Towards more granular choice set building in location choice models: A graph-structured meta-topology as an alternative to matrices in agent-based travel demand models

Kurzfassung

Microscopic and agent-based travel demand models frequently rely on precomputed matrices describing travel times, distances, costs, or interchange counts between traffic analysis zones (TAZ). While these matrices simplify computational efforts, their reliance on spatial aggregation means that finer-grained, more behaviorally realistic insights are sacrificed. To achieve a more accurate microscopic view, one would need to divide large TAZs into smaller ones. However, this solution quickly becomes problematic, as it leads to a growth in matrix size, increasing both the memory footprint and the amount of precomputation necessary to maintain these matrices. To address these limitations, we present a method that employs a precomputed meta topology, a graph structure where each node can be annotated with additional information. As a case study, we implement a constrained Dijkstra-based graph traversal algorithm on this meta topology to generate location choice sets for agent-based models. This approach significantly reduces memory requirements while introducing additional computational overhead, as it involves constructing multiple subgraphs. Although it enhances granularity and offers a more behaviorally nuanced perspective than matrix-based methods, certain trade-offs and performance considerations must be acknowledged. In the final section, we discuss how the meta topology can be further leveraged, as well as strategies for improving the efficiency and robustness of the proposed solution.