University of Minnesota
The Weighted Mile: Understanding Spatial Accessibility Through Electric Vehicle Charger Optimization and Purpose-Specific Spatial Interaction Modeling
Abstract
dc:description.abstractSpatial accessibility, the ease with which individuals can reach destinations, is a contested and evolving construct whose measurement has shifted from simple proximity buffers to complex, network-aware models that account for individual travel behavior. This thesis argues that accessibility can only be adequately understood at the intersection of three fields that have developed complementary but largely siloed toolkits: spatial econometrics, which provides the gravity models and friction-of-distance coefficients that quantify the cost of spatial separation; network science, which provides the graph-theoretic structures that model flow and optimization over transportation networks; and geographic information science (GIS), which provides the georeferenced socioeconomic and point-of-interest (POI) data that anchor models to observable urban reality. No single tradition is sufficient: a gravity model without network topology treats all directions as equivalent, a shortest-path algorithm without socioeconomic context optimizes routes without asking for whom, and a GIS layer without a behavioral model of spatial interaction counts nearby facilities without asking whether anyone actually uses them. Chapter 1 develops this interdisciplinary framework through a synthesis organized around three eras of friction. The static era measured access as binary proximity within fixed-radius catchments. The impedance era introduced continuous distance decay governed by a calibrated friction coefficient (β). The path-dependent era, motivated by trajectory data and the practical demands of EV charging, redefines access as a property of trips rather than locations; measured by deviation distance (Tdₑᵥ) rather than residential proximity. This progression frames a path-dependent equity argument: the transition to electric vehicles introduces new spatial frictions—finite range, non-trivial charging time, sparse infrastructure—that compound existing socioeconomic disparities in distance tolerance. For populations whose friction coefficients are already steep, any additional deviation for charging constitutes a disproportionate spatial tax. Chapter 2 implements the network science and GIS dimensions of this agenda by formulating EV charger siting as a graph-based sequential decision problem. A graph convolutional network (GCN) parameterized deep Q-network places chargers on a road network graph of Hennepin County, Minnesota, optimizing a Gaussian two-step floating catchment area (G2SFCA) accessibility score computed over deviation distance rather than Euclidean distance. Evaluated against random, greedy, and MLP-DQN baselines across 34,089 commuters and 7,558 candidate nodes, the GCN-RL agent achieves consistently higher G2SFCA scores, reduces aggregate route deviation by approximately 30% relative to the MLP-DQN, and produces qualitatively different—and more spatially dispersed—placement solutions, as confirmed by low Jaccard similarity and higher nearest-neighbor distances between placement sets. These results demonstrate that encoding road network topology via graph convolution enables the agent to identify corridor-aligned placements that serve commuters across distributed residential areas, a capability that flat, tabular models lack. Chapter 3 implements the spatial econometrics dimension by deconstructing the gravity model’s origin mass term and friction coefficient for healthcare trips. A purpose-specific mass (PSM) specification replaces undifferentiated residential population with the senior population (ages 65+) as the origin mass for Health and Medicine trips, and is evaluated across three structurally distinct U.S. metropolitan areas: the Twin Cities, Chicago, and Miami. The PSM substitution produces a null result—model fit, estimated coefficients, and predicted flows are essentially unchanged—a finding attributed to the limited spatial variation in senior population share across census block groups, which renders the two mass variables near-collinear in log-space. Despite this null result, a complementary per-CBG friction analysis reveals that healthcare distance decay is not a universal parameter but is systematically associated with neighborhood-level socioeconomic characteristics. High-poverty, high-minority, transit-dependent neighborhoods exhibit flatter distance decay—not because distance is less costly for their residents, but because sparse nearby healthcare options force longer travel to reach appropriate care. This finding reframes flat friction not as insensitivity to distance but as a signature of poor local accessibility, connecting gravity model estimation directly to the spatial equity concerns that motivate the thesis. Together, these chapters demonstrate that the friction of distance is neither fixed nor uniform, but is shaped by network topology, trip purpose, and socioeconomic context. Chapter 2 shows how infrastructure can be optimally sited to minimize path-dependent friction for EV charging; Chapter 3 shows that friction itself is socioeconomically stratified, establishing the empirical basis for equity-aware weighting in future optimization frameworks. The thesis contributes both methodological tools—a GCN-RL charger siting framework and a purpose-specific spatial interaction modeling approach—and empirical evidence to the growing literature on sustainable, equitable urban accessibility.
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Harris, Jacob
Subjects
dc:subject × 6Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Citation
- Harris, Jacob. (2026). The Weighted Mile: Understanding Spatial Accessibility Through Electric Vehicle Charger Optimization and Purpose-Specific Spatial Interaction Modeling. Retrieved from the University Digital Conservancy, https://hdl.handle.net/11299/283435.