The Viriato Algorithm Platform

Global warming is one of the greatest challenges facing modern mankind. How we are to meet it is a discussion that must be held now. The central question is how to reduce the burning of fossil fuels without risking a global economic lockdown? The transport of goods and people is a crucial element of our economic system and today’s prosperity. Maintaining sustainable mobility powered by renewable energy sources is therefore a basic prerequisite for achieving a transformation to an economy that is sufficiently efficient to sustain the necessary technological research. Fortunately, there is a solution to this important problem: the railways.

Rail has decisive advantages in local transport in conurbations, both in competition with motorised individual transport, as well as in high-speed long-distance transport as an alternative to aviation. Only perhaps in peripheral areas will there be mobility concepts in the future that are economically superior to the railway.

The decisive factor in solving the rail mobility problem is to make better use of existing resources and to optimise the allocation of additional infrastructure in the network-wide offer in terms of capacity utilisation. One approach to solving this problem is the consistent implementation of first-class long-term planning in network development, with the greatest possible automation of planning and operations, as well as the comprehensive optimisation of the use of the required resources.

These two classes of problems, optimal use of personnel and rolling stock resources and automatic resolution of capacity allocation conflicts in the broadest sense have been extensively addressed by algorithmic research over the past three decades. As experience and theory have shown, solutions within these problem classes are difficult to find. So far, research has produced practically relevant solutions for only a few specific subproblems.

Researchers have often been hindered by the need to solve recurring practical difficulties. For example, the procurement and preparation of required data or the visualisation of identified solutions to analyse the quality of the solution and their practical feasibility. In addition, the algorithms require common software components, such as running time calculation, conflict detection or basic simulation methods.

The development of appropriate software systems that provide such functionalities with professional levels of quality typically exceeds the capabilities of academic research institutes in terms of time and money. In fact, such systems are very complex and extremely expensive to develop and maintain. This is particularly the case because of the high performance requirements and the range of functionality to be covered, which are necessary for the development and investigation of real timetables.

Given the importance of these challenges, SMA decided three years ago to open up Viriato to algorithmic research by implementing an algorithm platform and making it available to the interested community.

The Algorithm Platform provides the research community with three basic functions:

• Viriato takes over the middleware functionalities for loading and integrating large amounts of data that are needed to solve practical problems (data acquisition).
• The algorithm platform offers a comprehensive domain-oriented data model, the Abstract Intermediate Data Model (AIDM), which provides infrastructure and timetable data in a format suitable for algorithmic research (data provision). This model, together with a number of supporting functions, serves to make Viriato’s algorithm API available to the research community. This means that they no longer need to implement functions whose role is to merely support the actual algorithmic work.
• The algorithm platform provides interfaces to basic software components such as running time calculation or conflict detection and takes over the task of communicating with these components as well as with Viriato itself.

SMA undertook the initial development of the algorithm platform using its own resources and is supporting the community with close cooperation in the development of the API. This enables us to understand and acquire the knowledge required for the continuous improvement of the interfaces. The independence of SMA makes it possible to carry out such initiatives without being forced to restrict ourselves to specific use cases. This enables Viriato to be opened up to a broad spectrum of algorithmic applications. In turn, researchers are provided with a tool that allows them to analyse and evaluate the solutions to various algorithmic problems using a professional and widely used system.

For the customers of Viriato, the algorithm platform offers the possibility to benefit directly from the results of the research community, and to tackle specific problems more efficiently with their existing research partners, benefiting from the industrialisation of algorithm development. In addition, suitable automation and optimisation algorithms that are seamlessly linked to Viriato support the planning process in becoming better and faster.

SMA has established contacts with universities and research institutes continuously over the past few years. The focus here is on internationally competitive research groups whose core competences are algorithmic research and the use of mathematical methods for the purposes of automation and optimisation. The initial collaborations took place with the École polytechnique fédérale de Lausanne (EPFL) and the University of Novi Sad. Here, the focus was on the use of the algorithm platform in teaching: SMA supported a dissertation on the topic “Automated rescheduling of trains in fault management” as the second supervisor. Currently, SMA is supporting the TU Delft in a project planning maintenance tasks for an existing vehicle roster. A thesis on line planning with the ETH Zurich is planned for later this year.

By using it in teaching, both academics and SMA can assure themselves of the accuracy of fit and usefulness of the provided functionality. During the ongoing collaborations, SMA has been continuously in close contact with the supervising university partners in order to gather knowledge such as new requirements or ideas for improvement and to incorporate them into the further development of the algorithm platform. The successful cooperation in teaching forms the basis for a subsequent cooperation in research. SMA is looking forward to supporting a first PhD at the University of Novi Sad, which will extend the algorithm developed at EPFL. With the first successful references from these collaborations, the circle of users is expected to grow continuously.