Query processing in complex modern traffic networks
Gregor Jossé · Elektronische Hochschulschriften der LMU München (Ludwig-Maximilians-Universität München) · 2016
The transport sector generates about one quarter of all greenhouse gas emissions worldwide. In the European Union (EU), passenger cars and light-duty trucks make up for over half of these traffic-related emissions. It is evident that everyday traffic is a serious environmental threat. At the same time, transport is a key factor for the ambitious EU climate goals; among them, for instance, the reduction of greenhouse gas emissions by 85 to 90 percent in the next 35 years. This thesis investigates complex traffic networks and their requirements from a computer science perspective. Modeling of and query processing in modern traffic networks are pivotal topics. Challenging theoretical problems are examined from different perspectives, novel algorithmic solutions are provided. Practical problems are investigated and solved, for instance, employing qualitative crowdsourced information and sensor data of various sources. Modern traffic networks are often modeled as graphs, i.e., defined by sets of nodes and edges. In conventional graphs, the edges are assigned numerical weights, for instance, reflecting cost criteria like distance or travel time. In multicriteria networks, the edges reflect multiple, possibly dynamically changing cost criteria. While these networks allow for diverse queries and meaningful insight, query processing usually is significantly more complex. Novel means for computation are required to keep query processing efficient. The crucial task of computing optimal paths is particularly expensive under multiple criteria. The most established set of optimal paths in multicriteria networks is referred to as path skyline (or set of pareto-optimal paths). Until now, computing the path skyline either required extensive precomputation or networks of minor size or complexity. Neither of these demands can be made on modern traffic networks. This thesis presents a novel method which makes on-the-fly computation of path skylines possible, even in dynamic networks with three or more cost criteria. Another problem examined is the exponentially growth of path skylines. The number of elements in a path skyline is potentially exponential in the number of cost criteria and the number of edges between start and target. This often produces less meaningful results, sometimes hindering usability. These drawbacks emphasize the importance of the linear path skyline which is investigated in this thesis. The linear path skyline is based on a different notion of optimality. By the notion of optimality, the linear path skyline is a subset of the conventional path skyline but in general contains less and more diverse elements. Thus, the linear path skyline facilitates interpretation while in general reducing computational effort. This topic is first studied in networks with two cost criteria and subsequently extended to more cost criteria. These cost criteria are not limited to purely quantitative measures like distance and travel time. This thesis examines the integration of qualitative information into abstractly modeled road networks. It is proposed to mine crowdsourced data for qualitative information and use this information to enrich road network graphs. These enriched networks may in turn be used to produce routing suggestions which reflect an opinion of the crowd. From data processing to knowledge extracting, network enrichment and route computation, the possibilities and challenges of crowdsourced data as a source for information are surveyed. Additionally, this thesis substantiates the practicability of network enrichment in real-world experiments. The description of a demonstration framework which applies some of the presented methods to the use case of tourist route recommendation serves as an example. The methods may also be applied to a novel graph-based routing problem proposed in this thesis. The problem extends the family of Orienteering Problems which find frequent application in tourist routing and other tasks. An approximate solution to this NP-hard problem is presented and evaluated on a large scale, real-world, time-dependent road network. Another central aspect of modern traffic networks is the integration of sensor data, often referred to as telematics. Nowadays, manifold sensors provide a plethora of data. Using this data to optimize traffic is and will continue to be a challenging task for research and industry. Some of the applications which qualify for the integration of modern telematics are surveyed in this thesis. For instance, the abstract problem of consumable and reoccurring resources in road networks is studied. An application of this problem is the search for a vacant parking space. Taking statistical and real-time sensor information into account, a stochastic routing algorithm which maximizes the probability of finding a vacant space is proposed. Furthermore, the thesis presents means for the extraction of driving preferences, helping to better understand user behavior in traffic. The theoretical concepts partially find application in a demonstration framework described in this thesis. This framework provides features which were developed for a real-world pilot project on the topics of electric and shared mobility. Actual sensor car data collected in the project, gives insight to the challenges of managing a fleet of electric vehicles.