EXPLORING BATTERY USAGE IN ELECTRIC VEHICLES THROUGH GRAPH BASED CASCADED CLUSTERING

Dhanashree Balaram, Brett Dufford, Gianina Alina Negoita, William A. Paxton, Matthew Yen

Keywords: cascaded clustering, time-series clustering, graph neural networks, AI and batteries, electric vehicles, two-level clustering

TL;DR: We propose a graph based cascaded clustering approach that leverages battery specific features to identify usage patterns that affect the battery across multiple timescales.

Abstract: Recently, electric vehicles (EVs) have gained popularity over internal-combustion engine vehicles (ICEV) because of their convenience and ability to use clean- energy sources. Fueling an EV is fundamentally different than an ICEV and thus, the driving and charging patterns associated with EVs are novel and not well understood. For example, where filling the tank of an IECV is a standard process, charging an EV occurs at different speeds (L1, L2, and DC Fast Charging) and intermittently while driving by regenerative braking. Understanding these usage patterns for EVs is important because the performance and longevity of the battery is dependent on the driving, charging, and idling patterns it is subjected to over its lifetime. We propose a scalable cascaded clustering approach that leverages battery- specific features to identify usage patterns that affect the battery across multiple timescales. We analyze 3,100 EVs over the course of a year using multivariate time- series data consisting of but not limited to state of charge (SOC) and mileage. First, we apply clustering to weekly multivariate data segments and extract usage profiles at that timescale. Then, we use these weekly cluster assignments to generate an EV battery meta-sequence that is unique to every vehicle, which reveals longer- term patterns. We apply a novel graph based clustering technique at the vehicle meta-sequence level to associate groups of vehicles that are operated similarly. Our approach reveals fine-grained usage patterns and helps identify salient themes across a vehicle’s lifetime. While limited to a relatively small selection of vehicles, our work reveals a unique representation of vehicles and their weekly usage pattern that can potentially aid in battery lifecycle management.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 6447