The Impact of Parallel U-Turns on Urban Intersection: Evidence from Chinese Cities

Abstract

As the number of motor vehicles in China’s cities continues to increase, the imbalance between the capacity that existing urban roads have for construction and the demand for motor vehicles is becoming increasingly evident. Indeed, the design of the intersection U-turn scheme has garnered significant attention from researchers. However, as the number of vehicles requiring U-turns increases, the traditional U-turn in the median or U-turn at the intersection fails to meet the timely demand for U-turns. In such cases, vehicles needing to make U-turns are required to queue first. As the queue length grows, it ultimately impacts the operational efficiency of the intersection. To optimize the imbalance between supply and demand at these intersections and promote the sustainable development of intersections, an innovative form of U-turn organization called the Parallel U-turn has been developed. In the engineering practice of reconstructing existing intersections or constructing new ones, it is crucial to investigate the compatibility between various U-turn design forms and traffic volumes. This exploration helps ensure that the chosen U-turn design effectively accommodates the specific traffic demands at the intersection. Therefore, in this paper, a typical intersection in Xi’an was chosen as the study intersection to investigate traffic data. The researchers calibrated and simulated four U-turn organization schemes using VISSIM microsimulation software. The four schemes included a traditional U-turn at the intersection, a Parallel U-turn at the intersection, a traditional U-turn in the median, and a Parallel U-turn in the median. Then, the researchers used the entropy-weighted TOPSIS method (EWTM) to evaluate the compatibility of each U-turn organization scheme for different traffic combinations. This assessment was conducted based on three criteria: operational efficiency, environmental protection, and safety performance. The results of this study indicate that the Parallel U-turn design is advantageous for the XiaoZhai intersection in Xi’an, China, under specific traffic conditions. When the traffic volume at the intersection exceeds 5940 vehicles per hour but remains below the intersection’s maximum capacity, implementing the Parallel U-turn design could yield positive outcomes in terms of operational efficiency, safety performance, and a reduction in intersection pollution. In summary, by enhancing operational efficiency, safety, and environmental impact, the Parallel U-turn design promotes the overall performance and sustainability of the XiaoZhai intersection and the transportation system in Xi’an, China.

1. Introduction

The U-turn is the maneuver of a vehicle that makes a 180° turn in a specific area to reverse its direction [1]. Vehicles that need to make U-turns often create merging points with traffic approaching from other directions [2]. This not only reduces the efficiency and stability of the intersection but also increases the probability of accidents, taking into account the lane-changing and merging behavior of vehicles [3]. For this reason, studies generally accept that the building of U-turn lanes at an intersection has the greatest impact on safety and efficiency [4]. Researchers have analyzed the median opening in terms of factors such as conflicting traffic volume [5], critical gaps [6,7], car-following time, and lateral attitude [8].

Scholars have discussed the key influences affecting the position of median openings, highlighting the growing demand for U-turns as a crucial factor [9]. At the same time, researchers have utilized the critical gap [10,11] and various configurations of lanes [12] as the parameter, respectively, to calculate the optimum location for the median opening. Some researchers have assessed the performance of median openings from the safety and crash perspective [13]. In addition, it is imperative to study the delays incurred by U-turn vehicles [14] and the innovation of the mode of the intersection [15] to evaluate their impact on traffic congestion.

Furthermore, U-turns in medians are also widely applied to freeway interchanges [16,17,18]. Scholars have conducted analyses from the perspective of safety sight distances to understand their impact on safety [19,20]. Vehicles with the requirement of a U-turn need to wait for the acceptable gap, which may cause stopping delays [6,7,21,22,23]. The delays resulting from lateral vehicle movements have been extensively studied by numerous researchers [24,25]. In terms of forms of intersection innovation, several researchers have proposed innovative U-turn forms, such as the Innovative U-turn Design (UALT) [26], addressing both the problem of changing lanes and the angle of vehicle U-turning.

In order to minimize intersection delays and queue lengths, an innovative intersection design called the Restricted Crossing U-turn (RCUT) [27,28,29,30,31,32] was developed, which redirects the left-turn and through traffic to U-turn intersections. Some researchers have assessed the effectiveness of the (ESUL) [33,34] intersection in terms of three indicators: travel time, delay, and number of stops. Some researchers constructed the elementary cellular automaton model to identify the impact of U-turning vehicles on traffic performance [35,36].

In a country like China, with a 65% urbanization rate, the rising number of vehicles has led to an imbalance between infrastructure and vehicle requirements [37]. Specifically, the design of U-turn schemes at intersections has garnered significant attention from researchers. However, considering the increasing number of vehicles with the requirement of U-turns, the traditional U-turn in the median or the U-turn at the intersection designs are unable to provide a timely solution to meet U-turn demand. Vehicles needing to perform U-turns must undergo a queuing process, which ultimately affects the operational efficiency of the intersection as the length of the queue increases [34]. Along with the accelerated urbanization process, the supply of existing roads in the country is unable to fulfill the demand of urban residents for vehicles, and the problem of imbalance between supply and demand is becoming increasingly prominent [38].

In order to enhance the efficiency of U-turn operations and further exploit the potential of road infrastructure and traffic management tools, some Chinese cities have innovatively proposed organizations. In this arrangement, vehicles stop in front of the stop line within the U-turn space and wait for the green light to proceed, enabling multiple U-turns to be completed simultaneously before the next light cycle begins (Figure 1). In the Figure 1, the pink curve and the blue curve represent the trajectories of two vehicles making U-turns simultaneously.

While parallel U-turns have been implemented in select cities in China, standardized guidelines for their implementation have not yet been developed by researchers. Therefore, it is necessary to analyze the feasibility of incorporating parallel U-turns in order to improve intersection operational efficiency, enhance safety performance, and reduce environmental pollution.

This study focuses on a typical intersection located in Xi’an, China, where the feasibility of implementing parallel U-turns is discussed in terms of enhancing intersection operational efficiency, safety performance, and environmental pollution reduction. The peak-hour traffic volume of the intersection was measured, and microsimulation software was utilized for analysis. It is worth noting that this intersection is equipped with a pedestrian overpass for the passage of pedestrians and non-motorized vehicles.

The remainder of the paper includes organized as follows: Section 2 includedes the problem statement and data collection; Section 3 includes the establishment and simulation of the model using VISSIM and SSAM; Section 4 includes the sensitivity analysis of operational performance; Section 5 includes the analysis of results based on the EWTM to supply an overall score to the models; Section 6 includes the conclusions to the paper.

2. Materials and Methods

2.1. Problem Statement

In the Chinese mountainous city of Chongqing and the Chang’an city of Xi’an, experts in the field of traffic have set up Parallel U-turns at intersections that are congested, as shown in Figure 1, Figure 2 and Figure 3. According to government departments, as of January 2023, the number of vehicles in Xi’an was 4.9 million, ranking eighth in China. This increase in the number of vehicles on urban roads and the limitations in the size of urban roads pose a challenge to the efficiency of intersections. Meanwhile, the additional percentage of vehicle U-turns increases the burden on the intersection.

Although Parallel U-turns are already in use and have demonstrated effectiveness ay improving the efficiency of intersections, the suitability of Parallel U-turns for traffic volumes has not been studied in depth. Improving the adaptability of intersection turnaround organization to traffic volumes is of great significance in terms of saving intersection land, improving intersection efficiency, protecting the urban environment, and ensuring traffic safety.

2.2. Data Collection

Located at the intersection of Chang’an Road and XiaoZhai Road, the XiaoZhai intersection is situated in the city’s bustling economic district, with an average daily traffic flow of 500,000 people and approximately 10,000 vehicles passing through the intersection every hour. This normalized traffic flow directly affects the travel experience of the public.

In particular, the traffic flow at the northbound entrance of the XiaoZhai intersection exhibits a significant imbalance between U-turns, left turns, and straight traffic. The periodic characteristics of U-turn vehicles are particularly pronounced, and during peak hours, the accumulation of vehicles can easily lead to congestion. In severe cases, long queues may prevent straight vehicles from entering this intersection, resulting in poor traffic efficiency along Chang’an Road. To address these issues, the researchers relocated the intersection stop line forward and implemented Parallel U-turns. The geometric configuration of this intersection and the location of data collection is shown in Figure 4. The lane function of the XiaoZhai intersection is divided as follows:

• There are three lanes in the eastbound entrance and westbound entrance.
• There are seven lanes at the southbound entrance.
• There are six lanes at the northbound entrance.
• There is a U-turn Lane at the northbound entrance, which is used to carry the U-turn demand of the traffic flow.
• The speed of the entire intersection is designed at 40 km/h.

According to the statistics of Auto Navi traffic in China Company, the traffic congestion index in Xi’an can be obtained, and it is easy to find that the morning peak hour in Xi’an is from 7:00 a.m. to 9:00 a.m., while the evening rush hour is from 5:00 p.m. to 7:00 p.m. The full-day congestion index for Xi’an on Monday, 22 May 2023, is shown in Figure 5.

Investigators stood on the pedestrian overpass at the intersection and used cameras to measure traffic volume at the intersection during the three peak hours on 22 May 2023. Based on these measurements, the evening peak flow was selected as the target flow, which is shown in

Table 1. Collected data during one peak hour (5:30 p.m. to 6:30 p.m.) on 22 May 2023.

Direction	Turn	Flow	Car	Bus	Total
N-S	S	1	1406	34	1440
	L	2	206	12	218
	R	3	242	4	246
	U	4	360	2	362
S-N	S	5	1560	80	1640
	L	6	102	16	248
	R	7	242	20	222
	U	8	18	0	18
E-W	S	9	390	26	416
	L	10	212	14	226
	R	11	186	14	200
W-E	S	11	412	40	452
	L	12	198	4	202
	R	13	188	20	208

.

The collected traffic data have the following characteristics:

• The roads of the intersection have primary and secondary characteristics, where North–South is the major road and East–West is the secondary road, and the traffic volume in the North–South is higher than the traffic volume in the East–West.
• The total traffic volumes in the northbound and southbound entrances are nearly identical, and they are nearly the same for the eastbound and westbound entrances.
• A larger proportion of vehicles U-turn in the northbound entrance, which accounts for about 20%.
• In Xi’an, new energy vehicles account for about 14.1%. Therefore, this paper focuses on conventional vehicles.
• This intersection is equipped with a pedestrian bridge, which allows the impact of pedestrians to be disregarded and non-motor vehicles on the U-turn, even at the intersection.
• The proportion of vehicles conducting a U-turn at the southbound entrance is so low that is explored in depth in this paper.

3. Establishment and Simulation of the Model by VISSIM

3.1. Preparation for Establishing the VISSIM Model

Microsimulation provides an efficient and accurate methodology to describe dynamic traffic behavior. VISSIM, the widely recognized software in the traffic field, can complete the simulation of U-turn lanes [39,40,41,42,43,44,45,46,47,48].

Combined with measured data, this paper formulates four schemes based on the location and form of U-turn organization, in combination with the overall layout of the XiaoZhai intersection. Figure 6 shows the layout of each scheme. Configuration 1 is the original U-turn design, where vehicles complete a U-turn before the stop line for the intersection The prerequisite for accurately replicating real traffic conditions using VISSIM simulation software requires the input of specific parameters. Based on the driving habits of Chinese drivers and in connection with available research [49,50,51,52], some of these parameters are specified as follows:

• The maximum deceleration (trailing vehicle) is set to −4 m/s².
• The accepted deceleration (own and trailing vehicle) is set to 2 m/s².
• The safety–distance reduction factor is set to 0.5.
• The maximum deceleration for the cooperative is set to −4 m/s².
• The rest of the parameters were used as defaults, and the Wiedemann 74 heel–chase model was selected.

The Synchro software has been used by many traffic engineers in order to determine signal cycles and to optimize and coordinate signal timing [53,54]. Synchro has demonstrated good performance and practicality in optimizing signal timing at intersections. As Synchro’s processing is iterative, it calculates the delays, queuing, and vehicle stops of the network while adjusting signal timing. It then assigns a score to each iteration based on these effectiveness metrics to achieve optimal network signal timing. This study used Synchro to complete intersection signal timing optimization.

3.2. Calibration of the VISSIM Model

In order to ensure the accuracy of the VISSIM model, we referred to previous research and selected the traffic volume as the calibration index. The validity of the model was evaluated by calculating the absolute percentage error (MAPE) between the measured flow and the simulated model flow. If MAPE < 15% [55], the model was valid. The formula for calculating MAPE is depicted in Equation (1):

$$MAPE=\frac{{\displaystyle \sum _{a=1}^n{C}_v^a}-{\displaystyle \sum _{a=1}^n{C}_f^a}}{{\displaystyle \sum _{a=1}^n{C}_f^a}}$$

(1)

where $i$ denotes the traffic flow, $n$ denotes the total of traffic flows, $C_v^a$ denotes the simulated capacity of VISSIM (veh/h), and $C_f^a$ denotes the collected capacity (veh/h). This paper uses Configuration 1 as the representative model. It can be seen that the total error between the VISSIM simulation model and the actual traffic volume was −0.21%, which shows the effectiveness of the VISSIM model in simulating traffic flow (

Table 2. The results of VISSIM simulation calibration.

Dir	Turn	Flow	Investigated Traffic Volume (veh/h)	Simulated Traffic Volume (veh/h)	Total MAPE (%)
N	S	1	1440	1361	−0.21%
	L	2	218	259
	R	3	246	216
	U	4	362	382
S	S	5	1640	1512
	L	6	248	274
	R	7	222	245
	U	8	18	22
E	S	9	416	446
	L	10	226	238
	R	11	200	216
W	S	12	452	475
	L	13	202	230
	R	14	208	209
Total			6098	6085

).

3.3. Constructing Comprehensive Evaluation Indicators

The VISSIM simulation model can output nine indicators on maximum queue length $x_1$, vehicle delay $x_2$, stopping delay $x_3$, the number of stops $x_4$, CO emissions $x_5$, NOx emissions $x_6$, VOC emissions $x_7$, fuel consumption $x_8$ and maximum travel time $x_9$. The specific explanation of the indicators is as follows:

Maximum queue length ($Mql$): The maximum value of all queue lengths within the node. In VISSIM, the maximum queue length of four reverse traffic streams was measured upstream from the intersection.

Delay: Delay refers to the gap between the actual travel time of the vehicle and the expected travel time of the driver. The reasons for this include traffic disruptions, traffic management, and traffic organization. Delays include vehicle delay ($Vd$) and stopping delay ($Sd$) in this paper.

Number of stops ($Ns$): In VISSIM, the number of stops is expressed as the number of all stops.

CO emissions ($CO$), NOx emissions ($NO$), and VOC emissions ($VOC$): Three indicators represent the sum of CO emissions, NOx emissions, and VOC emissions for all vehicles passing through each traffic flow, and the exhaust emissions affect the results of the nodal assessment. In VISSIM, these data were based on calculations by the US Department of Energy.

Fuel Consumption ($Fu$): The fuel consumption indicator represents the sum of fuel consumption for all vehicles passing on each traffic stream in VISSIM.

Travel Time (TT): Travel time represents the average time that it took for all vehicles to pass a given distance. In VISSIM, the average travel time of a vehicle crossing a road section to another road section, which included waiting time, was calculated by the distance between the starting section and the destination section.

Considering that there may be a strong correlation between the above indicators, this paper used factor analysis [56,57] to reduce the dimensionality of the relevant indicators.

3.3.1. Factor Model

Assuming that $X={\left(X_1,X_2,\dots X_p\right)}^T$ is a common factor random variable, the mathematical model of the factors is as follows:

$$X_i={\alpha }_{i1}{F}_1+{\alpha }_{i2}{F}_2+\dots +{\alpha }_{im}{F}_m+{\epsilon }_i i=1,2,\dots ,p$$

(2)

where $F_m$ denotes the public factor; ${\epsilon }_i$ denotes the special factor.

The factor model matrix takes the following form:

$$\mathit{X}=\mathit{A}\mathit{F}+\mathit{\epsilon }$$

(3)

$$\mathit{X}=\left[\begin{array}{c}{X}_1\\ ⋮\\ X_p\end{array}\right], \mathit{A}=\left[\begin{array}{ccc}{\alpha }_{11}& \cdots & {\alpha }_{1m}\\ ⋮& \ddots & ⋮\\ {\alpha }_{p1}& \cdots & {\alpha }_{pm}\end{array}\right], \mathit{F}=\left[\begin{array}{c}{F}_1\\ ⋮\\ F_p\end{array}\right], \mathit{\epsilon }=\left[\begin{array}{c}{\epsilon }_1\\ ⋮\\ {\epsilon }_p\end{array}\right]$$

(4)

3.3.2. Solving for Factor Loading Arrays

The factor loading matrix can usually be solved using various methods, such as principal component analysis, maximum likelihood, etc.

This paper applies the principal component method to obtain the factor-loading matrix and the solution process as follows:

Calculating the covariance matrix of the original variables and the covariance of the variables $X_i$ and $F_j$ can be conducted as follows:

$$\begin{array}{ll}Cov\left(X_i,F_j\right)& =Cov\left({\displaystyle \sum _{k=1}^m{\alpha }_{ik}{F}_k+{\epsilon }_i,F_j}\right)\\ & =Cov\left({\displaystyle \sum _{k=1}^m{\alpha }_{ik}{F}_k,F_j}\right)+Cov\left({\epsilon }_i,F_j\right)={\alpha }_{ij}\end{array}$$

(5)

Compute the characteristic roots of the covariance matrix $\sum$ as ${\lambda }_1\ge \cdots \ge {\lambda }_p\ge 0$ and the corresponding unit eigenvectors as ${\mathit{T}}_1,{\mathit{T}}_2,\dots ,{\mathit{T}}_p$;

The corresponding $T_i$ and ${\lambda }_i$ of the covariance matrix $\sum$ were used to solve the matrix $\mathit{A}$ as follows:

$$\mathit{A}=\left(\sqrt{{\lambda }_1}{T}_1,\sqrt{{\lambda }_2}{T}_2,\dots ,\sqrt{{\lambda }_p}{T}_p\right)$$

(6)

3.3.3. Factor Rotation

The factor loading matrix obtained by the principal component method is not unique, and the variables might not be well differentiated in terms of common factor loadings, which would be detrimental to the interpretation of common factor meanings. In this case, the load matrix needed to be factored, which could be conducted using orthogonal or oblique rotation, etc., so that the values of the load matrix were bifurcated to 0 and 1 to facilitate the naming of common factors.

The factor model matrix is of the form $\mathit{X}=\mathit{A}\mathit{F}+\mathit{\epsilon }$. If the effect of factor 3 on this model is ignored, the score of sample $\mathit{X}$ on the corresponding common factor $\mathit{F}$ can be calculated when $m=p$ and $A$ is invertible. However, $m$ is usually smaller than $p$ and does not give an exact value of $\mathit{X}$.Then, parameter estimation methods can be used to obtain factor score estimates. Next, Thompson regression is used to estimate the factor scores and the expression, which is as follows:

$$\widehat{\mathit{F}}={\mathit{B}}^T{\sum }^{-1}\mathit{X}$$

(7)

where $\mathit{W}$ is the factor score coefficient matrix: $\mathit{W}={\mathit{B}}^T{\sum }^{-1}$.

Finally, the factor expression was derived based on the factor score coefficients and the standardized values of the original variables as follows:

$$\left\{\begin{array}{ll}{F}_1=& 0.151x_1-0.025x_2-0.007x_3-0.031x_4+0.232x_5\\ & +0.232x_6+0.232x_7+0.232x_8-0.031x_9\\ F_2& =-0.072x_1+0.258x_2+0.254x_3+0.257x_4-0.004x_5\\ & -0.004x_6-0.004x_7-0.004x_8+0.249x_9\end{array}\right.$$

(8)

where $F_1$ denotes the efficiency index; $F_2$ denotes the environmental index.

3.4. Evaluation of Simulation Results

The actual measured data from this intersection were input into the four simulation configurations above and the simulation results are shown in

Table 3. Simulation results for four schemes.

	Configuration 1	Configuration 2	Configuration 3	Configuration 4
maximum queue length(m)	81.05	83.81	90.93	92.38
vehicle delay(s)	21.96	21.64	17.68	18.61
stopping delay(s)	15.59	15.34	11.67	12.32
number of stops	0.75	0.74	0.7	0.74
CO emissions(g)	1081.31	1073.58	1030.7	1061.24
NOx emissions(g)	210.384	208.879	200.537	206.478
VOC emissions(g)	250.604	248.811	238.875	245.952
fuel consumption(gallon)	15.469	15.359	14.745	15.182
traveling time(s)	43.76	42.94	37.21	38.03

.

Combining the results of the factor analysis in Section 3.3, four indicators were obtained including the values of the efficiency indicator $F_1$ and environmental indicator $F_2$ which are shown in

Table 4. Combined evaluation results of the four configurationss.

	Configuration 1	Configuration 2	Configuration 3	Configuration 4
$F_1$	371.60	369.47	356.52	366.89
$F_2$	8.65	8.14	4.48	4.82

.

As shown in Table 4, regarding the operational efficiency of the intersection, Configuration 1 outperformed Configuration 2 in terms of the U-turn at the intersection, indicating that the traditional U-turn design is superior to the Parallel U-turn design. For the U-turn in the median, Configuration 4 outperformed Configuration 3, meaning that the Parallel U-turn design is better than the traditional U-turn design. In terms of reducing pollutant emissions, there was not much difference in the performance between Configuration 1 and Configuration 2 for U-turns at the intersection. However, Configuration 1 scores were slightly higher, possibly due to their ability to improve intersection operational efficiency to a greater extent. Similarly, there was not much difference in performance between Configuration 3 and Configuration 4 for U-turns in the median, with Configuration 4 scoring slightly higher for the same reasons as U-turns at the intersection.

3.5. Evaluation of Security Indicators

In addition to the aforementioned access efficiency and environmental indicators, the safety of level crossings is also crucial. The Surrogate Safety Assessment Model (SSAM) is commonly used by the Federal Highway Administration (FHWA) to assess road safety when evaluating the safety of these crossings. In this paper, the VISSIM microsimulation output trajectory files, which, in combination with SSAM [58,59], analyze the trajectory files, were used to evaluate the safety of the U-turn model. Commonly used safety indicators include time to rear intrusion (PET), time to conflict (TTC), crossing, rear end, and lane change (

Table 5. Safety analysis of four SSAM U-turn design configurations.

Configuration Number	Crossing	Rear End	Lane Change	Total
Configuration 1	92	65	21	178
Configuration 2	90	62	20	172
Configuration 3	62	56	23	141
Configuration 4	112	90	34	236

).

In terms of the intersection analysis, Configuration 1 demonstrated superior performance compared to Configuration 2 for U-turns at the intersection. This indicates that Parallel U-turns at the intersection outperform traditional U-turns, with an optimization degree of 2%. Similarly, for U-turns in the median, Configuration 3 outperformed Configuration 4, revealing that traditional U-turns in the median outperform Parallel U-turns with an optimization degree of 45%.

Regarding vehicle rear-end collisions, Configuration 1 outperformed Configuratione 2 for U-turns at this intersection, indicating that Parallel U-turns are more effective than traditional U-turns at the intersection, with an optimization degree of 4.8%. Additionally, for U-turns in the median, Configuration 3 outperformed Configuration 4, indicating that traditional U-turns are more effective than Parallel U-turns in the median, with an optimization degree of 37.7%.

In terms of the number of vehicle lane changes, Configuration 1 outperformed Configuration 2 for U-turns at the intersection, suggesting that Parallel U-turns are superior to traditional U-turns at the intersection, with an optimization degree of 5.8%. Similarly, for U-turns in the median, Configuration 3 outperformed Configuration 4, indicating that traditional U-turns are more effective than Parallel U-turns in the median, with an optimization degree of 40%.

Overall, the implementation of Parallel U-turns positively influenced the intersection index, vehicle tailgating, and the number of vehicle lane changes for U-turns at the intersection. However, for U-turns in the median, the adoption of Parallel U-turns did not result in substantial improvements in the intersection index, vehicle tailgating, and the number of vehicle lane changes.

Based on the previously mentioned simulation results, it is clear that both traditional U-turn designs and the Parallel U-turn design have their strengths and weaknesses in terms of intersection efficiency, environmental impact, and safety. To assess the optimization degree of the Parallel U-turn design configuration compared to different forms of traditional U-turn designs, specific traffic data from the XiaoZhai intersection in Xi’an City, China, were utilized.

4. Establishment and Simulation of the Model by VISSIM

4.1. Sensitivity Factor Determination and Different Traffic Scenarios Establishment

The efficiency of a signalized intersection is highly correlated with the saturation of the intersection. Taking the saturation characteristics of the intersection at different times of the day and the traffic volume in all directions of the intersection as independent variables, the performance of the four configurations can be performed in relation to three aspects: efficiency, environment, and safety.

According to the investigation on site and the description of AASHTO, the maximum service flow at the junction was 4042 pcu/h in the north–south entrance and 1581 pcu/h in the east–west entrance. Therefore, the intersection saturation range was defined as 0.3 v/c to 0.8 v/c in order to ensure a normal traffic flow at the XiaoZhai intersection and according to the relevant level of service regulations in the sensitivity analysis. The traffic in sensitivity analysis combinations is shown in

Table 6. VISSIM volume in sensitivity analysis. (Different traffic volumes in the four directions were combined to perform VISSIM simulation tests under different traffic combinations, and the whole sensitivity analysis included 36 groups of different traffic combinations.).

Item	Value
N/S-Volume	1213/1617/2021/2425/2829/3234
E/W-Volume	474/632/791/949/1107/1265

.

According to the investigation of site data, the traffic volumes were almost the same in the north–south direction and almost the same in the east–west direction. Therefore, it was assumed that traffic volumes in the north–south, and east–west directions were the same. The investigation of the peak hour traffic volumes at this intersection identified the proportion of turning vehicles, and therefore, the proportion of turning vehicles on the north entrance road remained unchanged in the sensitivity analysis.

At the same time, the signal timing software Sychro7 was used to optimize the operation of the intersection for different combinations of traffic volumes.

4.2. Sensitivity Analysis

Combining the comprehensive factors obtained from the above factor-based analysis, including the efficiency indicator F1, environmental indicator F2, and safety indicator F3, these were used as indicators for the sensitivity analysis.

The degree of optimization of the four configurations is represented in Figure 7. The optimization rate was calculated by Ratio = (Configuration 1-Configuration 2)/Configuration 1 × 100%, and if the results were positive, Configuration 2 was better than Configuration 1.

As shown in Figure 7a, in terms of intersection operation efficiency, when the north–south entrance traffic flow gradually increased to 3234 pcu/h, the performance of Configuration 2 was always better than Configuration 1, and the optimization degree interval of Configuration 2 was 20~70%, showing an increasing trend, indicating that the optimization degree of Configuration 2 is more significant when the north–south entrance traffic flow increases. When the east–west entrance traffic gradually increased to 1265 pcu/h, the performance of Configuration 2 was always better than Configuration 1, and the optimization degree interval of Configuration 2 was 50~70%, showing a slight upward trend, which indicates that the optimization degree of Configuration 2 is more significant when the east–west entrance traffic increases. Taken together, when the north–south entrance traffic flow was 3234 pcu/h, and the east–west entrance traffic flow was 1265 pcu/h, the optimization degree of Configuration 2 was the largest, and the maximum value was 70%, which indicates that Parallel U-turns are feasible strategies to improve intersection operation efficiency.

As shown in Figure 7b, in terms of protecting the environment, when the traffic flow at the north and south entrances gradually increased to 3234 pcu/h, the performance of Configuration 2 was always better than that of Configuration 1, and the optimization degree interval of Configuration 2 was 11~5%, which presents a decreasing trend, indicating that when the traffic flow at the north and south entrances decreases the optimization degree of Configuration 2 is more significant. When the east–west entrance traffic gradually increased to 1265 pcu/h, the performance of Configuration 2 was always better than Configuration 1, and the optimization degree interval of Configuration 2 was 6~5%, showing a slight downward trend, which indicates that the optimization degree of Configuration 2 is more significant when the east–west entrance traffic decreases. Taken together, when the north–south entrance traffic flow was 1213 pcu/h and the east–west entrance traffic flow was 474 pcu/h, the optimization degree of Configuration 2 was the largest, with a maximum value of 11%, which indicates that Parallel U-turns are feasible strategies to improve the environmental protection of intersections.

As shown in Figure 7c, in terms of safety performance, when the traffic flow at the north and south entrances gradually increased to 3234 pcu/h, the performance of Configuration 2 was always better than that of Configuration 1, and the optimization degree of Configuration 2 was in the interval of 20~50%, showing an upward trend, which indicates that the optimization degree of Configuration 2 was more significant when the traffic flow at the north and south entrances increased. When the east–west entrance traffic gradually increased to 1265 pcu/h, the performance of Configuration 2 was always better than Configuration 1, and the optimization degree of Configuration 2 was 28~20%, showing a slight downward trend, indicating that the optimization degree of Configuration 2 is more significant when east–west entrance traffic decreases. Taken together, when the north–south entrance traffic flow was 3234 pcu/h and the east–west entrance traffic flow was 1213 pcu/h, the optimization degree of Configuration 2 was the largest, and the maximum value was 50%, which indicates that Parallel U-turns are feasible strategies to improve the safety performance of intersections.

As shown in Figure 8a, in terms of intersection operation efficiency, when the north–south entrance traffic flow gradually increased to 3234 pcu/h, the performance of Configuration 2 was always better than Configuration 1, and the optimization degree interval of Configuration 2 was 20~70%, showing an upward trend, which suggests that the optimization degree of Configuration 2 is more significant when the north–south entrance traffic flow increases. When the east–west entrance traffic gradually increased to 1265 pcu/h, the performance of Configuration 2 was always better than Configuration 1, and the optimization degree of Configuration 2 was 59~70%, showing a slight upward trend, which indicates that the optimization degree of Configuration 2 is more significant when the east–west entrance traffic increases. Taken together, when the north–south entrance traffic flow was 3234 pcu/h and the east–west entrance traffic flow was 1265 pcu/h, the optimization degree of Configuration 2 was the largest, and the maximum value was 70%, which indicates that Parallel U-turns are feasible strategies to improve the operational efficiency of the intersection.

As shown in Figure 8b, in terms of protecting the environment, when the traffic flow at the north and south entrances gradually increases to 3234 pcu/h, the performance of Configuration 2 is always better than that of Configuration 1, and the optimization degree of Configuration 2 is in the interval of 10% to 1%, showing an upward trend, which suggests that when the traffic flow at the north and south entrances decreases the optimization degree of Configuration 2 is more significant. When the east–west entrance traffic gradually increases to 1265 pcu/h, the performance of Configuration 2 is always better than Configuration 1, and the optimization degree of Configuration 2 is 4~1%, showing a slight downward trend, which indicates that the optimization degree of Configuration 2 is more significant when the east–west entrance traffic decreases. Taken together, when the north–south entrance traffic flow is 1213 pcu/h, and the east–west entrance traffic flow is 474 pcu/h, the optimization degree of Configuration 2 is the largest, and the maximum value is 10%, which indicates that Parallel U-turns are feasible in purifying the intersection environment.

As shown in Figure 8c, in terms of intersection safety performance, when the north–south entrance traffic flow gradually increases to 3234 pcu/h, the performance of Configuration 2 is always better than Configuration 1, and the optimization degree interval of Configuration 2 is 25~65%, showing an upward trend, which suggests that the optimization degree of Configuration 2 is more significant when the north–south entrance traffic flow increases. When the east–west entrance traffic gradually increases to 1265 pcu/h, the performance of Configuration 2 is always better than Configuration 1, and the optimization degree of Configuration 2 is 25~27%, showing an upward trend, which indicates that the optimization degree of Configuration 2 is more significant when the east–west entrance traffic decreases. Taken together, when the north–south entrance traffic flow is 3234 pcu/h, and the east–west entrance traffic flow is 474 pcu/h, the optimization degree of Configuration 2 is the largest, and the maximum value is 65%, which indicates that Parallel U-turns are feasible in improving the safety performance of intersections.

5. Analysis of the Results Based on the EWTM

The results of the comprehensive evaluation of fours are significant for people, vehicles, roads, and the environment. This paper formulates a comprehensive evaluation method based on the entropy-weighted-TOPSIS method [60] (EWTM).

5.1. Data Pre-Processing

5.1.1. Analysis of Sensitivity Analysis Simulation Results

In the sensitivity analysis, simulations were carried out for a total of 36 different sets of simulations for four configurations. And 10 indicators were obtained, including maximum queue length, vehicle delays, stopping delays, the number of stops, CO emissions, NOx emissions, VOC emissions, fuel consumption, traveling time, and TTC using VISSIM and SSAM. Each configuration had a set of simulation results for each case, and the simulation results for all configurations were combined into a matrix of 36 rows and 10 columns, as denoted by $A_i$:

$$A_i=[Mq{l}_i^k,V{d}_i^k,S{d}_i^k,N{s}_i^k,C{O}_i^k,N{O}_i^k,VO{C}_i^k,F{u}_i^k,T{T}_i^k,TT{C}_i^k]$$

(9)

where $i$ denotes number; $k$ denotes different combinations of traffic volumes from 1 to 36; $Mql$ denotes maximum queue length; $Vd$ denotes vehicle delay; $Sd$ denotes stopping delay; $Ns$ denotes the number of stops; $CO$ denotes CO emissions; $NO$ denotes NOx emissions; $VOC$ denotes $VOC$ emissions; $Fu$ denotes fuel consumption; $TT$ denotes journey time; and $TTC$ denotes post-intrusion time. The specific expression of the formula can be found in Appendix A.

These four matrices represent the simulation results for each configuration and for all traffic combinations. To compare the differences in indicators between the configurations for the same traffic combinations, the simulation results for different scenarios needed to be recombined and then analyzed for calculation. The simulation results of different configurations were combined under the same traffic volume combination.

$$\begin{array}{c}{X}^1=\left[\begin{array}{l}{A}_1(1,\cdot )\\ A_2(1,\cdot )\\ A_3(1,\cdot )\\ A_4(1,\cdot )\end{array}\right]\cdots X^k=\left[\begin{array}{l}{A}_1(k,\cdot )\\ A_2(k,\cdot )\\ A_3(k,\cdot )\\ A_4(k,\cdot )\end{array}\right]\cdots X^{36}=\left[\begin{array}{l}{A}_1(36,\cdot )\\ A_2(36,\cdot )\\ A_3(36,\cdot )\\ A_4(36,\cdot )\end{array}\right]\\ X={\left[X^1,X^2,X^3\cdots X^k\cdots X^{36}\right]}^T\end{array}$$

(10)

where $A$ denotes the $k$th row of $A_1(k,\cdot )$, which the simulation results obtained for the first configuration in the $k$th case. After rearranging the matrix, it was finally possible to summarize the simulation results for each traffic condition for four configurations containing results for 10 indicators, as shown by Equation (11).

$$\begin{array}{l}{X}^k={\left[\begin{array}{l}Mq{l}_1^k,V{d}_1^k,S{d}_1^k,N{s}_1^k,C{O}_1^k,N{O}_1^k,VO{C}_1^k,F{u}_1^k,T{T}_1^k,TT{C}_1^k\\ Mq{l}_2^k,V{d}_2^k,S{d}_2^k,N{s}_2^k,C{O}_2^k,N{O}_2^k,VO{C}_2^k,F{u}_2^k,T{T}_2^k,TT{C}_2^k\\ Mq{l}_3^k,V{d}_3^k,S{d}_3^k,N{s}_3^k,C{O}_3^k,N{O}_3^k,VO{C}_3^k,F{u}_3^k,T{T}_3^k,TT{C}_3^k\\ Mq{l}_4^k,V{d}_4^k,S{d}_4^k,N{s}_4^k,C{O}_4^k,N{O}_4^k,VO{C}_4^k,F{u}_4^k,T{T}_4^k,TT{C}_4^k\end{array}\right]}_{4\times 10}\\ k=1 to 36\end{array}$$

(11)

Finally, data from the simulation results for one of the traffic volume combinations were initially summarized in matrix $X^k$.

5.1.2. Standardization of Data

In the process of carrying out the evaluation of configurations, especially in multi-indicator evaluation systems, raw indicator data need to be standardized in order to eliminate the influence of different scales and orders of magnitude between individual indicators and to ensure the validity of the results. As each indicator is evaluated on different criteria, with some indicators being larger and others smaller, data were first positive for the same. Here, $y_{ij}$ represents the size of each element of the matrix $X^k$, which is the results of the simulation; $i$ denotes the $i$th configuration; $n$ denotes the total number of configurations, which was four; $j$ denotes the $j$th indicator; $m$ denotes the number of indicators, which was ten.

$$Y^k=\left[y_1,y_2\cdots y_j\cdots y_{10}\right]$$

(12)

$$y_j=\left[y_{1j},y_{{}_{2j}},y_{3j},y_{4j}\right]$$

(13)

For each of the 10 selected indicators, numbered 1 to 10. Among these indicators, the smaller the value of TTC, the better, and the larger the value of the remaining indicators, the better. The indicators, therefore, needed to be forwarded or reversed in order to facilitate subsequent scoring. In this paper, nine indicators were forwarded for maximum queue length, vehicle delay, stopping delay, the number of stops, CO emissions, NOx emissions, VOC emissions, fuel consumption, and journey time. Subsequently, higher values for the nine indicators after treatment represented better solutions.

$$y_j^{\prime }=\left[\begin{array}{l}\max \left\{y_{1j},y_{2j},y_{3j},y_{4j}\right\}-y_{1j}\\ \max \left\{y_{1j},y_{2j},y_{3j},y_{4j}\right\}-y_{2j}\\ \max \left\{y_{1j},y_{2j},y_{3j},y_{4j}\right\}-y_{3j}\\ \max \left\{y_{1j},y_{2j},y_{3j},y_{4j}\right\}-y_{4j}\end{array}\right],j\ne 10 ,j=1 to 9.$$

(14)

Thereby, $Y^{k^{\prime }}$, $Y^{k^{\prime }}=\left[y_1^{\prime },y_2^{\prime },\cdots ,y_j^{\prime },\cdots ,y_{10}^{\prime }\right]$ was obtained, where $y_{10}^{\prime }=y_{10}$.

The normalization method was used to standardize $Y^{k\prime }$ and thus standardize the scale. A linear transformation of data was performed, such that the data results were all mapped to the interval [0, 1], which converted absolute indicators into relative indicators. For $Y^{k^{\prime }}$, where each element was denoted by $y_{ij}^{\prime }$, each indicator result $y_{ij}^{\prime }$ was normalized using the formula (Equation (15))

$$y_{ij}^{″}=\frac{y_{ij}^{\prime }-\min \left\{y_{1j}^{\prime },\cdots ,y_{4j}^{\prime }\right\}}{\max \left\{y_{1j}^{\prime },\cdots ,y_{4j}^{\prime }\right\}-\min \left\{y_{1j}^{\prime },\cdots ,y_{ij}^{\prime }\right\}}, i=1 to n, j=1 to m.$$

(15)

The combination provided $Y_{4\times 10}^{″}$, and this matrix was the final datum after the original data had been integrated and normalized. The subsequent process of calculating the weights required the application of a ${\left[Y^{K″}\right]}_{4\times 10}$ data matrix.

5.2. Entropy Method for Calculating Objective Weights of Indicators

After data pre-processing, the entropy method was applied to the processed matrix ${\left[Y^{K″}\right]}_{4\times 10}$ to calculate the objective weights of the indicators. In the sensitivity analysis, there were a total of 144 traffic configurations, so $k=1~144$. The entire process of calculating indicator weights by the entropy method is analyzed below for one of the traffic scenarios with data ${\left[Y^{K″}\right]}_{4\times 10}$ as the object.

The share of indicator in all indicators under the configuration is shown in Equation (16):

$$P_{ij}=\frac{y_{ij}^{″}}{{\displaystyle \sum _{i=1}^n{y}_{ij}^{″}}}, i=1 to 4, j=1 to 10$$

(16)

The entropy value of the indicators can be calculated, as shown in Equation (17):

$$e_j=-k{\displaystyle \sum _{i=1}^n{p}_{ij}\ln \left(p_{ij}\right)}$$

(17)

where $k=1/\ln \left(n\right)$ and $e_j\ge 0$.

The information entropy redundancy of the indicator $j$ could be calculated as shown in Equation (18):

$$d_j=1-e_j$$

(18)

The weight of the indicator $j$ could be calculated, as shown in Equation (19):

$${\omega }_j=\frac{d_j}{{\displaystyle \sum _{j=1}^m{d}_j}}$$

(19)

The weights of all indicators under group $k$ traffic conditions (Equation (20)):

$$W_k=\left[w_{k,1},w_{k,2},\cdots ,w_{k,j},\cdots ,w_{k,10}\right]$$

(20)

The combined weight of the $j$th indicator in all cases is shown in Equation (21):

$$W_{entropy\left(j\right)}=\frac{{\displaystyle \sum _k^{\infty }\left(W_{k,j}\right)}}{{\displaystyle \sum _j^{10}{\displaystyle \sum _k^{36}\left(W_{k,j}\right)}}}$$

(21)

The final objective weight $W$, calculated by the entropy method, is shown in

Table 7. The results of the calculation of entropy method weighting.

Indicators	$\mathit{M}\mathit{q}\mathit{l}$	$\mathit{V}\mathit{d}$	$\mathit{S}\mathit{d}$	$\mathit{N}\mathit{s}$	$\mathit{C}\mathit{O}$	$\mathit{N}\mathit{O}$	$\mathit{V}\mathit{O}\mathit{C}$	$\mathit{F}\mathit{u}$	$\mathit{T}\mathit{T}$	$\mathit{T}\mathit{T}\mathit{C}$
Weighting	10.025	5.685	5.497	5.873	11.406	11.406	11.406	11.405	3.956	23.341

:

5.3. Improving the TOPSIS Method for Scoring

Based on the combined weights obtained, each solution was scored in combination with the TOPSIS method.

• Constructing the weighting matrix $\mathit{Z}$.

For each traffic condition, a weighting matrix existed, taking any one traffic condition as the object of analysis, corresponding to processed indicator data as ${\left[Y^{K″}\right]}_{4\times 10}$ and the combined weight as $W_{com}$. The weighting matrix $Z_k$ for the $k$ traffic conditions is shown in Equation (22):

$$Z^k=\left[\begin{array}{cccc}{y}_{11}^{″}\cdot W_{com.1}& y_{12}^{″}\cdot W_{com.2}& \cdots & y_{1m}^{″}\cdot W_{com.m}\\ y_{21}^{″}\cdot W_{com.1}& y_{22}^{″}\cdot W_{com.2}& \cdots & y_{2m}^{″}\cdot W_{com.m}\\ ⋮& ⋮& \ddots & ⋮\\ y_{n1}^{″}\cdot W_{com.1}& y_{n2}^{″}\cdot W_{com.2}& \cdots & y_{nm}^{″}\cdot W_{com.m}\end{array}\right], n=1 to 4, m=1 to 10$$

(22)

where $n$ denotes the number of configurations; $m$ denotes number of indicators; $k$ denotes group $k$ traffic combinations, $k=1~144$.

• Finding the best and worst ideal points

The maximum and minimum values of each column of the weighted matrix species were chosen as the optimal and inferior ideal points, respectively (Equation (23)):

$$Z^{k,+}=\left[Z_1^{k,+},Z_2^{k,+},\cdots ,Z_j^{k,+},\cdots Z_m^{k,+}\right]$$

(23)

where $Z_j^{k,+}=\max \left(Z_j^k\right)$. The calculation of the virtual worst ideal point is shown in Equation (24).

$$Z^{k^{*,-}}=\left[Z_1^{k^{*,-}},Z_2^{k^{*,-}},\cdots ,Z_j^{k^{*,-}},\cdots ,Z_m^{k^{*,-}}\right]$$

(24)

where $Z_j^{k^{*,-}}=2Z_j^{k,-}-Z_j^{k,+}.$

• Calculation of optimal and inferior distances

The distance of the $i$th configuration to the virtual optimal ideal point is defined in Equation (25):

$$D_i^{+}=\sqrt{{\displaystyle \sum _{j=1}^n{\left(Z_j^{k,+}-z_{ij}^k\right)}^2}},\left(i=1,2,\cdots ,n\right)$$

(25)

The distance of the $i$th configuration to the virtual worst ideal point is defined in Equation (26):

$$D_i^{-}=\sqrt{{\displaystyle \sum _{j=1}^n{\left(Z_j^{k,+}-z_{ij}^k\right)}^2}},\left(i=1,2,\cdots ,n\right)$$

(26)

• For $k$ combinations of traffic volumes, the score of the $i$th configuration is shown in Equation (27):

  $$S{c}_i^k=\frac{D_i^{-}}{D_i^{+}+D_i^{-}}$$

  (27)

The scores for all configurations under k combinations of traffic volumes are shown in Equation (28):

$$S{c}^k={\left[S{c}_1^k,S{c}_2^k,S{c}_3^k,S{c}_4^k\right]}^T$$

(28)

• For all traffic combinations, the scores for all configurations are shown in Equation (29):

$$Sc=\left[S{c}^1,S{c}^2,\cdots ,S{c}^k,\cdots ,S{c}^{36}\right]$$

(29)

After these steps, the combined score for each configuration could be obtained for each case based on the results of the simulation calculations (

Table 8. Design configurations score for the intersection at 3234 veh/h to the north.

Traffic Mix	U-Turn at the Intersection	Parallel U-Turn at the Intersection	U-Turn in the Median	Parallel U-Turn in the Median
3234–474	0.47	0.42	0.38	0.34
3234–632	0.44	0.40	0.36	0.32
3234–791	0.45	0.37	0.29	0.35
3234–949	0.36	0.41	0.29	0.34
3234–1107	0.32	0.38	0.25	0.32
3234–1265	0.30	0.36	0.24	0.32

).

The final scores for each of the four configurations were calculated for each of the 144 traffic volume combinations, and the best-performing design solution for each traffic volume combination was selected, as shown in Figure 9.

From Figure 9, it is evident that the XiaoZhai intersection in Xi’an serves as an example. Parallel U-turns are proven suitable for intersections with high inbound traffic volumes. When the traffic volume at the north–south entrance ranged from 2021 pcu/h to 3234 pcu/h, and the traffic volume at the east–west entrance ranged from 949 pcu/h to 1265 pcu/h, the design of Parallel U-turns outperformed the traditional U-turn design, leading to significant enhancements in intersection operational efficiency, environmental quality, and safety performance. On the other hand, in situations where inbound traffic volumes are lower, the traditional U-turn design is more effective. Specifically, when the traffic volume at the north–south entrance falls below 2021 pcu/h but remains equal to or greater than 1213 pcu/h, and the traffic volume at the east–west entrance falls below 949 pcu/h but remains equal to or greater than 474 pcu/h, the traditional U-turn design demonstrates superior performance.

The simulation experiment and calculations conducted in this paper specifically pertain to the XiaoZhai intersection in Xi’an. However, the same methodology can be applied to the selection of turnaround design options for other intersections across China. Implementing a Parallel U-turn design in new or reconstructed intersections yields favorable outcomes by boosting operational efficiency, promoting environmental cleanliness, and enhancing intersection safety performance. The EWTM evaluation method was employed in this study, and the findings can be effectively leveraged in practical engineering applications.

6. Conclusions

U-turn vehicles demonstrate distinct time-of-day patterns, resulting in significant congestion at intersections during peak periods. This congestion manifests in the form of extensive queues, and in more severe cases, it may impede the entry of straight-ahead vehicles into the intersection during periods of intense traffic flow. The selection of both traditional U-turn design configurations and Parallel U-turn design configurationsare an important matter that warrants comprehensive research and thorough discussion. By evaluating the U-turn design configuration based on intersection operation efficiency, safety performance, and environmental preservation, this paper can not only offer practical engineering solutions but also align with the current era’s development and propose innovative ideas for sustainable transportation. This comprehensive approach ensures the consideration of all relevant factors and contributes to the advancement of green transportation initiatives.

This study comprehensively assessed four distinct U-turn design alternatives (traditional U-turn at the intersection, Parallel U-turn at the intersection, traditional U-turn in the median, and Parallel U-turn in the median) using the EWTM (Evaluating U-turn Traffic Models) approach. This paper employed output traffic parameters obtained from VISSIM microsimulation software and SSAM safety assessment software for a comprehensive evaluation of these alternatives.

Taking the XiaoZhai intersection located in Xi’an, China, as an example, this paper draws the following conclusions:

• Parallel U-turns are suitable for intersections experiencing high inbound traffic volumes, ranging from 2021 pcu/h to 3234 pcu/h for the north and south entrances, and from 949 pcu/h to 3234 pcu/h for the east and west entrances.
• The traditional U-turn design is appropriate for intersections with lower traffic volumes, where the north and south entrances have traffic volumes below 2021 pcu/h but still equal to or greater than 1213 pcu/h, and the east and west entrances have traffic volumes below 949 pcu/h but still equal to or greater than 474 pcu/h.
• The number of U-turn vehicles plays a decisive role in the choice of the turnaround form.
• In actual engineering practice, traffic survey data can be obtained through engineering feasibility studies to determine the optimal organizational design form of intersection U-turns.
• Although a Parallel U-turn can still meet the demand for U-turn vehicles when the intersection saturation is high and performs well, with the growth of traffic volume, the congestion at intersections may not only come from U-turn vehicles and further research is needed to find out a better and more comprehensive solution to intersection congestion in order to further promote the development of green transportation.

Due to the limitations of space in this article, only one representative intersection has been analyzed in depth. Before applying Parallel U-turns on a large scale, a certain number of aspects should also be taken into consideration, including but not limited to 1. This paper focuses on the organization of Parallel U-turns at intersections with dedicated U-turn lanes, questioning if the same applies at intersections without U-turn lanes? 2. Does the capacity of Parallel U-turns affect the operational performance of intersections? What are the factors influencing the capacity of Parallel U-turns and their underlying mechanisms? The authors aim to focus on these issues in future research.