The design and optimization of public transport systems is a highly complex and challenging process. Here, we focus on the trade-off between two criteria which shall make the transport system attractive for passengers: their travel time and the robustness of the system. The latter is time-consuming to evaluate. A passenger-based evaluation of robustness requires a performance simulation with respect to a large number of possible delay scenarios, making this step computationally very expensive. For optimizing the robustness, we hence apply a machine-learned oracle from previous work which approximates the robustness of a public transport system. We apply this oracle to bi-criteria optimization of integrated public transport planning (timetabling and vehicle scheduling) in two ways: First, we explore a local search based framework studying several variants of neighborhoods. Second, we evaluate a genetic algorithm. Computational experiments with artificial and close to real-word benchmark datasets yield promising results. In all cases, an existing pool of solutions (i.e., public transport plans) can be significantly improved by finding a number of new non-dominated solutions, providing better and different trade-offs between robustness and travel time.

Due to the significant growth in passenger numbers, higher vehicle load factors and crowding become more and more of an issue in public transport. For safety reasons and because of an unsatisfactory discomfort, standing of passengers is rather limited in high-speed long-distance trains. In case of delays and (partially) cancelled trains, many passengers have to be rerouted. State-of-the-art rerouting merely focuses on minimizing delay at the destination of affected passengers but neglects limited vehicle capacities and crowding. Not considering capacities allows using highly efficient shortest path algorithms like RAPTOR or the connection scan algorithm (CSA). In this paper, we study the more complicated scenario where passengers compete for scarce capacities. This can be modeled as a piece-wise linear, convex cost multi-source multi-commodity unsplittable flow problem where each passenger group which has to be rerouted corresponds to a commodity. We compare a path-based integer linear programming (ILP) model with a heuristic greedy approach. In experiments with instances from German long-distance train traffic, we quantify the importance of considering vehicle capacities in case of train cancellations. We observe a tradeoff: The ILP approach slightly outperforms the greedy approach and both are much better than capacity unaware rerouting in quality, while the greedy algorithm runs more than three times faster.

Providing attractive and efficient public transport services is of crucial importance due to higher demands for mobility and the need to reduce air pollution and to save energy. The classical planning process in public transport tries to achieve a reasonable compromise between service quality for passengers and operating costs. Service quality mostly considers quantities like average travel time and number of transfers. Since daily public transport inevitably suffers from delays caused by random disturbances and disruptions, robustness also plays a crucial role. While there are recent attempts to achieve delay-resistant timetables, comparably little work has been done to systematically assess and to compare the robustness of transport plans from a passenger point of view. We here provide a general and flexible framework for evaluating public transport plans (lines, timetables, and vehicle schedules) in various ways. It enables planners to explore several trade-offs between operating costs, service quality (average perceived travel time of passengers), and robustness against delays. For such an assessment we develop several passenger-oriented robustness tests which can be instantiated with parameterized delay scenarios. Important features of our framework include detailed passenger flow models, delay propagation schemes and disposition strategies, rerouting strategies as well as vehicle capacities. To demonstrate possible use cases, our framework has been applied to a variety of public transport plans which have been created for the same given demand for an artificial urban grid network and to instances for long-distance train networks. As one application we study the impact of different strategies to improve the robustness of timetables by insertion of supplement times. We also show that the framework can be used to optimize waiting strategies in delay management.

The classical planning process in public transport planning focuses on the two criteria operating costs and quality for passengers. Quality mostly considers quantities like average travel time and number of transfers. Since public transport often suffers from delays caused by random disturbances, we are interested in adding a third dimension: robustness. We propose passenger-oriented robustness indicators for public transport networks and timetables. These robustness indicators are evaluated for several public transport plans which have been created for an artificial urban network with the same demand. The study shows that these indicators are suitable to measure the robustness of a line plan and a timetable. We explore different trade-offs between operating costs, quality (average travel time of passengers), and robustness against delays. Our results show that the proposed robustness indicators give reasonable results.

Frequent train delays make passenger-oriented train dispatching a task of high practical relevance. In case of delays, dispatchers have to decide whether trains should wait for one or several delayed feeder trains or should depart on time. To support dispatchers, we have recently introduced the train dispatching framework PANDA (CASPT 2015). In this paper, we present and evaluate two enhancements which are also of general interest. First, we study the sensitivity of waiting decisions with respect to the accuracy of passenger flow data. More specifically, we develop an integer linear programming formulation for the following optimization problem: Given a critical transfer, what is the minimum number of passengers we have to add or to subtract from the given passenger flow such that the decision would change from waiting to non-waiting or vice versa? Based on experiments with realistic passenger flows and delay data from 2015 in Germany, an important empirical finding is that a significant fraction of all decisions is highly sensitive to small changes in passenger flow composition. Hence, very accurate passenger flows are needed in these cases. Second, we investigate the practical value of more sophisticated simulations. A simple strategy evaluates the effect of a waiting decision of some critical transfer on passenger delay subject to the assumption that all subsequent decisions are taken according to standard waiting time rules, as usually employed by railway companies like Deutsche Bahn. Here we analyze the impact of a higher level of simulation where waiting decisions for a critical transfer are considered jointly with one or more other decisions for subsequent transfers. We learn that such "coupled decisions" lead to improved solution in about 6.3% of all considered cases.

Passenger-friendly train disposition is a challenging, highly complex online optimization problem with uncertain and incomplete information about future delays. In this paper we focus on the timing within the disposition process. We introduce three different classification schemes to predict as early as possible the status of a transfer: whether it will almost surely break, is so critically delayed that it requires manual disposition, or can be regarded as only slightly uncertain or as being safe. The three approaches use lower bounds on travel times, historical distributions of delay data, and fuzzy logic, respectively. In experiments with real delay data we achieve an excellent classification rate. Furthermore, using realistic passenger flows we observe that there is a significant potential to reduce the passenger delay if an early rerouting strategy is applied.