 2 and 5 are near failure conditions.

Xi’an Square is one of the most important intersections in Changchun. As this is a signalized roundabout, the average delay is used to assess its performance [1]. In this case, individual stop delay is measured. The level of service is determined based on the Highway Capacity Manual (HCM) [3].

The average delay and level of service at Xi’an Square are shown in Table 4. The roundabout has not reached capacity, although the traffic flow are rather high in Xi’an Square.

Xinmin Square is an example of a more complex signalized roundabouts in Changchun. The configuration of the roundabout is shown in Figure 6.

The average delay and service level of each leg is listed in Table 5.

The results show that during the afternoon peak, the traffic is high but the roundabout is operating just below capacity.

There has been past work done on simulating roundabouts [4] and Kotusevski, etc. [5] have compared several software in modeling roundabouts. Septarina [6] has made use of AIMSUN in modeling roundabouts in Sweden. In this paper, AIMSUN is used in developing the microsimulation model.

For the purpose of this paper, only the roundabout in Weixing Square is modeled to illustrate how changes to the roundabout can improve performance. Basic roundabout model has been established as is shown in Figure 7.

The traffic settings in the model are determined based on the traffic data collected. In order to define OD matrix and allocate the generation and attraction, centroid configuration is also necessary. Figure 8 demonstrates the four centroids of this roundabout.

Blue lines represent generation, while green lines represent attraction. In this picture, every single line corresponds with a specific percentage.

The OD matrices for car and bus traffic are shown in Figure 9 and Figure 10 respectively.

A screenshot taken during the simulation process is shown in Figure 11.

The last step was to carry out error checking, which includes three stages: software error checking, input data error checking and animation error checking.

After checking the software and input data, animation is to be reviewed. Compared with the actual condition, the queue length in the simulation is slightly longer, which probably results from the default “offside priority rule”. In addition, the traffic conditions in weaving section is slightly smoother than that in real-life scenario. This may be because the roundabout is affected by temporary construction within the central island resulting in a reduction in effective capacity in some weaving sections less than 5. Nonetheless, in general, the animation is very close to reality.

AIMSUN utilizes the Gipps model for updating the vehicle movements. This is shown in Equation (1) below.

where:

a_n = max acceleration of vehicle n;

b_n = max deceleration of vehicle n;

b_n-1 = max deceleration of vehicle n-1;

T = reaction time;

θ = reaction time to ensure safety;

v_n = max expected speed of vehicle n in time t;

v_n(t) = speed of vehicle n in time t;

x_n(t) = driven distance of vehicle n in time t;

S_n-1 = effective length of vehicle n-1.

Some of the parameters such as reaction time, speed limit, maximum acceleration, and maximum deceleration will need to be tested for sensitivity for the computation of performance criteria, e.g. queue length. From the earlier experimental runs, reaction time does not have a significant influence on average queue length. Moreover, given the “give way” protocol, the maximum give-way time should be taken into consideration.

The Root Mean Square Error method is used to obtain as the calibration criterion between the model measurements and the field measurements of V/C ratio. The calibrated parameters are shown in Table 6, along with the default values used in AIMSUN.

The roundabout can be improved by adding lanes in weaving sections. In this case, additional lanes are applied to the north and south weaving sections as shown in Figure 12.

The resulting changes in V/C ratios of each weaving section are shown in Table 7.

It is obvious that V/C ratios have dropped significantly in the north and south weaving sections, an improvement in service level from F to C.

Other performance measures have also improved more or less as is shown in Table 8.

In this scenario, besides adding lanes to the north and south weaving sections the

roundabout is subjected to fixed-time signal control as is demonstrated in Figure 13.

The resulting V/C ratios are shown in Table 9 and measures are shown in Table 10.

According to Table 9, the V/C ratio has dropped tremendously, while other indexes have deteriorated like Table 10 shows.

All in all, signal control, which is generally considered as an effective method, is actually not suitable for this intersection, whereas addition of lane is much more appropriate.