This protocol describes the process of solving a microscopic traffic problem with simulation. The whole process contains a detailed description of data collection, data analysis, simulation model build, simulation calibration, and sensitive analysis. Modifications and troubleshooting of the method are also discussed.
Traditional U-turn designs can improve operational features obviously, while U-turn diversions and merge segments still cause traffic congestion, conflicts, and delays. An exclusive spur dike U-turn lane design (ESUL) is proposed here to solve the disadvantages of traditional U-turn designs. To evaluate the operation performance of ESUL, a traffic simulation protocol is needed. The whole simulation process includes five steps: data collection, data analysis, simulation model build, simulation calibration, and sensitive analysis. Data collection and simulation model build are two critical steps and are described later in greater detail. Three indexes (travel time, delay, and number of stops) are commonly used in the evaluation, and other parameters can be measured from the simulation according to experimental needs. The results show that the ESUL significantly diminishes the disadvantages of traditional U-turn designs. The simulation can be applied to solve microscopic traffic problems, such as in single or several adjacent intersections or short segments. This method is not suitable for larger scale road networks or evaluations without data collection.
Some traffic problems, such as traffic congestion at an intersection or short segment, can be solved or improved by optimizing the road design, change signal timing, traffic management measurements, and other transportation technologies1,2,3,4. These improvements either have a positive or negative effect on traffic flow operations compared to the original situations. The changes in traffic operations can be compared in traffic simulation software rather than in actual reconstruction of the intersection or segment. The traffic simulation method is a quick and cheap option when one or more improvement plans are proposed, especially when comparing different improvement plans or evaluating the effectiveness of improvements. This article introduces the process of solving a traffic problem with simulation by evaluating traffic flow operational features of an exclusive spur dike U-turn lane design5.
U-turn movement is a widespread traffic demand that requires a U-turn median opening on the road, but this has been debated. Designing a U-turn opening can cause traffic congestion, while closing the U-turn opening can cause detours for the U-turn vehicles. Two movements, U-turn vehicles and direct left-turn vehicles, require a U-turn opening and cause traffic delays, stops, or even accidents. Some technologies have been proposed to solve the disadvantages of U-turn movements, such as signalization6,7, exclusive left turn lanes8,9, and autonomous vehicles10,11. Improvement potential still exists on U-turn issues, due to the above solutions having restrictive applications. A new U-turn design may be a better solution under certain conditions and be able to address existing problems.
The most popular U-turn design is the median U-turn intersection (MUTI)12,13,14,15, as shown in Figure 1. A significant limitation of the MUTI is that it cannot distinguish U-turn vehicles from passing vehicles and that traffic conflict still exists16,17. A modified U-turn design called the exclusive spur dike U-turn lane (ESUL; Figure 2) is proposed here and aims to diminish traffic congestion by introducing an exclusive U-turn lane on both sides of a median. The ESUL can significantly reduce travel time, delays, and the number of stops due to its channelization of the two flows.
To prove that the ESUL is more efficient than the normal MUTI, a rigorous protocol is needed. The ESUL cannot be actually constructed before a theoretical model; thus, simulation is needed18. Using traffic flow parameters, some key models have been used in simulation research19, such as driving behavior models20,21, car following models22,23, U-turn models4, and lane change models21. The accuracy of traffic flow simulations is widely accepted16,24. In this study, both the MUTI and ESUL are simulated with collected data to compare improvements made by the ESUL. To guarantee accuracy, a sensitive analysis of the ESUL is also simulated, which can apply to many different traffic situations.
This protocol presents experimental procedures for solving real traffic problems. The methods for traffic data collection, data analysis, and analysis of overall efficiency of traffic improvements are proposed. The procedure can be summarized in five steps: 1) traffic data collection, 2) data analysis, 3) simulation model build, 4) calibration of simulation model, and 5) sensitivity analysis of operational performance. If any one of these requirements in the five steps is not met, the process is incomplete and insufficient to prove effectiveness.
1. Preparation of the equipment
2. Testing of the equipment
3. Data collection
4. Data analysis
5. Building the simulation model
6. Simulation model calibration
7. Sensitivity analysis
NOTE: Sensitivity analysis process is shown in Figure 8b. The collected data can only reflect its own performance (Figure 9, Table 4, Table 5, and Table 6). To prove the effectiveness under all situations, all possible traffic situations and different combinations were input into the simulation model to ensure that all situations are covered between the MUTI and ESUL (Figure 10 and Table 7).
Figure 2 shows the illustration of the ESUL for U-turn median opening. WENS mean four cardinal directions. The main road has six lanes with two directions. Greenbelts divide non-motorized lane on both sides and divide the two directions in the middle. Flow 1 is the east to west through traffic, flow 2 is east to east U-turn flow, flow 3 is west to east through traffic, and flow 4 is west to west U-turn traffic.
The functions of the inner 2 lanes of the ESUL are to divert, decelerate, U-turn, accelerate, look for headway, and merge the U-turn vehicles. The spur dike part is the core part and is different from ordinary U-turn designs. This part has the potential to force traffic flow to move outward slightly (one-lane width) and separate the through-traffic and U-turn traffic after the spur dike.
The spur dike design has three significant differences. First, it provides a specific U-turn lane to avoid influence from through traffic by moving the whole lanes outward. Compared to the markings, drivers cannot across the spur dike and must follow the lanes to divide the two flows apart36,37. Second, it maximally uses the land by symmetrically designing both two-direction U-turn demands. Third, the spur dike adjusts different U-turn radii of vehicles and uses the land flexibly.
Figure 3 shows the data collection location, which is a typical median opening at the northwest corner of the second loop road of Xi'an City in Shaanxi Province, China. The loop road in this research consists of six lanes, and the speed limitation in the loop road is 80 km/h (Figure 3a). The width of the lane is 3.5 m and median width is 1.2 m on average. The median opening section is 10 m wide and 17 m long. Two non-motor vehicle lanes (9 m width) are on both sides, and a 1.5 m greenbelt divides them from the main lanes (Figure 3b).
The distance between the upstream and downstream interchanges near the median opening is 5.1 km (Figure 3a). Since there is no entrance or exit for this section, the operation speed can reach the speed limit after the median opening reaches 200 m. From the median opening, it is 1.4 km to the upstream interchange and 3.6 km to the downstream interchange. Vehicles make a detour of 10 km (delay of 9 min at most) if no U-turn opening is designed. U-turn vehicles must wait for a long time when meeting at the intersection or are forced to join, resulting in delays or stopping of through-traffic. Figure 4 shows that the morning peak appears from 7:00 A.M. to 9:00 A.M., the evening peak appears from 17:00 to 19:00, and the valley (excluding late night) appears from 12:00 to 14:00.
The speeds of all traffic from east to west are shown in Figure 6a. The U-turn opening occurs at ~70 m at the horizontal axis. The deceleration and acceleration are obvious near 70 m, which indicates that the vehicles were affected by U-turn vehicles. The peak value in Figure 6a is under 80 km/h, and points are mainly centralized under 40 km/h, which indicate that the operating speed was much lower than the speed limit (80 km/h). Figure 6b shows the trajectories of traffic flow from east to west. The three lanes and U-turn vehicles trajectories are identified easily in the figure. The lowest trajectory is dark blue and wider than the two trajectories above it, indicating merging between the U-turn vehicles and through vehicles. The merging movement starts at 60 m and ends at 40 m, which represents a 20 m merge segment. The through-traffic in the inner lane was affected seriously by U-turn vehicles.
Figure 6c is the speed of flows from west to east. When the running speed reaches 80 m at the U-turn opening, it starts to increase. The result indicates that the WW U-turn flow had a smaller influence on WE via flow that was due to diversion movement (rather than merge movement; Figure 6b. The points starting from 0 km/h indicate that the WW U-turn vehicles caused stops and deceleration for whole vehicles. Figure 6d shows the trajectories from west to east of through-traffic and U-turn traffic. The U-turn section has high trees, which block the radar signals for detecting U-turn movements.
Figure 7 shows one-half of the ESUL design. Lanes 1 and 4 are through-traffic lanes, and lanes 2 and 3 are U-turn lanes. The calculation of each section is based on previously published guidelines35,38 and studies39,40. Section AB is based on a road alignment process, section BC is dependent on the drivers' reaction times and movement procedures, section CD is the diversion part, and section DE contains deceleration and safety distance. Section EF provides enough space to U-turn. Section FH and HI contain acceleration, headway finding, and combined motion separately. All sections are described in Table 2 according to a design speed of 80 km/h.
Figure 10a shows that the flow 1 travel time ratio decreased with ESUL under all traffic combinations within 20%-40%. The delay greatly decreased by 35%−70% (Figure 10b). The number of stops decreased slightly, with a maximum value of 0.4 (Figure 10c). The ESUL showed a significant improvement for EW through-traffic in all situations. Figure 9e,f and Figure 10d show the sensitivity results of flow 2 (EE U-turn vehicles). All three indexes of EE U-turn vehicles were improved greatly. The travel time shown in Figure 10d decreased by 20%-70% with increasing traffic volume. The delays in Figure 10e decreased more than the travel time and reached nearly 100% at the peak value. The minimum improvement ratio was larger than 70%. A significant improvement for number of stops shown in Figure 10f reached six at most.
Figure 9i,j and Figure 10h show the sensitivity results of flow 3 (WE through vehicles). With a similar trend to flow 1, flow 3 improved a lot with ESUL. Travel time decreased by 40%-50% in Figure 10h. Delays decreased by 50%-90% in Figure 10i. The number of stops only decreased 0.4x at most in Figure 10j. In flow 4, the WW U-turn vehicles and sensitivity results are shown in Figure 9l,m and Figure 10k. The travel time decreased by ~20%-60% with traffic volume increases (Figure 10k). In Figure 10l, delays increased 1% when traffic volume was 1,386 veh/h, and the U-turn ratio was 0.06. Delays decreased significantly by 54%-97% in the rest scope. The number of stops decreases up to 6x at most (Figure 10m).
Figure 1: Examples of median U-turn intersections (MUTIs). Two designs represent the common U-turn opening on the road, but it should be noted that the U-turn vehicles may cause traffic conflicts with passing vehicles, whether in the same or opposite direction flow. Please click here to view a larger version of this figure.
Figure 2: Illustration of the ESUL design on provincial trunk highway. W = west, E = east, N = north, S = south. Please click here to view a larger version of this figure.
Figure 3: Data collection location at a median at the northwest corner second loop road in Xi'an. Coordinates: 108.903898, 34.301482. (a) The investigation location schematic. (b) The MUTI of the U-turn median opening. The image was taken by a drone at the height of 150 m. Please click here to view a larger version of this figure.
Figure 4: 24 h congestion index. (a) The 24 h congestion trend of major cities from 2015 to 201725. (b) The 24 h congestion delay index for Xi'an on May 22nd, 201925,26. The data in panel a comes from the 2017 Traffic Analysis Reports for Major Cities in China25, which is provided by a Chinese web mapping navigation provider41. The data in panel b comes from the real-time congestion index in Xi'an on May 22nd, 201926. Please click here to view a larger version of this figure.
Figure 5: Data collection with radar on a pedestrian bridge at the U-turn location. Please click here to view a larger version of this figure.
Figure 6: Speed and trajectories of traffic flows. (a) Speed of vehicles from east to west. (b) Trajectories of vehicles from east to west. (c) Speed of vehicles from west to east. (d) Trajectories of vehicles from west to east. Please click here to view a larger version of this figure.
Figure 7: Geometry of ESUL design. The blue arrow represents vehicles traveling straight through, and the red arrow represents U-turn vehicles. Please click here to view a larger version of this figure.
Figure 8: Flowchart of calculating MAPE and sensitive analysis. (a) Calculation process of MAPE. (b) Process of sensitive analysis. Please click here to view a larger version of this figure.
Figure 9: Comparison between MUTI and ESUL with collected data. Comparison of travel time (a), delay (b) and number of stops (c) with morning peak (h). Comparison of travel time (d), delay (e) and number of stops (f) with middle noon valley (h). Comparison of travel time (h), delay (i) and number of stops (j) with evening peak (h). Please click here to view a larger version of this figure.
Figure 10: Sensitivity analysis of all flows, including EW through, EE U-turn, WE through and WW U-turn. X-axis = different traffic volumes, Y-axis = U-turn ratio, and Z-axis = improvement ratio (ratio = [MUTI – ESUL]/MUTI x 100%) in travel time and delay, reduced times (reduced times = MUTI – ESUL) in number of stops. (a-c) EW through flow, (d-f) EE U-turn flow, (h-j) WE through flow, and (k-m) WW U-turn flow. Every three figures are travel time (a,d,h,k), delay (b,e,i,l) and the number of stops (c,f,j,m), respectively. Please click here to view a larger version of this figure.
Item | Morning (07:00−08:00) | Middle Noon (13:00−14:00) | Evening (17:00−18:00) | |||||||||
Direction | EW | WE | EW | WE | EW | WE | ||||||
Flow | i = 1 | i = 2 | i = 3 | i = 4 | i = 1 | i = 2 | i = 3 | i = 4 | i = 1 | i = 2 | i = 3 | i = 4 |
Car | 4195 | 172 | 3442 | 504 | 3001 | 176 | 2460 | 402 | 1665 | 287 | 3296 | 394 |
Truck | 86 | 10 | 56 | 16 | 79 | 7 | 60 | 41 | 11 | 6 | 38 | 35 |
U-turn Ratio | 4281:182 | 3498:520 | 3080:183 | 2520:443 | 1676:293 | 3334:429 | ||||||
Aver. Speed | 21.5 | 11.5 | 22.2 | 10.5 | 36.7 | 12.3 | 23.7 | 11.8 | 29.3 | 12.8 | 22.9 | 12.1 |
Max. Speed | 73.8 | 13.4 | 63.7 | 12.8 | 90.4 | 15.6 | 75.9 | 13.5 | 76.7 | 14.6 | 63.7 | 13.3 |
Min. Speed | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 |
Table 1: Collected vehicle information. A minimum speed of 0 km/h indicates that some vehicles were stopped before starting to move.
Item | Description |
LAB | 166 m. Length for all flows to move outward slightly |
LBC | 185 m. Length for flow i = 4 to recognize the U-turn sign and take action |
LCD | 50 m. Diversion length to separate flow i = 3 and flow i = 4 |
LDE | 42 m. Flow i = 4 deceleration length |
LEF | Radius = 7.26 m. U-turn width for passenger cars |
LFH | 180 m. Acceleration length of flow i = 4 |
LHI | 140 m. The length to seek a headway for flow i = 4 and merge into flow i = 1 |
Table 2: Geometric parameters of ESUL. The calculation of each section is based on previously published guidelines35,38 and studies39,40. The value in Table 2 is input into the simulation model to evaluate ESUL performance at a design speed of 80 km/h.
Item | Morning (07:00−08:00) | Noon (13:00−14:00) | Evening (17:00−18:00) | |||||||||
Direction | EW | WE | EW | WE | EW | WE | ||||||
Flow | i = 1 | i = 2 | i = 3 | i = 4 | i = 1 | i = 2 | i = 3 | i = 4 | i = 1 | i = 2 | i = 3 | i = 4 |
Invest. Capacity (veh/h) | 4281 | 182 | 3498 | 520 | 3080 | 183 | 2520 | 443 | 1676 | 293 | 3334 | 429 |
Simulated Capacity (veh/h) | 4115 | 127 | 3571 | 501 | 3000 | 169 | 2484 | 360 | 1814 | 268 | 3381 | 409 |
Individual MAPE (%) | -3.9 | -30.2 | 2.1 | -3.6 | -2.6 | -7.4 | -1.4 | -18.7 | 8.2 | -8.5 | 1.4 | -4.6 |
MAPE (%) | -9.9 | -7.5 | -5.7 |
Table 3: Simulation calibration results. Calibration between the investigation and simulation is shown in the table. The MAPE is calculated using Equation 2, and the results are acceptable27,30.
Item | Travel time (s) | Delay (s) | Number of stops | ||||||
Flow | MUTI | ESUL | Rate (%) | MUTI | ESUL | Rate (%) | MUTI | ESUL | Rate (%) |
i = 1 | 17.7 | 12.5 | -29.4 | 7.2 | 4 | -44.4 | 0.19 | 0 | -100 |
i = 2 | 33.2 | 14.1 | -57.5 | 17.4 | 0.4 | -97.7 | 1 | 0 | -100 |
i = 3 | 18.3 | 9.9 | -45.9 | 5.6 | 1.6 | -71.4 | 0.06 | 0 | -100 |
i = 4 | 27.8 | 15.7 | -43.5 | 14.8 | 3 | -79.7 | 0.89 | 0 | -100 |
Table 4: Simulation results of MUTI and ESUL with morning peak data. In the morning peak, the ESUL improves significantly more than the MUTI. Travel time decreased by 29.4%-57.5%. Delay decreased by 44.4%-97.7%. The number of stops is completely diminished.
Item | Travel time (s) | Delay (s) | Number of stops | ||||||
Flow | MUTI | ESUL | Rate (%) | MUTI | ESUL | Rate (%) | MUTI | ESUL | Rate (%) |
i = 1 | 16.3 | 11.2 | -31.3 | 5.6 | 2.8 | -50 | 0.1 | 0 | -100 |
i = 2 | 26.7 | 15.6 | -41.6 | 10.1 | 1.3 | -87.1 | 0.5 | 0 | -100 |
i = 3 | 17.8 | 10 | -43.8 | 4.9 | 1.6 | -67.3 | 0.1 | 0 | -100 |
i = 4 | 24.4 | 15 | -38.5 | 11.7 | 2.9 | -75.2 | 0.7 | 0 | -100 |
Table 5: Simulation results of MUTI and ESUL with middle noon data. At noon, the travel time decreased by 31.3%-43.8%. Delay decreased by 50.0%-87.1% and no number of stops exist with ESUL.
Item | Travel time (s) | Delay (s) | Number of stops | ||||||
Flow | MUTI | ESUL | Rate (%) | MUTI | ESUL | Rate (%) | MUTI | ESUL | Rate (%) |
i = 1 | 13 | 9.4 | -27.7 | 2.9 | 1.1 | -62.1 | 0 | 0 | 0 |
i = 2 | 37.6 | 16.3 | -56.6 | 20.7 | 1.7 | -91.8 | 5.9 | 0 | -100 |
i = 3 | 18.3 | 10.6 | -42.1 | 5.6 | 2.2 | -60.7 | 0.2 | 0 | -100 |
i = 4 | 23 | 15.5 | -32.6 | 9.5 | 3.1 | -67.4 | 1.4 | 0 | -100 |
Table 6: Simulation results of MUTI and ESUL with evening peak data. With the evening peak data, travel time decreased by 27.7%-56.6%. Delay decreased by 60.7%-91.8%. The number of stops also diminishes with the ESUL.
Item | Value | ||||||||
Car/Truck(bus) ratio | 4281:182 (EW) / 3498:520 (WE) | ||||||||
U-turn ratio (%) | 0.03/0.06/0.09/0.12/0.15 | ||||||||
V/C | 0.2 | 0.3 | 0.4 | 0.5 | 0.6 | 0.7 | 0.8 | 0.9 | 1 |
Volume (veh/h) | 1386 | 2079 | 2772 | 3465 | 4158 | 4851 | 5544 | 6237 | 6930 |
Table 7: Parameters input into sensitivity analysis in simulation.
In this article, the procedure of solving a traffic problem at an intersection or short segment using simulation was discussed. Several points deserve special attention and are discussed in more detail here.
Field data collection is the first thing deserving attention. Some requirements for data collection location are as follows: 1) Finding a suitable location for data collection. The location should be similar to the road geometric shape in the study, which is the premise of data collection. 2) Determination of the set location of radar and other equipment by finding an enough clearance, where radar signals cannot be blocked. Some state-of-art technologies can be used, such as drones, to detect traffic operations. The entire observation area should be clear of barriers, like trees or architecture. 3) Finally, the data collection time should be at least 3 h in one location. The time should reflect the morning and evening peaks as well as the valley situation in the day. The time of congestion index can be obtained from observation or from another reliable transportation publisher.
Simulation model building is another critical step. The accuracy of the simulation model will lead to different simulation errors. The first thing in the simulation model is the connector. If one link on one side of the connector moves, the connector may be out of shape and intrude the adjacent link or connector, which may result in errors. So, it is important to recalculate the connector whenever moving a link is necessary.
Another key step is the conflict rule in conflict area. Use conflict areas instead of priority rules to simulate the right of way at intersections. Compared to priority rules, conflict areas are automatically displayed, and therefore are easier to edit and better reflect the driving behavior. The conflict rule should be the same as data collection, and every conflict area should be set with corresponding rules. The last critical step is adjustment of the parameters regarding driving behaviors when simulation error (MAPE) is big. Driving behaviors have several individual parameters, and a small change in each parameter may lead to a positive or negative impact on the results. It is key to adjust the various parameters carefully and repeatedly.
Usually, the travel time, delay and the number of stops is most common used indexes in evaluating the operational features in simulation. Many other indexes can also be obtained from the simulation (i.e., vehicle volume, exhaust emission, fuel consumption, pedestrian record, safety evaluation, vehicle behaviors, vehicle routes, coordinates, etc.). It is important to select the corresponding evaluation indexes according to the different experimental needs. Other indexes, except the three above that are most commonly used, may lead to new research findings or methods.
Using "quick mode" when performing the simulation may allow the simulation to reach the highest speed and save time, especially during the sensitive analysis. Thus, dozens of simulations are needed. The simulation result stays the same no matter which simulation speed is chosen.
There are two major areas for future applications. One application is the solving of traffic problems and evaluating one or more traffic designs at an intersection or short segment. The simulation helps to evaluate microscopic traffic behaviors, whether it includes vehicles, pedestrians, infrastructure modifications, or traffic management measurements. Second, the process provides a sufficient practice guide for those conducting traffic research. The provisions help to obtain accurate and robust data on traffic simulation measurements.
This method also has some limitations. First, the radar can detect a straight direction, and this requires that the target segment is also straight. The radar cannot be used for curved segments, like ramps. Second, the radar requires sufficient clearance to detect the vehicles. However, in the real environment, there are always trees or billboards that block the signal. It is difficult to find a suitable place for radar settlement. In addition, when the traffic volume is large or vehicles are close to each other, radar cannot distinguish the vehicles, and counting manually from the video is the only option, which is a lot of work. Efficiency and accuracy can be improved if the protocol also uses a method that can count and classify vehicles automatically.
The authors have nothing to disclose.
The authors would like to acknowledge the China Scholarship Council for partially funding this work was with the file No. 201506560015.
Battery | Beijing Aozeer Technology Company | LPB-568S | Capacity: 3.7v/50000mAh. Two ports, DC 1 out:19v/5A (max), for one laptop. DC 2 out:12v/3A (max), for one radar. |
Battery Cable | Beijing Aozeer Technology Company | No Catalog Number | Connect one battery with one laptop. |
Camera | SONY | a6000/as50r | The videos shot by the cameras were 1080p, which means the resolution is 1920*1080. |
Camera Tripod | WEI FENG | 3560/3130 | The camera tripod height is 1.4m. |
Laptop | Dell | C2H2L82 | Operate Windows 7 basic system. |
Matlab Software | MathWorks | R2016a | |
Radar | Beijing Aozeer Technology Company | SD/D CADX-0037 | |
Radar Software | Beijing Aozeer Technology Company | Datalogger | |
Radar Tripod | Beijing Aozeer Technology Company | No Catalog Number | Corresponding tripods which could connect with radars, the height is 2m at most. |
Reflective Vest | Customized | No Catalog Number | |
VISSIM Software | PTV AG group | PTV vissim 10.00-07 student version |