Enhancing Work Zone Safety: Evaluating Static Merge Strategies Through Microscopic Traffic Simulation

2.1. Early Merge and Late Merge Simulation Studies

Several scholarly articles examined the efficacy of EM and LM control via traffic simulations. EM is a strategic approach that informs vehicles in advance of an impending lane closure in a work zone, allowing users sufficient time to identify a gap and merge successfully before the closure. EM has been observed to be particularly effective when capacity is insufficient compared to traffic demand [1]. However, the system may fail when high demand and gaps are limited [2, 3].

LM is a strategy aimed at minimizing queue length by maximizing the utilization of the entire roadway leading to a construction zone [4]. This involves directing vehicles to utilize all available lanes, and upon reaching the merge point, they must adhere to a “take turn” strategy [1]. The “zipper” strategy, as outlined by Delft University in the Netherlands, requires each driver to maintain a specific distance from the lane drop, directly behind the follower of the driver who initiated the maneuver. When implemented accurately, the LM system has the potential to significantly improve throughput and reduce queue length by as much as 50% [4, 5].

In a study by Venkatesan et al. [6], an approach was proposed to optimize traffic flow using a FORTRAN simulation model. This model determined the optimal merging strategy for vehicles navigating a construction zone with two lanes instead of three. According to the results of their simulations, the most effective merging method for reducing travel time did not involve a significant number of early mergers. Conversely, the LM practice significantly reduced trip time as traffic volume increased.

Beacher et al. [7] employed the microscopic highway simulation model WZSIM to simulate static EM and LM control strategies in a scenario involving a reduction from two to one lane. Their simulations yielded a significant result: providing vehicles with notice of the lane closure led to reduced delays and enhanced productivity within the work zone. This finding has practical implications, as it suggests that early merging, by offering drivers more opportunities to transition into the available lane, can effectively mitigate delays. This, in turn, can lead to increased speed and density in the unobstructed lane, thereby enhancing overall productivity. Comparative analysis between the LM approach and a no-control scenario revealed that only under substantial traffic demand did the LM strategy outperform the no-control scenario. Their simulation analysis, nevertheless, failed to incorporate a comparison between static EM and LM control strategies.

Chatterjee et al. [8] assessed the correlation between work zone capacity values and VISSIM driving behavior parameters in a conventional EM system. In evaluating the impact of lane closure scenarios ranging from two to one with two lanes closed it was determined that the VISSIM parameters CC1, CC2, and SDRF had the most significant influence on work zone throughput.

2.2. Car-following Model

Wiedemann's driver behavior model forms the cornerstone of the VISSIM car-following model. The fundamental concept underlying the car-following model suggests that as a vehicle approaches a slower vehicle in front, the approaching driver will decelerate to preserve a designated safe following distance [9]. Nevertheless, due to the driver's inability to precisely gauge the speed difference with the leading vehicle, an excessive decele- ration may occur, resulting in a reduction in velocity relative to the leading vehicle. Consequently, the driver will continue accelerating to uphold the safety distance [10].

Consequently, the car-following process in VISSIM is characterized by a combination of iterative deceleration and acceleration caused by vehicles whose perceptions of the desired speed, speed difference, and safety distance fluctuate, as well as the unique attributes of each driver and vehicle [10]. To simulate freeway traffic in VISSIM, utilizing the Wiedemann 99 model is advisable. The user can modify ten parameters in this model; these parameters are displayed in VISSIM in Fig. (1) and are briefly described in Table 1.

Due to the requirement that drivers follow other vehicles at a safe distance (i.e., safety distance), safety distance-related parameters (CC0, CC1, CC2, CC4, and CC5) will have a substantial effect on road capacity during periods of heavy traffic demand [11].

2.3. Lane-changing Model

Lane changes are categorized into two distinct categories in VISSIM: necessary and free lane changes. In addition to making necessary lane changes to adhere to a predetermined path (e.g., due to an intersection or lane closure), drivers make free lane changes to achieve the desired speed or take advantage of improved driving conditions in the adjacent lanes. Predetermined gap acceptance criteria determine lane change decisions. These criteria comprise the following: the availability of the gap in the opposite lane, the wish to change lanes, and improved driving conditions [11].

Fig. (2) illustrates the vehicle-specific variables that impact “necessary lane change” behavior: maximum deceleration, maximum deceleration reduction rate, and accepted deceleration. These variables are represented in VISSIM. These variables influence the aggressiveness of the lane-changing driver and the following vehicle. Each driver will begin by decelerating at their accepted deceleration rates. Nevertheless, if the driver changing lanes cannot do so within a specified distance before the final necessary lane change position, they will progre- ssively quicken their deceleration until they achieve their maximum deceleration value. The driver will stop and wait for a potential lane change gap if he cannot change lanes at the final required lane change position. The vehicle will be removed from the network, and an error report will be generated if the stopped lane-changing driver cannot change lanes within the time limit specified by the parameter waiting time before diffusion [11].

Fig. (2) illustrates the additional parameters that influence all lane changes. In this analysis, the adjustment was focused solely on the safety distance, as outlined in the car-following model, which impacts the level of aggression during free lane changes [12]. This adjustment is accomplished by regulating the Safety Distance Reduction Factor (SDRF). SDRF decreases the safety distance defined by the car-following model for both the leading and trailing vehicles until the lane change maneuver is complete [12]. Consequently, the driver executing the lane change becomes more aggressive as SDRF increases.

This study presents findings from a thorough EM and LM strategy analysis using microscopic traffic simulations. The study encompasses meticulous computer simulations of static EM and LM controls in a scenario involving a reduction from two lanes to one, alongside a compre- hensive examination of traffic performance within the work zone area, focusing on metrics such as queue length, delay, and total travel time. Notably, safety considerations are not explicitly addressed in this analysis but will be in subsequent studies. The principal objectives of this study are threefold: firstly, to gain a comprehensive under- standing of the functionality and advantages of EM and LM control through traffic simulations; secondly, to identify simulation parameters that exert influence on the performance of these control systems; and thirdly, to furnish recommendations on the optimal conditions for implementing EM or LM strategies based on simulation outcomes.

The paper is structured as follows: the Methods section introduces the methodology utilized; the Simulation Results and Discussion section analyzes the results; and the Conclusion section summarizes the significant findings of the study.

3.1. Parameter Selection

The consideration of this study was directed solely toward the parameters deemed to exert the most significant influence on work zone performance, as identified in prior literature. Previous studies [13]– [16] have collectively established that the parameters CC0, CC1, CC2, CC4, and CC5 within the car-following model, alongside SDRF within the lane-changing model, can substantially impact the estimated capacity of freeway construction zone areas.

The computation of the minimum safety distance is derived as {dx_safe distance, min = CC0 + CC1 × FFS}. Consequently, CC1 exerts a more pronounced effect on the minimum safe distance, mainly when vehicles operate at free-flow speed. Additionally, findings from a study conducted by Chitturi and Benekohal [17] indicate that for CC1 values exceeding 0.8, the contribution of CC0 to capacity becomes negligible. Consequently, CC0 was disregarded in this study, and the minimum value for CC1 was established at 0.9 seconds.

As explained in the literature review subsection, Chatterjee et al. [8] concluded that the VISSIM parameters CC1, CC2, and SDRF were of the utmost impor- tance regarding work zone throughput. Addi- tionally, it was determined that the absolute values of parameters CC4 and CC5 had no statistically significant effect on capacity when they remained below 3.0. Their visual analysis of the simulations determined that an absolute value greater than 3.0 for CC4 and CC5 may result in an incorrect car-following procedure. As a result, the values CC4 and CC5 were excluded from the parameter set utilized in the present study.

Following this, the significant parameters CC1, CC2, and SDRF were chosen for additional analysis. Table 2 specifies the possible values for the three input parameters. During the analyses, the preset values were maintained for all other parameters.

By simulating eight scenarios using the data in Table 2, it is possible to determine which parameter or parameters impact the work zone's throughput under varying traffic demands and the two merge control schemes. As a result, recommendations regarding the optimal conditions under which the EM or LM should be contemplated can be generated based on the simulation outcomes. Table 3 details every conceivable scenario.

Table 3.

Scenario ID	Input Parameter Values
1 2 3 4 5 6 7 8	{0.9, 13.12, 0.6} (More aggressive driver) {0.9, 13.12, 0.2}
	{0.9, 20, 0.6} {0.9, 20, 0.2} {2, 13.12, 0.6} {2, 13.12, 0.2}
	{2, 20, 0.6} {2, 20, 0.2} (Less aggressive driver)

3.2. Building VISSIM Network

Coding involves the building of the simulation network. The VISSIM model was utilized to evaluate the potential benefits of employing static EM and LM controls in a scenario involving a reduction from two lanes to one within a work zone area. Furthermore, the model aimed to identify the optimal conditions for implementing each control strategy. The configuration of the EM and LM control schemes is depicted in Fig. (3), while their respective VISSIM designs are illustrated in Figs. (4a and 4b).

In the EM control scheme, vehicles are directed to merge into the open lane within 0.5 miles upstream of the lane closure. Conversely, in the open lane, vehicles are prohibited from using the closed lane for overtaking until they have passed the work zone area. In the LM scenario, the necessary lane change occurs 200 feet upstream of the work zone; before this point, vehicles are free to utilize both lanes for travel and overtaking. The designated speed limit is 75 mph on the freeway and 55 mph in the work zone. The traffic composition consists of 80% passenger cars and 20% heavy vehicles. The Wiedemann 99 car-following model was employed to replicate freeway driving behavior. Each simulation was executed for a duration of 3600 seconds.

To thoroughly assess the performance of EM and LM controls under varying traffic demand conditions, the vehicle inputs in the simulation were set to range from 2,000 vehicles per hour (veh/h) to 2400 veh/h at increments of 100 veh/h for the freeway segment, and from 270 veh/h to 410 veh/h randomly assumed for the on-ramp segment. These volume inputs are distributed at 5-minute intervals from 0 to 3600 seconds, as depicted in Fig. (5).

3.2.1. Vehicle Travel Time and Delay Measurement

A vehicle travel time measurement encompasses both a “From” and “To” Section. It calculates the mean travel time from traversing the “From” Section to traversing the “To” Section, including any waiting and stop times encountered across all lanes [11]. In our network, three travel time sections were defined, as depicted in Fig. (6a). One was designated for the work zone, while the other two were allocated for the merging areas, as shown in Fig. (6b). On the other hand, delay measurements capture various metrics, including the number of vehicles, average delay, average stop delay, and the number of stops of a turn relation. These measurements may consist of one or several travel time measurements.

4.1. Vehicle Network Performance Evaluation Results

Table 4 presents the results of the network performance evaluation for both merge controls. A comparative analysis of simulation results is conducted for each scenario input to identify the most influential parameters. Specifically, comparing scenarios (1 with 2), (3 with 4), (5 with 6), and (7 with 8) enables assessment of the impact of the Safety Distance Reduction Factor (SDRF) on network performance, as depicted in Table 5a. Similarly, comparing scenarios (1,2 with 3,4) and (5,6 with 7,8) allows for examination of the influence of the CC2 parameter on network performance, as shown in Table 5b. Furthermore, comparing scenarios (1,2,3,4 with 5,6,7,8) helps determine whether CC1 significantly affects network performance, as illustrated in Table 5c. Fig. (8) illustrates the comparison tree for these scenarios. Tables 5a, b, and c indicate that CC1 and CC2 substantially impact the network performance compared to the SDRF parameter.

The comparative analysis highlights the critical role of parameters CC1 and CC2 in simulations utilizing both EM and LM controls. Increasing values of CC1 and CC2 result in decreased work zone throughput in both EM and LM scenarios, as indicated in Tables 5b and 5c. Additionally, the analysis reveals that the Safety Distance Reduction Factor (SDRF) has a relatively minor impact on network performance for both merge controls, as indicated in Table 5a.

4.2. Average Queue Results for all Scenarios

The EM scenario exhibits a significantly shorter average queue length than the LM scenario. This indicates that the queue in the EM scenario experiences slower growth than the LM scenario. Detailed queue results for both merge strategies are presented in Table 6.

When drivers exhibit aggression and maintain minimal safety distances between vehicles (i.e., low CC1 values), opting for EM control yields advantages such as shorter queue lengths and lower average delay time per vehicle in comparison to LM control. Conversely, implementing EM control is not advisable when drivers are cautious and maintain significant distances from leading vehicles (i.e., high CC1 values). This is because, with LM control, queues tend to grow slower, resulting in shorter average delay times per vehicle than EM control.

5.1. Study Limitations and Future Work

While this study offers significant insights into the performance of the merge control strategies within a freeway work zone, it also presents several limitations that merit attention in future research. Our simulations are primarily grounded in conventional vehicles, overlooking the presence or impact of Connected and Autonomous Vehicles (CAVs). The introduction of CAVs, with their advanced communication capabilities and adaptive behaviors, could potentially reshape traffic dynamics, thereby potentially reducing the applicability of our findings to future traffic scenarios.

Another limitation is the reliance on specific VISSIM parameters (CC1 and CC2) to assess the impact on work zone throughput. While these parameters are pivotal, other factors, such as vehicle heterogeneity, driver behavior variability, and environmental conditions, were not extensively investigated. It is imperative to incorporate a wider range of parameters and scenarios in future research to foster a more comprehensive understanding of the effectiveness of merge control techniques. Furthermore, our study concentrated on a specific work zone configuration (2-to-1 lane closure). Diverse configurations, such as multi-lane closures or varying work zone lengths, could potentially yield different outcomes. Future studies should encompass these variations to augment the generalizability of the findings.

To build upon this study's findings, future research should explore the impact of CAVs on merge control techniques in work zones. CAVs can communicate with each other and the infrastructure, leading to more coordinated and efficient merging strategies. Investigating how these vehicles interact with human-driven vehicles under different merge control schemes could provide insights into optimizing traffic flow and safety in mixed-traffic environments.

Moreover, developing and simulating new merge control algorithms designed explicitly for CAVs should be considered. These algorithms could leverage real-time data and predictive analytics to adjust merging strategies dynamically based on current traffic conditions. Another avenue for future work is to conduct field experiments to validate the simulation results. Real-world data can help refine simulation models and ensure the findings apply in practical scenarios. Lastly, considering the varying driver behavior across different regions and cultures, future studies should include a diverse range of driver profiles to better understand the global applicability of the merge control techniques. Extending this research to include another case study dealing with CAVs would be highly beneficial. Such an extension could involve dynamic control models like the lane-based signal merge (LBSM) system, which uses lane-based signals or variable signs to optimize merge operations in work zones. The relevance of this approach is underscored by studies like those by Fan et al. [18], Hussain et al. [19], Nassef et al. [20], and Yuan et al. [21], which offer valuable insights into developing merge strategies using deep reinforcement learning and cooperative lane merging for CAVs.

By addressing and exploring these limitations and future research directions, we can advance our understanding of work zone traffic management and develop more robust and adaptable solutions for the evolving transportation landscape.