Nationwide Analysis of Spatial-Temporal EV Charging Patterns Using Machine Learning Techniques

Abstract: Electric Vehicles (EVs) offer a promising solution to reduce carbon emissions and promote climate-friendly transportation. However, the rapid adoption of EVs has raised concerns about the adequacy and equity of EV charging infrastructure. Challenges such as high charging costs and the unequal distribution of EV Charging Stations (EVCS) limit access for vulnerable populations, posing barriers to widespread EV usage. To address these challenges, a deeper understanding of EV charging behaviors, spatial disparities in EVCS placement, and potential future locations for infrastructure development is critical for stakeholders including local governments, energy providers, and manufacturers.

This study presents a comprehensive nationwide analysis of EV charging usage patterns, incorporating both spatial and temporal dimensions. By utilizing unsupervised machine learning techniques, such as the Clustering LARge Applications (CLARA) algorithm and agglomerative hierarchical clustering, we analyze real-time charging data to uncover distinct usage profiles. CLARA, an extension of k-medoids for large datasets, is applied to classify user charging behavior based on both the magnitude and shape of usage profiles. This method captures temporal trends and variation across daily, monthly, and seasonal time scales. Additionally, agglomerative hierarchical clustering is used to group charging stations based on similarities in their usage profiles.

The results are expected to reveal clusters of EV charging behaviors that differ in terms of peak usage times, magnitudes, and responses to various factors, such as socioeconomic conditions and the impact of natural hazards. These findings will allow for a more detailed understanding of EV power demand, providing critical insights for optimizing the development of EV charging infrastructure to meet increasing demand. The study’s large-scale, data-driven approach offers a significant improvement over prior research, which often relied on synthetic or survey-based data. By leveraging real-time usage data across the US, this analysis provides a clearer picture of how to improve the resilience and equity of the EV charging network.

The outcomes of this research hold potential for guiding investment and policy decisions related to EV infrastructure, contributing to a more sustainable and efficient energy grid that is responsive to emerging challenges and opportunities in the EV market.