APPLICATIONS OF MACHINE LEARNING IN UTILITY PLANNING FOR CLIMATE RESILIENCE

Utility services are critical infrastructures that shape how urban populations experience the impact of planning policies. Climate change and increasing extreme weather events challenge utilities to maintain service levels while adapting to changing environmental and technological regimes. This dissertation investigates the intersection of utility planning, climate resilience, and applied machine learning through three interconnected papers. The research addresses three critical utility planning challenges: workforce development amid the clean energy transition, water service accessibility for mobile home park residents, and exposure to flooding risk to solar photovoltaic infrastructure. The study pursues two overarching objectives: methodologically, it examines the efficacy of applied machine learning techniques in utility planning applications; substantively, it applies these methods to address specific climate resilience challenges. Paper I uses Natural Language Processing and unsupervised clustering to identify occupational transition pathways for clean energy workforce development, demonstrating the utility of the Occupational Information Network (O*NET) framework as a reconnaissance tool for workforce planning. Paper II employs computer vision to identify 8,460 Mobile Home Parks in North Carolina and characterizes their water utility service accessibility, finding distinct patterns of accessibility inside and outside urban area cluster boundaries. Paper III evaluates flood risk exposure of small and utility scale solar installations in Central and Coastal North Carolina using 100-year digital flood maps and hurricane flood footprints, revealing complementary risk characterizations between the two assessment methods. The dissertation demonstrates how applied machine learning can address utility planning data gaps through information organization, prediction, and structured information extraction. The methods include unsupervised clustering for reorganizing occupational information, computer vision for infrastructure identification, and Large Language Models for information extraction. The research generates datasets to be publicized for use by the climate resilience community of practice. The findings contribute to understanding workforce development pathways, spatial patterns of utility service accessibility, and patterns of exposure to infrastructure risks, while highlighting opportunities for future research in climate resilient utility planning.