The field of automated vehicles gains more and more attention each year as the effective implementation of such vehicles in real life is imminent. The increasing amount of scientific studies addressing the impact of automated vehicles on the traffic network shows the same trend. While much research focuses on regular intersections, only a few investigate the impact of turbo-roundabouts. In particular, studies researching the obtainable capacity increase on turbo-roundabouts under fully automated conditions are scarce. A turbo-roundabout is a multi-lane roundabout with separated lanes that require vehicles to choose their desired destination before entering the turbo-roundabout, effectively eliminating most weaving conflicts on the roundabout itself. Only a single paper researches how much additional merging capacity can be obtained by automated vehicles on a turbo-roundabout compared to human-driven vehicles (Fortuijn and Salomons, 2020). This study takes a theoretical approach and uses existing theory about turbo-roundabout capacity to generate five graphs depicting the merging capacity of the left lane of a minor access branch for varying circulating traffic volumes on the turbo-roundabout: four models for differing automated vehicle settings and one human-driven vehicle model. These four automated vehicle models differ in complexity, where the most complex model features gap synchronisation, headway optimisation, and path guidance. Each model, therefore, has different following distances, vehicle speeds, critical gaps, etc. The results of the theoretical models are capacity curves depicting the merging traffic capacity over the circulating traffic volumes. This thesis investigates whether the capacity increase obtained from the most simplified automated vehicle model of the theoretical study can be obtained through microscopic traffic simulation in VISSIM. What must be noted is that the most simplified model of this theoretical study features communication between vehicles, which is not the case in the simulation models of this thesis. Therefore, higher follow-up times must be maintained in the simulation models, whereas the theoretical model uses minimal follow-up times to obtain a capacity increase of 58%. Initially, this thesis aimed to determine the achievable capacity increase for Connected and Automated Vehicles. However, due to limitations in VISSIM, the capacity increase could only be obtained for Automated Vehicles as the software cannot model communication between vehicles. Therefore, this report does not obtain the highest potential capacity increase for automated vehicles but serves as a stepping stone for more complex AV models. It determines the additional merging capacity of automated vehicles on turbo-roundabouts, only considering key parameters such as the car-following behaviour, critical gap and follow-up times. Gap synchronisation, an important aspect of the study of Fortuijn and Salomons, is not included in the simulation models due to shortcomings in VISSIM. Before making the AV models, this thesis aims to calibrate the VISSIM model for human drivers with real traffic data at a turbo-roundabout in Nieuwerkerk aan den IJssel, the Netherlands. Only when the VISSIM model is capable of accurately modelling human behaviour can it be used to determine the capacity increase through AVs. The thesis starts with developing a human-driven vehicle model to determine the merging capacity of the turbo-roundabout in current traffic conditions. The calibration is conducted on the traffic flow level through an iterative process where the critical gap and follow-up times are adjusted. The root-mean-normalised-squared error (RMNSE) determines the difference between regression lines of the merging capacities of the simulation model and the real traffic measurements. Based on the calculated differences between both lines, the critical gap and follow-up time parameters are adjusted. For the left lane, the iteration process reduced the RMNSE from 0.427 to 0.072, equaling an improvement of 72.5%. For the right lane, the iteration procedure gave an unsatisfactory result. The RMNSE increased from 0.408 to 0.427, which signifies an increase of 4.7%. The final capacity curve after seven iterations was unrealistic. It was concluded that the VISSIM model performs adequately at simulating realistic behaviour on the left lane but that the same accuracy could not be achieved for the right lane. The calibration procedure is only applied to the minor branch of the turbo-roundabout as the data for the major branch could not be used for calibration. Subsequently, the calibrated model serves as the basis for developing three automated vehicle models: cautious, normal, and aggressive, each representing a different aggression level of automated vehicles. Each of these models features a different car-following model to simulate the various levels of aggression. Furthermore, the critical gap times of the human model are changed to model Avs properly, and for each of the three models, a penetration rate of 100% of AVs is applied. For each of the three automated vehicle models, this thesis measures the increase in capacity. It compares it to the theoretically obtained capacity increase of 58% for the left lane of the minor access branch of the turbo-roundabout. This thesis finds capacity increases of 7.47%, 38.62%, and 24.32% for the cautious, normal, and aggressive automated vehicle models, respectively. For the right lane, the obtained capacity increases were not calculated as the final HDV model is inaccurate for this lane. An unexpected result is that the normal automated vehicle model outperforms the aggressive model for the left lane, even though the parameter settings should result in a higher capacity for the aggressive model. Comparing the resulting capacity increases to the 58% increase of Fortuijn and Salomons, one can conclude that the simulation model does not reach the same increase in capacity as the theoretical model. This is not an unexpected result, as a theoretical simulation model has fewer interactions, and the theoretical model assumes communication between vehicles, whereas the simulation models do not. Therefore, concluding that the microscopic traffic simulation models in VISSIM cannot accurately model the behaviour of automated vehicles is premature. The models leave room for improvement by fine-tuning various parameters that were ignored in this thesis. This thesis concludes that, while VISSIM is adequate at simulating human-driven vehicles, it has various limitations for modelling automated vehicles. First of all, gap synchronisation and headway optimisation cannot be modelled, while these are the most critical elements to unlock the full potential of automated vehicles on turbo-roundabouts. If one wants to incorporate these into VISSIM, an external mathematical software package is necessary. Considering other software packages for studying (C)AVs on turbo-roundabouts is advised. This thesis is a foundation for future research on turbo-roundabouts, both for the calibration process and for simulating automated vehicles.

With the steadily increasing road traffic demand, congestion on freeways has become a major problem. With the development of communication technology and automated vehicles, there are more opportunities for DTMs to be studied extensively at the coordination level and cooperation level. In the existing literature, the coordination of RM (ramp metering) + VSL (variable speed limit) + RG(route guidance) has rarely been studied. Besides, some coordination studies have limited discussion on the interaction between DTMs in non-coordinated cases. Based on these research gaps, the research question of this thesis is: What is the impact of coordinating RM, VSL and RG on a road structure where they have potential counter-effects on each other when following local objectives? This research aims to explore the network performance on the non-coordination cases and coordinated cases. The simulation is built in SUMO, while the DTM control measures are implemented via the Traffic Control Interface (TraCI). In the coordination, all controller parameters (metered flows, speed limits, and split rate) work together to improve traffic in this two-bottleneck system. The coordination strategy has a model predictive control structure. The cell transmission model (CTM) is used as the prediction model. In this case, all the effects of considered DTM measures can be integrated into the expression of the inflow to the first cell on the basis of certain assumptions. Considering that there are 5 control variables, genetic algorithm (GA) is used for the optimisation process in the MPC control. The objective is maximising the total outflow after bottlenecks on two routes. The conclusion of the research is that there is benefit for applying DTM coordination on such a road structure with two parallel freeways. However, the improvement caused by a predictive RG is the major part of the benefit. This is because that the simulation cannot capture the capacity drop phenomenon. The simulation results indicate that a predictive RG is more effective compared to coordinated control. It also leads to the best travel time equity among routes than the coordinated control. Based on the findings, it is more prior for this kind of road structure to improve the in-vehicle RG on this road structure than to apply coordinated control.

Traffic congestion is a challenge that frequently emerges due to changes in the roadways such as tunnels and sags, causing capacity reduction. The capacity drop phenomenon exacerbates traffic congestion, due to decreased queue discharge rates. Among the strategies employed for traffic management, Variable Speed Limit (VSL) control is a common approach to alleviate congestion and mitigate capacity drop. The control is expected to be more powerful when integrated with Connected Vehicles (CVs). However, the intricate interplay between the Market Penetration Rate (MPR) of CVs and the parameters governing VSL remains under-explored. This study seeks to quantify the relationship between the MPR of CVs and the key VSL parameters, encompassing factors like the optimal acceleration length and speed limits. The VSL control is applied to CVs at the road section upstream of a tunnel bottleneck modelled by a continuum car-following model with bounded acceleration. The findings of this research suggest a minimum MPR of CVs essential for preventing capacity drop and reaching the maximal outflow. Intriguingly, this challenges the conventional notion that regulating solely the leading vehicle suffices to govern the behaviours of all following vehicles. The value of the minimum MPR threshold is affected by the acceleration behaviours vehicles exhibit. Furthermore, this study underscores the importance of considering the MPR of CVs when devising the optimal speed limit and acceleration length for the VSL. It shows that while a higher speed limit can lead to higher throughput, the optimal acceleration length increases exponentially with the increasing speed limit, particularly in cases with low MPR of CVs. While adopting a relatively lower speed limit, the acceleration length can be reduced to 0m for all levels of MPR of CVs. In summary, it suggests that when implementing VSL in practice, a balance between the resulted throughput, the required acceleration length and the robustness of VSL across different levels of MPR of CVs should be considered.

This thesis aims to answer the question "What are the properties of a resilient road network?" The resilience of a road network indicates the magnitude and consequences of a disruption, which can be events such as a traffic accident or a flood. The literature is not in agreement about the exact definition of road network resilience, and the way to quantify it. A literature search is done to look for definitions and resilience metrics. Eleven metrics are found that quantify resilience in road networks to accidents. These eleven metrics are compared based on whether they are on a network level, and how many variables are used. Two metrics are chosen, one based on travel time, and one based on space-mean flow. Another metric, based on the outflow of the network is added. The three chosen metrics are then compared in small networks to see if they give the same results. The small network consist of five nodes with different link configurations. The conclusion about which network is the most resilient changes with the resilience metrics. The metric based on travel time is deemed the best for this type of research. The metric based on network outflow can have distorted results due to high peaks, and the metric based on space-mean flow is better suited for a different experiment design. To see how network parameters such as link density influence the resilience, a simulation with random networks is done. 266 networks with nine nodes in random locations, and random links between them are simulated. It is concluded that networks with a lower link density have a higher resilience. This is not only illustrated by the relation between resilience and link density, but also by several other network parameters which are related to link density such as average number of lanes and connectivity. A reason that these networks are more resilient could be that there is less spillback, the links in the high density networks are longer and have more lanes, so congestion will not spread to other links as fast.