Understanding the Performance of Multilane Expressway Exit Design and a Traffic Organization Strategy Based on VISSIM Micro-Simulation and a Comprehensive Evaluation Method

Abstract

Frequent consecutive lane changes and the barrier effect produced by trucks can always contribute to rapid access to the capacity bottleneck at expressway exits, thereby reducing operational performance. This paper proposes to retrofit the conventional interchange design to reduce lane changes and compares them with conventional design and passenger vehicle and truck separation (PVTS) design from multiple fields via VISSIM numerical simulation, which is developed and calibrated with traffic data collected on the eight-lane expressway in China at four levels of service (LOS). Comprehensive evaluation (CE) results reveal that the PVTS strategy improves operational performance by 10–20% at multilane expressway exits, while lane separation around interchanges also yields a similar gain. However, ramps on both the left and right sides show less effective improvement and are even negative at LOS-A and LOS-B. All PVTS and improved designs produce a better comprehensive improvement ratio with a LOS decrease, and improved designs achieve optimal performance below LOS-B with construction cost taken into consideration.

1. Introduction

As the exclusively controlled hub connecting the expressway to the surrounding road network, the interchange plays a vital role in transferring traffic flow. However, the exit diverging area on interchange has also been regarded as a bottleneck section that produces local delay and congestion, ultimately leading to decreased transport efficiency along the entire road mainline [1]. The bottleneck can be attributed to drivers’ difficulty in identifying exit locations and changing lanes consecutively within specific deadlines, exacerbating their workload, increasing error probability, and disrupting the smooth traffic flow on the mainline due to frequent consecutive lane changes [2,3,4,5]. Furthermore, this issue has been associated with a higher probability of accidents in the diverging area, as stated in various literatures [6,7,8]. This problem may be particularly prominent on multi-lane expressways, where diverging traffic requires longer lane change maneuvers. Heavy congestion can also exacerbate pollutant emissions, despite upgraded vehicle emission standards, with current studies commonly associating them [9,10]. Mechanistic explanations for this phenomenon have pointed to a significant increase in fuel consumption and emissions during the transition from free flow to congestion [11]. Obviously, this issue contradicts the sustainable development concept that has attracted public attention in recent years, leading scholars to focus on research on improvement measures [12,13].

The capacity bottleneck of the expressway diverging area can be affected by traffic characteristics, geometric design, and management strategy, while current research has focused on traffic control measures and demand management [14,15]. Some researchers believe that congestion alleviation can be achieved through traffic demand reduction and emphasize the gains from expressway charges and traffic restrictions. However, this approach fails to deal with the problem in the long term, and the travel restriction can be a nuisance for travel, as suggested by other articles [16,17,18,19]. Optimization of traffic management strategies such as ramp metering, sign optimization, and dynamic guided detours can also positively contribute to congestion alleviation; these parameters were extensively investigated in research [20,21,22]. Nevertheless, the optimization of road geometry design can provide the most direct and effective approach to achieving a fundamental solution; thus, this paper presents an upgraded geometric design and traffic organization strategy for a multi-lane expressway diverging area with the aim of reducing lane changes. In addition, the authors also address the disturbance to exiting vehicles caused by the barrier effect produced by trucks, which can be effectively tackled with the PVTS strategy, which helps improve a transport environment and provide an essential idea for this paper [23,24].

The emerging technology of numerical simulation has found extensive application in various disciplines. In the field of transportation, numerous studies have shown the reliable, effective, and economic advantages of using VISSIM numerical simulation [25,26]. To ensure reliability and realism, it is essential to incorporate real traffic data in the development of the simulation model and the calibration of parameters to recreate specific traffic scenarios [27]. Furthermore, benefiting from the emergence of target detection algorithms, the use of unmanned aerial vehicles (UAV) in data measurement can achieve better efficiency compared to conventional radar measurements [28].

Despite the fact that the VISSIM micro-simulation approach has been widely adopted in previous studies, those studies mainly focused on the evaluation of one aspect, such as conflicts, emissions, or delays [29,30,31]. To meet the sustainability objective, the design and construction of road infrastructure must balance factors that are not limited to traffic performance, such as construction costs and consumption. Therefore, comprehensive evaluation (CE) was introduced into this study.

The core of the CE method, which aims to assign weights to indicators with different amounts of information, has been successfully applied in various fields, including economics [32], medicine [33], environment [34], and transportation [27]. Among those, the entropy method, which determines the weights based on the variability of indicators, is the most frequently used type of method [35]. Inevitably, however, the entropy method also possesses certain limitations as it ignores the correlations between indicators. The CRITIC method, on the other hand, compensates for this deficiency, which matches the data characteristics of the organic system of transportation [36].

To address the capacity bottleneck of the diverging area and provide sustainable expressway transportation, the authors explore from the perspective of the interchange horizontal layout, carriageway elements, and traffic organization optimization. Specifically, the paper discusses PVTS and the lane restriction (LR) strategy that is commonly used in China for eight-lane expressways, and presents two innovative design strategies that minimize lane changes in diverging areas. Traffic data from LOS-A to LOS-D is collected and processed to develop and calibrate the VISSIM model. Aiming at achieving an objective and scientific understanding of the sustainable performance of different designs, the authors used the CRITIC method to assign weights and score to different indicators and compared it with the traditional entropy method, pointing out the outstanding advantages of the CRITIC method used in the transportation field. This initiative achieves a breakthrough in research methodology in the transportation field. For one thing, the organic combination of micro-simulation and comprehensive evaluation provides an effective approach for engineers to make efficient scheme decisions. For another, it explores the potential of the CRITIC method in the transportation discipline, addresses the deficiencies of the traditional entropy method, and effectively upgrades the research method for related problems. The final results demonstrate the advantages and adaptability of each design under different traffic conditions.

The remainder of this paper is structured as follows: Section 2 provides an overview of the research methodology, including design description, data collection and analysis, VISSIM simulation, and the process of CE. Section 3 presents the simulation and CE results of the investigation scenarios and analyzes the results for dynamic traffic conditions. In Section 4, the research findings are discussed in detail. Finally, Section 5 summarizes the conclusions drawn from the study.

2. Methodology

2.1. Scheme Design

In the conventional design for an eight-lane expressway under the management strategy in China, four lanes are equally allocated to passenger vehicles and mixed traffic in each direction, but without a guardrail separating them. This lane restriction (LR) scheme is frequently taken on mainlines with a design speed of 120 km/h, and restrictions are removed around interchanges to provide conditions for diverging traffic to change lanes. However, passenger vehicles, especially those in the fast lane, must make consecutive lane changes within diverging times. In addition, trucks in the curb lanes disturb the interweaving of traffic. Consequently, congestion frequently occurs around interchanges during periods of heavy traffic. Figure 1 shows the conventional design for an interchange exit ramp, which features two-lane ramps in each direction of the 8-lane mainline.

Figure 2 presents a solution that employs PVTS along the mainline. This approach allows passenger vehicles and trucks to travel and diverge continuously on their dedicated lanes. For one thing, that eliminates the barrier effect produced by trucks that LR failed to fully address, promoting a considerably smoother traffic flow. For another, the separated section facilitates diverging traffic, allowing them to change lanes only once at most to exit into the deceleration lane. However, the implementation of PVTS requires significant extra construction, including an auxiliary lane and the addition of segregated subgrade or barrier facilities.

Research has elucidated the mechanisms underlying capacity bottlenecks resulting from consecutive lane changes occurring within seconds at the diverging areas, which provide a clear path for optimizing the design by reducing the lateral range of lane changes. While PVTS provides an advantageous solution, an optimization approach can be attempted to avoid the drawbacks of significant consumption. Specifically, as shown in Figure 3, optimized scheme A proposes partitioning the four-lane main-line into dual two-lane sections only around interchanges, enabling vehicles to exit with only one lane change, while the basic section still applies the LR strategy. This design obviates substantial segregated subgrade and barrier facilities and also omits auxiliary lanes, thereby reducing the required mainline pavement area. However, the barrier effect produced by trucks can still hardly be prevented in the mixed trac section.

Along this line, the authors also propose an alternative design and organization strategy to eliminate the necessity of section separation for each direction. Scheme B, as depicted in Figure 4, features a single-lane exit ramp on each side of the mainline. In this way, merely a short, separated central divider is needed around diverging areas, which obviates section separation in the same direction. Inevitably, the left-side exit ramp would have a lower capacity and a higher crash risk since the speed differences between vehicles in the diverging areas of the left-side ramp can always be higher under the right-hand drive rule in China [37].

2.2. Data Collection and Analysis

Since the optimized design has never been constructed yet, it is essential to develop and calibrate an accurate and stable simulation model to recreate the real traffic scenario. Thus, actual traffic data should be collected to create a simulation model for evaluating each design. The following data were collected:

• Vehicle speed and volume, which were used to calculate traffic density.
• Time headway and lane change trajectories are used to calibrate vehicles’ behavior in simulation.
• The diverging ratio of the investigated interchange.
• The proportion of passenger vehicles and trucks in the investigated section.

Furthermore, optimal conditions required to conduct this task include peak hours of several levels of service (LOS), clear weather, free from any unforeseen crashes or facility maintenance, and avoidance of obvious congestion [38]. Therefore, a preliminary investigation is imperative to determine specific details. Notably, roadside radar equipment can be replaced with UAVs for efficient data collection, as it always attracts drivers’ attention and disrupts the normal traffic flow.

The use of UAVs provides a cost-effective solution for data collection since their compact and portable size enables up to 30 min of continuous aerial photography per battery charge. Additionally, it offers a wide field of view, enhancing the spatial coverage of data measurement. However, this method necessitates more stringent image processing requirements, which can be addressed through the use of an image recognition tool based on the YOLOv3 target detection algorithm. To establish an accurate coordinate network in the video, it is necessary to first employ a GPS-RTK mobile station to measure the coordinate information of string feature points along the roadside. Figure 5 shows the used devices and tools.

Obtained vehicle trajectory data frequently contains errors that could result from UAV in-flight jitter, and these errors may be further amplified at speed and acceleration. To eliminate this effect, the authors propose the application of the Kalman filter, which serves as an optimal autoregressive data processing algorithm that leverages the linear system state equation to estimate the system state optimally from input and output observations [39]. Equation (1) shows the Kalman filter process.

$$\left\{\begin{array}{l}{\widehat{x}}_t^{-}=F{\widehat{x}}_{t-1}+B{u}_{t-1}\\ P_t^{-}=F{P}_{t-1}{F}^T+Q\\ K_t=P_t^{-}{H}^T{(H{P}_t^{-}{H}^T+R)}^{-1}\\ {\widehat{x}}_t={\widehat{x}}_t^{-}+K_t(z_t-H{\widehat{x}}_t^{-})\\ P_t=(I-K_{t}H)P_t^{-}\end{array}\right.$$

(1)

The Kalman filter process involves several components. ${\widehat{x}}_t^{-}$ and ${\widehat{x}}_t$ denote the predicted and optimal estimate values of the system state at moment t, respectively. $U_{t-1}$ denotes the acceleration at moment t. F and B serve as the state transfer matrix, $P_t^{-}$ and $P_t$ represent the covariance estimation matrix and the optimal estimation matrix, respectively. The Kalman coefficient is denoted as $K_t$, and Q is the predicted noise covariance matrix. The observed matrix and observed noise covariance matrix are represented by H and R, respectively. $Z_t$ denotes the measured position value, and I serves as the unit matrix.

The optimal predicted trajectory can be obtained by utilizing the Kalman filter, which further solves for several critical parameters, including vehicle speed, density, volume, and time headway.

2.3. VISSIM Simulation

Numerical simulation technology has proven to be the most expeditious and efficient approach for comparing various schemes within a given context. Additionally, certain optimized schemes have not yet been implemented, which curtails the ability to observe and analyze operational performance in reality. Consequently, an assessment would necessitate a numerical simulation, for which VISSIM presents an appropriate alternative.

2.3.1. Model Development

Realistic recreation of traffic flow is contingent upon precise model development and calibration; therefore, the appropriate geometric parameters should be selected firstly during the construction of the road network. Additionally, setting certain significant traffic behavior parameters such as speed distribution, time headway characteristics, and lane change distance is also necessary to ensure reality. The model will be constructed based on the guidance in the Green Book for the geometric design of an expressway interchange with a design speed of 120 km/h, which includes the following parameters [40].

• For a common ramp design speed of 60 km/h in China, the deceleration lane length should not be less than 490 ft (150 m).
• For the two-lane exit ramp, an auxiliary lane not less than 1500 ft (450 m) should be provided.
• The taper of the deceleration lane and auxiliary lane should be 300ft (90 m) in length.
• The lane separation site of LSI should match a decision sight distance of 12–14s travel from the start of the taper of the exit, which comes to approximately 1500 ft (450 m).

2.3.2. Model Calibration

Mean absolute percent error (MAPE), which gauges the error between the VISSIM model and reality, has been frequently utilized as an accuracy metric in calibration. Capacity and density are commonly examined indicators, with the calculations shown in Equations (2) and (3):

$$MAP{E}_C=\left({\displaystyle \sum _{i=1}^n{C}_v^i}-{\displaystyle \sum _{i=1}^n{C}_f^i}\right)/{\displaystyle \sum _{i=1}^n{C}_f^i}$$

(2)

$$MAP{E}_K=\left({\displaystyle \sum _{i=1}^n{K}_v^i}-{\displaystyle \sum _{i=1}^n{K}_f^i}\right)/{\displaystyle \sum _{i=1}^n{K}_f^i}$$

(3)

where $C_v^i$ and $K_v^i$ denote the simulated capacity and density in the VISSIM model (veh/h) respectively, while $C_f^i$ and $K_f^i$ denote the investigated result, i denotes the traffic flow, and n denotes the total simulation times.

Moreover, the Greenshields model will also serve as an additional benchmark to verify the simulation accuracy, which depicts the objective connection among the three traffic parameters in real traffic. As shown in Figure 6, volume increases towards its maximum capacity at a density equivalent to half of the jam density. Subsequently, volume experiences a decrease, eventually leading to the traffic flow being divided into smooth and congested areas.

The VISSIM output will be fitted with Greenshields formula, and the goodness of fit (R²) will be utilized to determine whether the simulated traffic flow matches the actual. This confirms the rationality and accuracy of the simulation setup.

2.3.3. Results Indicators

In order to achieve a comprehensive evaluation, the indicators selected to assess the simulation result should cover multiple perspectives, including efficiency, environmental impact, and safety. The following indicators have been considered by the authors:

Capacity, which is a frequently applied road design indicator, and its improvement should be the primary concern. Optimized design aims to breach the capacity bottleneck of interchanges, which is often aggravated by the barrier effect produced by trucks. Among the VISSIM simulation results, the capacity can be captured using the indicator of the number of vehicles on a certain section.

Travel time expresses transport efficiency and reflects vehicles speed on a certain section, as well as supports the estimate of delay and congestion.

Fuel consumption represents the environmental impact of carbon and nitrogen emissions as well as petrochemical resource consumption. Since the UN initiative for energy conservation and emission reduction, green sustainability has attracted much attention.

Conflicts serve as safety hazards and risks in characterized road sections and can be predicted and evaluated using the surrogate safety assessment model (SSAM) introduced by the Federal Highway Administration (FHWA). The total number of conflicts will be specifically expressed as an indicator.

Delay expresses the smoothness or congestion of the section, and also serves as a significant indicator to assess transport quality.

Additionally, to ensure accuracy and equality in design evaluation, VISSIM result output nodes should be equidistant and provide complete coverage of the diverging area.

2.4. Comprehensive Evaluation

Conventional CE typically involves the handling of indicators across various fields, each containing varying amounts of information. Among these fields, weight assignment plays a significant role. However, since each expert always holds individual views on the importance of design performance indicators in various aspects, reaching a unified and accepted conclusion is difficult, resulting in subjective evaluation that can lead to excessive randomness. In such a case, an objective approach can be more appropriate.

Table 1. Comparison of three typical weight assignment methods in CE.

Method	Data Size	Data Volatility	Data Correlation	Remarks
Entropy	N/A	N/A	N/A	entropy, information amount
CRITIC	N/A	√	√
Factor Analysis Method	N/A	N/A	√	information concentration

presents a comparison of weight assignment methods frequently used in CE. This comparison provides insight into the mechanics of these methods and reflects the adaptation scene.

Road transportation has been commonly recognized as a complex organic system consisting of interdependent factors across various aspects, whereas the CRITIC method holds features that match handling the volatility and correlation of traffic data. The two crucial parameters, volatility (contrast intensity) and conflict (correlation), are positively associated with the assigned weight. The standard deviation and correlation coefficient are used to express the two parameters, respectively. Higher contrast intensity stands for higher volatility, whereas the correlation coefficient represents the opposite of the conflict. The specific calculation is shown from Equation (4) to Equation (13).

• Indicator results matrixing.

$$A_k=\left[\begin{array}{c}{V}_{1,k},T_{1,k},F_{1,k},C_{1,k},D_{1,k}\\ V_{2,k},T_{2,k},F_{2,k},C_{2,k},D_{2,k}\\ ⋮ ⋮ ⋮ ⋮ ⋮\\ V_{i,k},T_{i,k},F_{i,k},C_{i,k},D_{3,k}\end{array}\right]$$

(4)

As shown in Equation (4), Matrix A_k for original data in the scene k contains the five indicators for each design, where V_i,k, T_i,k, F_i,k, C_i,k, and D_i,k denote capacity, travel time, fuel consumption, conflicts, and delay of the ith design.

2.

Data normalization.

$$X=\left[x_1,x_2,x_3,x_4,x_5\right]$$

(5)

$$x_j={\left[x_{1j},x_{2j},\cdots ,x_{ij}\right]}^T$$

(6)

Data normalization aims to eliminate the effects of dimensionality and measure data with a uniform standard, before which each indicator for different designs will be transferred to x_j, as shown in Equations (5) and (6).

$${x^{\prime }}_{ij}=\frac{x_{ij}-\max \left\{x_{1j},x_{2j},\cdots ,x_{ij}\right\}}{\max \left\{x_{1j},x_{2j},\cdots ,x_{ij}\right\}-\min \left\{x_{1j},x_{2j},\cdots ,x_{ij}\right\}}$$

(7)

Equation (7) shows forward normalization for x_ij, which denotes capacity and is expected to perform at a higher value. Conversely, the other four that are expected to lower take the inverse normalization process in Equation (8).

$${x^{\prime }}_{ij}=\frac{\max \left\{x_{1j},x_{2j},\cdots ,x_{ij}\right\}-x_{ij}}{\max \left\{x_{1j},x_{2j},\cdots ,x_{ij}\right\}-\min \left\{x_{1j},x_{2j},\cdots ,x_{ij}\right\}}$$

(8)

3.

Calculating information volume.

This part involves the initial calculation of two crucial parameters, among which contrast intensity can be expressed as a standard deviation by Equation (9), and the correlation coefficient represents the other, which is calculated by Equation (10):

$$S_j=\sqrt{\left({\displaystyle \sum _{i=1}^n{\left({x^{\prime }}_{ij}-(1/(n{\displaystyle \sum _{i=1}^n{x^{\prime }}_{ij}}))\right)}^2}\right)/\left(n-1\right)}$$

(9)

$$R_j={\displaystyle \sum _{j^{\prime }=1}^m\left(1-r_{{j,j}^{\prime }}\right)}$$

(10)

where S_j and R_j denote the standard deviation and correlation coefficient of the jth indicator, respectively, n and m denote number of design and indicators, respectively, and r_j,j′ denotes the correlation coefficient of the jth and j′th indicators.

After that, the volume of information can be determined using Equation (11).

$$C_j=S_j\times {\displaystyle \sum _{j^{\prime }=1}^m\left(1-r_{{j,j}^{\prime }}\right)}=S_j\times R_j$$

(11)

4.

Weight assignment and scoring

A higher value of C_j implies a more significant effect and corresponds to a higher weight assigned to the objective weight of the jth indicator in kth case, as expressed in Equation (12). Furthermore, the score of the ith design in kth case can be obtained using Equation (13).

$$w_{k,j}=C_j/{\displaystyle \sum _{j=1}^m{C}_j}$$

(12)

$$z_i={\displaystyle \sum _{j=1}^m\left(w_{k,j}\times \left({x^{\prime }}_{ij}/{\displaystyle \sum _{i=1}^n{x^{\prime }}_{ij}}\right)\right)}$$

(13)

3. Results and Analysis

3.1. Data Collection and Processing

The data collection for this study has been planned to occur at several sections approaching exits on the eight-lane G30 and G65 national expressways, which serve as critical traffic arteries in Shaanxi Province, China. These sections are comprised of a four-lane mainline and a two-lane ramp in each direction with design speeds of 120 km/h and 60 km/h, respectively. Prior to the investigation, the AMAP platform was used to determine the investigation period of various LOS with congestion delay indexes in the vicinity of Xi’an, as shown in Figure 7.

Table 2. Basic information of the investigated section and interchanges.

Interchange	Expressway	Design Speed (km/h)		Management Strategy
		Mainline	Ramp
Xiwu	G30	120	60	LR
Epanggong	G30	120	60	LR
Xinqu	G6522	120	60	PVTS

contains some fundamental information about the surveyed section, while Figure 8 displays geographical locations and actual views.

The data collection in this study will focus on several significant traffic situations, specifically from LOS-A to LOS-D, which represent the majority states of the surveyed expressway. Figure 9 shows vehicle speed distribution at the four LOS, which will serve as a critical parameter in the development and calibration of the VISSIM model. Additionally, Figure 10 illustrates the time headway, which is another calibration parameter.

The message conveyed in Figure 9 tends to suggest that there is no significant difference in speed distribution between the LR and PVTS on the basic section prior to the interchange. However, subtle variances can still be observed in the following ways:

• On passenger vehicle lanes, PVTS displays marginally slower vehicle speeds than the LR in smooth traffic conditions. However, as the LOS decreases, the PVTS regains its advantage, particularly noticeable on the curb lane.
• On truck lanes, the fast lane shows a similar trend to the aforementioned, while PTVS exhibits an advantage in the curb lane.

This phenomenon can be explained by the fact that certain slow-moving vehicles, or “turtle cars”, occupying the mixed lane have shifted to the passenger vehicle lane, producing a negative impact. Although this impact is negligible at low LOS, as all vehicle speeds will be suppressed, it is evident in the figure that the number of “turtle cars” on PVTS truck lanes is fewer than that on LR mixed lanes. After excluding this effect, the benefits of PTVS in mitigating the barrier effects produced by trucks can still be interpreted. The same explanation still applies to the clarification for Figure 10.

Lane changes are the most significant vehicle behavior in expressway diverging areas, involving two steps: gap queuing and acceptance and lane change execution. The VISSIM model typically defaults to a lane change execution length of 200 m. However, in practice, that will be influenced by various factors, such as speed, driver experience and style, and urgency. Thus, the authors conducted a dedicated investigation for calibration. The findings indicate that lane change execution time ranges intensively from 3.6 s to 9 s, with a moderately positive correlation between length and speed, as shown in Figure 11. The average lane change execution length, about 160 m, has been entered into the vehicle behavior parameters of the VISSIM model.

3.2. VISSIM Results

Table 3. MAPE of the calibration at the four LOS, where T denotes through traffic and D denotes the diverging flow.

Flow	LOS-A		LOS-B		LOS-C		LOS-D
	T	D	T	D	T	D	T	D
Investigated Capacity (veh/h)	2201	704	3379	1099	4811	1511	5681	1924
Simulated Capacity (veh/h)	2209	691	3353	1089	4708	1499	5434	1815
Individual MAPE (%)	0.36	−1.85	−0.77	−0.91	−2.14	−0.79	−4.35	−5.67
MAPE (%)	−0.17		−0.80		−1.82		−4.68
Investigated Density (veh/km)	27.75		43.21		63.60		80.92
Simulated Density (veh/km)	27.59		43.22		63.06		80.77
MAPE (%)	−0.57		0.02		−0.86		−0.19

presents the calibration results of the VISSIM model across the four investigated LOS, where the diverging ratio is about 25%, with trucks accounting for 30% of the traffic. MAPE for both capacity and density remains at acceptable levels, as previous studies have confirmed that MAPE values under 15% can provide accurate and credible guidance for engineering practice [41,42]. Furthermore, to demonstrate the accuracy of the VISSIM model, various volumes were inputted, and the density results presented in Figure 12 align with the Greenshields model, exhibiting a 99% R². This outcome indicates that the simulation matches the actual situation with fully plausible results.

The simulation results for the four investigation scenarios are depicted in Figure 13. The analysis shows that under LOS-A and LOS-B, LSI and PVTS significantly improve CI. Although PVTS has a lower mainline speed than LR at this LOS and its potential for exit diverging area has not been completely utilized, it still provides a significant improvement in overall performance. On the other hand, LRER failed in all aspects of the competition with CI. This can be explained by the fact that the significant speed dispersion produced in the diverging area on the fast lane of the left contributes to more conflicts and congestion at that LOS, and this issue cannot be resolved by reducing lane changes.

However, significant changes have occurred under LOS-C and LOS-D, with LRER regaining the lead over CI, while PVTS has topped the radar plot in comprehensive performance. Despite this, LSI continues to exhibit better performance than LRER, which again confirms the advantage of the right-side exit ramp over the left, as is the commonly accepted conclusion in the vast majority of studies. Another fact that has emerged is that multiple lane changes preceding the exit ramp still struggle to accommodate heavy traffic, and the barrier effect from trucks cannot be neglected, particularly in this scenario.

3.3. Evaluation Results

The results presented in Figure 13 enable a comprehensive evaluation of the four designs. Based on the findings, it can be concluded that PVTS and LRER may be more suitable for low LOS, whereas LSI has a wider application. However, these results might fail to provide sufficient decision support for expressway construction, as further details are not available. Thus, to address this issue, Section 2.4 of this paper undertakes a quantitative analysis, presenting a score for each design in each scenario, as shown in Figure 14.

The minimum indicator value has been scored as 0 via CE. Consequently, the final score of a design with weak performance in various aspects shows a significant difference from the best one. This hinderance makes it difficult to assess the improvement degree of optimized solution accurately. Therefore, the authors propose that the sum of the product of each indicator improvement ratio and weight would provide a better expression for the quantitative advantage of each design, as shown in Equation (14):

$$\mathrm{Im}={\displaystyle \sum _{j=1}^m\left[\left((B_j-A_j)/A_j\right)\times 100\%\times w_{k,j}\right]}$$

(14)

where B_j denotes the jth indicator of optimized design (LSI, PVTS, and LRER), A_j denotes that of CI, and Im denotes the total improvement ratio of the optimized design.

Table 4. Total improvement ratio of the LSI, PVTS, and LRER to the CI.

Improvement Ratio	LSI	PVTS	LRER
LOS-A	7.48%	11.77%	−9.08%
LOS-B	13.44%	16.21%	−6.93%
LOS-C	15.78%	21.58%	6.86%
LOS-D	12.27%	16.84%	8.65%

shows the final result.

Table 4 provides some valuable insights and conclusions. It indicates that PVTS and LSI achieve the best effect at LOS-C, but their effectiveness begins to decline subsequently, eventually leading to failure. This failure occurs at the stage when the eight-lane mainline becomes overwhelmed by the crowded traffic. In contrast, LRER produces negative side effects at LOS-A and LOS-B, while as the mainline speed decreases with the drop in LOS, it gradually catches up with the effectiveness of LSI.

3.4. Analysis

Furthermore, in engineering practice, construction cost is a crucial factor in the decision-making process.

Table 5. A rough score for the construction cost of each design.

Construction Cost	CI	LSI	PVTS	LRER
Segregated subgrade and barrier facilities	N/A	2nd	1st	3rd
Auxiliary Lane	1st	N/A	N/A	N/A
Ramp	2nd	1st	1st	1st
Score	4	2	1	3

presents the evaluation of the construction costs for each design. Since the specific cost will vary depending on various project conditions, it is challenging to provide a precise reference. Therefore, Table 5 provides a rough score of 1–4 based on the facility requirements of different designs, among which a higher value indicates a lower cost, which is more expected.

The latest CE result with construction cost taken into consideration is presented in Figure 15, which includes the weight assignment result of the entropy method for comparison to the CRITIC method. Figure 15a reveals that the entropy method focuses solely on the information and leads to a relatively even distribution of weight. On the contrary, the CRITIC method screens out the internal linkages between traffic indicators and classifies construction cost as the second category, thereby achieving a balanced view between cost and traffic performance. This also explains the rationality and scientific validity of the CRITIC method in engineering evaluation. Figure 15 and

Table 6. Final score for each design from LOS-A to LOS-D with a diverging ratio of 25% and a truck proportion of 30%.

LOS	CI	LSI	PVTS	LRER
A	35.3	28.9	19.3	16.5
B	24.8	33.5	27.7	14.0
C	24.4	26.6	25.1	24.0
D	24.9	25.2	22.0	27.9

show the final scoring results.

In summary, CI demonstrates a competitive balance between traffic performance and construction costs, particularly at LOS-A, in which the advantages of optimization have yet to be fully realized. LSI still remains the optimal choice at LOS-B and LOS-C, while LRER becomes the preferred option at LOS-D.

However, in reality, the proportion of diverging flows and trucks is constantly changing in the dynamic organic road traffic system. Therefore, this section proceeds to explore additional traffic scenarios and uses the available CE results in Figure 16 and Figure 17. The Appendix A provides specific simulation details that support CE analysis.

In actual engineering construction practice, the corresponding design scheme can be taken based on the design traffic volume, and in the meantime, the results of the traffic survey on the diverging ratio and truck proportion in the road network should also be taken into consideration.

4. Discussion

From the perspective of the methodology, despite the fact that the VISSIM micro-simulation model calibrated with traffic data basically recreates the real traffic scenario and provides accurate node evaluation, it is still difficult to support an accurate overall evaluation of the design schemes solely based on the single VISSIM results in Figure 13 and Appendix A, which can only provide a general basis for judgment. Meanwhile, the introduction of the CRITIC method improves that, referring to Table 4 and Table 6, and Figure 15, Figure 16 and Figure 17. Based on that, not only the comprehensive traffic performance of the scheme can be accurately and quantitatively evaluated, but also other factors such as the construction cost and consumption of the scheme can be taken into consideration, which is conducive to the comprehensive judgment of the sustainability of schemes and substantially improves the efficiency and accuracy of engineers’ decision-making. The comparison of the weight assignment results between the CRITIC method and the entropy method also clarifies the advantages of the CRITIC method for efficient and accurate application in the transportation field, referring to Figure 15 and its analytical description.

From the perspective of the results, according to Figure 16, the results consistently demonstrate that CI outperforms other designs at LOS-A. This can be attributed to the fact that, in smooth traffic, vehicles remain free to change lanes without excessive restrictions, resulting in an acceptable level of extra conflicts and delays associated with consecutive lane changes in diverging areas. Thus, the benefits cannot be matched by the massive renovation required in this condition. Similarly, the same reasoning applies to LSI and PVTS, although PVTS performs best in traffic indicators at LOS-A. LRER deservedly performs the worst due to its unsatisfactory transport performance, as well as maintaining this tendency up to LOS-B, and the explanation has already been discussed in Section 3.2.

LSI and PVTS become dominant below LOS-A, with the most pronounced dominance occurring between LOS-B and LOS-C. That indicates that reducing lane changes would achieve the best benefits as traffic converges and lane changes become more limited. From a transportation standpoint, PVTS makes significant sense for relieving the barrier effect from trucks at low LOS, although LSI may be the better choice when construction costs are considered. Subsequently, as the volume increases, the mainline traffic switches from stable flow to congested flow with LOS-D reached, which means the reaching of mainline capacity bottlenecks, which restrains the optimized design from accessing the desired efficiency. At this point, LRER profits from the less prominent mainline speed inferiority of the left-side exit ramp, highlighting its competitiveness due to lower construction requirements than LSI.

It appears that the diverging ratio always has a positive correlation with congestion and conflict, especially at lower LOS, but does not produce an excessive substantive impact on scheme choice. From a microscopic perspective, the success probability of a lane change depends on the time headway on the target lane. Thus, more diverging will take up more space on the curb lane and increase density, which serves as the mechanism to create more congestion and conflicts. Actually, vehicles previously on the curb lane can also change to the fast lane to achieve the desired LOS as satisfaction with the curb lane decreases, and this process will be much closer to being achieved with quite a few vehicles exiting. Thus, the impact of the diverging ratio can be considered limited, especially at the LOS that supports free flow.

Similarly, excess trucks will also exacerbate the barrier effect and impede smooth diversions, but PVTS responds to the opposite. Too few trucks contribute to oversaturation in the passenger vehicle lane and are accentuated at low LOS. Conversely, products will transfer to the truck lane, requiring an optimal proportion of approximately 30–35% to match the optimal effect of PVTS, as demonstrated in Figure 16 and the Appendix A.

In summary, LOS serves as a significant and decisive factor in design choice, governing the freedom of lane change to control the successful diversion. While diverging and truck proportion still play a certain role in altering some details. Thus, upgrading the interchange shows a significant necessity for expressways that frequently deviate from free flow, and specific options can be chosen depending on the specific performance.

5. Conclusions

This paper proposes two optimized design methods for multilane expressway exits, considering the frequently consecutive lane changes and barrier effect produced by trucks and with reference to PVTS. Extensive observations were taken for VISSIM simulation development and calibration, and the CRITIC method in CE is proposed to quantitatively score the comprehensive performance of each design scheme, covering various factors such as traffic performance and construction cost. The following useful conclusions were attempted to derive that provide references for engineering practice:

• LOS is a critical factor that determines smooth diverging on a multilane expressway.
• Reduction of lane changes can effectively improve transport efficiency in diverging areas, resulting in a 10–20% improvement ratio of LSI and PVTS at various LOS. However, left-side exit ramps have an unfavorable outcome in smooth traffic conditions, leading LRER to produce a negative impact at LOS-A and LOS-B.
• When taking construction cost and transportation quality into account, CI is sufficient at LOS-A, while LSI provides better sustainability benefits at LOS-B and LOS-C. LRER highlights strengths when the mainline remains at LOS-D throughout the term.
• PVTS delivers excellent transport performance, especially when the truck proportion is optimal at about 35%. However, its high construction costs offset its advantages.
• While diverging ratios and truck proportions can modify some details in scheme selection, LOS still dominates the process.
• The organic combination of CE and micro-simulation realizes accurate, efficient, and holistic evaluation of road design and traffic organization schemes, which has considerable application prospects with the background of sustainability objectives in the transportation field.
• For engineering assessment of road traffic disciplines, the CRITIC method possesses a distinct advantage over other weight assignment methods. It takes more internal links between indicators into account that match the characteristics of the transportation system.