Electrothermal modeling of lithium-ion batteries using an artificial intelligence technique-based approach for powerwall applications

Abstract

The Powerwall, an innovative home battery powered by solar energy, is gaining popularity in the United States and Australia due to its ability to provide complete power to homes and serve as a backup during electrical supply interruptions. Recently, lithium-ion (Li-ion) battery technology has received significant attention from the industry and academia, standing out for offering greater energy capacity, power density, efficiency, and lower self-discharge rate compared to other technologies such as NiCd and NiMH batteries. Initially, this research aimed to model the Li-ion cells of the Powerwall using the equivalent circuit model (ECM). However, a detailed review of the literature indicated that a data-based approach, employing Deep Learning (DL) techniques or deep neural networks, represents the state of the art in the field. DL, a subset of machine learning and part of Artificial Intelligence (AI), offers notable advantages over ECM, including: i) high precision in estimating the non-linear relationships between voltage, current, temperature, and hidden variables such as the State of Charge (SOC); ii) the ability of a single artificial neural network to model the behavior of a Li-ion cell across different temperature ranges (example: from 0°C to 40°C); and iii) overcoming the challenge of identifying ECM parameters. Therefore, with the advancement of the research, the objective was updated to investigate and develop DL models, specifically feed-forward neural networks, for estimating the SOC of lithium-ion batteries in Powerwalls. The results obtained focus on the precise estimation of the SOC of Li-ion cells, underlining the efficacy and innovation of this methodological approach.