Despite advanced communication, monitoring, and control facilities, train operations are still subject to uncertainties that can disturb train services, cause delay to multiple trains, and propagate through the network. One option is to mitigate delay propagation in the timetable design by adding buffer time to the minimum difference between the time two successive train of either direction enter a section. It is still common practice to design buffer times based on a deterministic value, decreasing operational capacity and requiring large amount of manual checking by planners. Existing approaches to effectively allocate buffer time in timetables lack flexibility and require an initial timetable. In this paper, a data-driven approach for determination of buffer time planning rules suitable for usage in an initial timetable is presented. These planning rules are not necessarily generic, but rather depend on timetable characteristic. Two metrics that describe delay propagation, mean secondary delay and hindrance percentage, are extracted from literature and predicted in a regression analysis with the use of timetable characteristics related to headway situations of two succeeding trains. The results of the regression analysis on a case study of the Dutch railway network between Haarlem, Leiden Centraal and Schiphol Airport are used to determine the amount of scheduled buffer time that would ensure a certain amount of hindrance percentage given a specific headway situation. The results show that the mean secondary delay and hindrance percentage for various headway situations can both be predicted with an accuracy of 90.7\% based on timetable characteristics and is quite heterogeneous. Mean secondary delay appeared not significantly impacted by the scheduled buffer time, contrary to hindrance percentage which is significantly influenced by the scheduled buffer time.

As the demand for the railways is expected to raise in the future, researchers are looking for ways to improve the railway capacity to increase the transport ability. Virtual coupling is a new solution based on the moving block technology, which further shortens train separation from absolute braking distance to relative braking distance. This will also affect the train service schedules for optimal operation under virtual coupling. In this study, a mixed integer quadratic programming method has been proposed to schedule the train services under virtual coupling on a network. Train operation as well as the headway and capacity benefit have been analyzed. In comparison to moving block, virtual coupling will change train operation on shared routes, and the capacity will be improved by different percentages for different timetable patterns and different service braking rates with or without speed limit restrictions.

With the increasing demand for public transport systems worldwide and also a lot of these systems running at their maximum capacity, there is a strong need for finding ways for these systems to operate in a more efficient way. In recent transportation research there is an increasing attention for operational conditions and the impact of passengervehicles interaction on the timetables of urban rail networks. Passengervehicle interaction can have a strong impact on the operational conditions of an urban rail line as a large part of the dwell time of a vehicle can be explained by the number of boarding and alighting passengers.

In busy passenger railway networks, a large amount of trains have to be parked in shunting yards off the mainline every night, where they will be cleaned, maintained, sorted and parked. This problem is known as the Train Unit Shunting Problem (TUSP), which is a hard combinatorial optimization problem faced by railway operators. The TUSP is currently solved using human heuristics, which is difficult and time-consuming. Reinforcement learning approaches have been developed in the last few years to efficiently approach this problem. In this research we develop, in a multi-agent deep reinforcement learning framework, an heuristic for random exploration to efficiently search the state-action space and two heuristics for train routing strategies, which aim at improving the performance and quality of the produced route plans. On one hand, we develop the Type-Based Routing Strategy, based on the idea of parking trains of the same rolling stock type on the same tracks. On the other hand, we develop the In-Residence Time Routing Strategy, based on parking trains ordered according to their departure time. Both routing strategies consists of four components: (1) standard parking rules, (2) combination and split rules, (3) conflict resolution rules and (4) unnecessary movements rules. The goal is to incorporate information from the logistics side of the problem into the framework. We demonstrated the performance of the heuristics developed in two real-life cases in the Dutch railway network. The results of the experiments carried out demonstrate the potential of the resulting framework to produce more efficient route plans and to adapt to different problems designs such as different matching problems types and different shunting yards.

Disruptions occur frequently in railway networks, requiring adjustments to the timetable, rolling stock planning and crew planning while causing delays and cancellations. Although the evolu­tion of system performance during a disruption can be visualized in the resilience curve, not much is known about performance during disruptions or the extent to which the curve applies in practice. The limited quantitative knowledge about the resilience of railway networks makes it hard to design appropriate recovery measures. In this thesis, a data-driven evaluation approach is presented to make an ex post assessment of the resilience of railway networks. Several resilience metrics are extracted from literature and two new resilience metrics are introduced. Using historical traffic realization data, resilience curves are reconstructed for a large and heterogeneous set of single disruptions and are quantified in terms of the resilience metrics. Among others, the values of the resilience metrics are compared across disruptions of different causes using Welch’s ANOVA and the Games-Howell test. The approach is applied to a case study of the Dutch railway network, with a focus on the five most common disruption causes. The results of the case study show that there is significant heterogeneity in the shape of the resilience curve, even within disruptions of the same cause. Train defects are found to be the least impactful disruptions on multiple resilience metrics, while collisions are found to be the most impactful disruptions on multiple resilience metrics. The successful application of the approach shows that it can be used by practitioners to assess which types and which parts of disruptions deserve attention to improve disruption management practices, and thus, improve resilience.

The current research addresses a problem found in the area of railway operations regarding the maintenance of rolling stock units. It focuses on the situation in The Netherlands and approaches the problem from the perspective of its main railway operator N.V. Nederlandse Spoorwegen (NS).
The increasing use of the capacity of the railway network leads to two issues. First, the complexity of the scheduling process is increasing, raising the need for tools that automate this process. Second, since NS is considering to perform more maintenance activities during daytime, raising the question at which locations maintenance teams needs to be stationed to perform daytime maintenance. These issues are interrelated.
The model development in the current research, tackling the aforementioned issues, can be understood as a three-stage framework. Assuming a given rolling stock circulation, the first stage aims to find the maintenance schedule and maintenance location choice minimizing the total number of nighttime maintenance activities. The second stage introduces a model to compute the required capacity. The third stage integrates the first and second stage, aiming to find a solution to the first-stage model that satisfies some predetermined maintenance location capacity constraints that can be determined by the second-stage model.
First, it is shown that, for a scenario with 20 maintenance locations for daytime maintenance, up to 42.0% of the work can be performed during daytime. Also, the second model can be used to efficiently (i.e. within seconds) compute required capacity and an accurate maintenance activity planning. Moreover, results for the third model have been generated showing, for example, that in one considered problem instance, the number of maintenance shifts for which the required capacity exceeds the available capacity can be reduced from 21 to 5 in 7.6 minutes. In addition to the aforementioned experimental results, a more practical approach is taken as well by constructing a small use case, demonstrating how the current research can be applied in practical situations. To this end, planning software Viriato is used, by which various visualizations of maintenance schedules can be provided.

Unreliability is a major source of discontent for passengers using public transport. Real-time rescheduling is one of the tools available to deal with it. Although passenger movements have an impact on the development of a disturbance in public transport operations, this effect is rarely accounted for in rescheduling models for metro networks. The aim of this thesis is to develop a rescheduling framework for metro networks that mitigate the impact of disturbances on passengers whilst considering the dynamic variation of demand and its effect on train operations. To achieve this, the Simulation-Based Traffic Management for Metro Network (SBTM-MN) framework is developed. This comprises two models that interact iteratively, the Transport Simulation Model (TSM), and the Train Rescheduling Model (TRM). The TSM integrates a train simulation in the microscopic train simulation tool OpenTrack, with a passenger module constructed to keep track of passenger allocation and compute train dwell times according to the passenger exchange. The TRM modifies arrival and departure times of trains according to a base schedule and the corresponding passenger allocation provided by the TSM. A series of experiments are carried out on the SBTM-MN, and on each of its components, the TSM and the TRM. These experiments provide insights into the effect of passenger allocation on the rescheduling measure, the trade-off between passenger and operators perspective, and the impact of demand on the development of disturbances. The case study consisted of lines D and E of the Rotterdam metro network, with northerly direction.

This MSc thesis research has the objective to empirically determine the perceived stochastic nature of the deceleration regimes and to determine the impact it has on infrastructure occupation within the network corridor. This research expands on the data-driven reconstruction model to estimate the speed profiles of realised train runs developed by N. Besinovic et al., calling it the 'Deceleraton Reconstruction' (DR) model. This model elaborates on the deceleration regimes of the speed profiles in a more dynamic and generalised manner to provide a more detailed description of the realised deceleration behaviour, through introducing the concept of sub-regimes to describe the deceleration regimes and the concept of a non-uniform braking rate in the braking regimes. This research expands on the number of data sources used for location tracking of the realised train runs and introduces the concept of 'Data Fusion' in which the different sources of location tracking are combined in order to provide a more reliable and more detailed input for the DR model. The results of the empirical and comparative analysis, provide interesting insights to the correlations between the departure delays, realised running times and deceleration loss time performance, and provides insights to the impact the deceleration behaviour has on infrastructure occupation. The DR model provides interesting results regarding the quality and detail of the estimated deceleration behaviour (e.g. braking rate, deceleration regime profile) and compares these in relation to the nominal characteristics used in timetabling tools and simulation models.

Passengers often complain about dirty trains indicating the relevance of interior cleaning of rolling stock (RS). Servicing tasks (i.e. interior and exterior cleaning and smaller technical checks) are executed on a daily basis at service locations (SLs). Currently, due to train operations during daytime, the current focus lies on night servicing. In this thesis daytime servicing is considered in order to tackle the capacity shortages at SLs. Therefore, the Rolling Stock Servicing Scheduling Problem (RS-SSP) is developed comprising a Mixed Integer Linear Programming (MILP) model. By complying with the planned timetable, the RS-SSP maximises the RS units being serviced during daytime. The RS-SSP allows RS exchanges between RS units having completed servicing and operating RS units requiring servicing. Due to this RS Exchange Concept, the number of RS units visiting the SL during daytime can be increased. Within the thesis three RS-SSP model versions have been developed: the RS-SSP Base Model and two model extensions. The RS-SSP Base Model considers trains running with a single RS unit per train and RS units to be immediately serviced when entering the SL. The first extension (RS-SSP-MU) considers multiple unit trains and the second extension (RS-SSP-MU-W) allows RS units to wait for servicing. The proposed RS-SSP models have been tested on a real-life case from the Dutch railways. The RS-SSP-MU-W yielded the most feasible and improved solutions as compared to the other two model variants. For multiple scenarios, the model was able to exchange all running RS. As a conclusion, the capacity usage at SLs can be increased by the RS-SSP by shifting the excessive workload to daytime, and thus solving the capacity shortages. As the RS-SSP model is a generic model, it may not only be applied to other railway operators, but also to other public transport companies. Further extensions on the model are suggested for an appropriate applicability on large scale.

Railway maintenance is currently planned according to a fixed schedule, mostly of one year. This is usually not optimal when considering the cost of maintenance, of possessions and of failure. However, an approach that minimizes the sum of these costs was not yet available. This paper presents a statistical analysis method and an optimization problem that aims to solve this problem. It is applied to fictitious problems and a realistic test case.

In a busy railway network such as the Netherlands, more and more maintenance activities are needed to be performed. These planned activities often lead to an infeasible timetable since infrastructure is temporary unavailable for operations. A macroscopic network model can roughly adjust the timetable for a complete network, while a microscopic model is needed to check for possessions in station areas and ensure feasibility of the timetable. In this thesis a microscopic routing model is proposed that adjusts the route plan in a station area while minimising passenger dissatisfaction. The model also finds new rolling stock connections, implements shunting movements in a station area and considers shorter rolling stock formations. To increase the quality of the route plan, its robustness is increased iteratively and small time shifts of arrivals and departures are applied to resolve small conflicts that would lead to cancellations. Several case studies demonstrate the high performance of the model. Finally, a feedback mechanism to a macroscopic model is proposed.