Data-Driven Approaches to Battery Health Monitoring in Electric Vehicles Using Machine Learning

Electric vehicles (EVs) have become a popular mode of transportation due to the decreasing cost of energy storage. However, this decrease in energy storage cost has led to an increase in the number of battery durability issues. The main issue with batteries is the uncertainty and unavailability of real-time trace capacity, which has hindered the development of new value-added business models based on the battery state of electric vehicles. This article introduces battery health monitoring in electric vehicles using machine learning techniques as a response to these business opportunities and analytical challenges in the energy transition. Specifically, this article presents an integrated virtual battery prototype that serves as a valuable mid-stream application example, a carbon balance optimization application. We use LENs-Lab's heterogeneous architecture to perform online, in-service, system-wide health monitoring for electric vehicle fleets while minimizing the carbon balance with grid optimization under constraints.Supervised learning methods use labeled data to learn a mapping between input features and output labels. For the virtual battery prototype integration, we employ a supervised Random Forest and Deep Neural Network regression on a labeled series of electric vehicle state data. The supervised learning model learns to link an estimated real-time energy slack to the corresponding state of health of the energy storage through a training dataset. This presents potential high-value end-use participation, from footprint-constrained drivers, fleet managers, and utility/grid operations managers. Furthermore, we provide a carbon balance minimization application. The goal of this application is to reduce the carbon balance, the amount of carbon spent on the electricity network, by minimizing the charging cost of the electric vehicles' electricity consumption. In this application, the carbon balance minimization application uses the output from the supervised learning model to determine the impact electric vehicle users will have on vehicle grid integrity, in two forecasting time intervals of one week and the present-day decision. The EUDC driving cycle for electric vehicle usage and the individual's token values of elasticity usage timing are used to determine this user impact. The model presented in this article can enable continuous monitoring of the battery state of health, which can provide new use cases and commercial models to empower electric vehicle users.