A review on data-driven SOC estimation with Li-Ion batteries: Implementation methods & future aspirations

The goal to minimize global carbon emissions from fossil fuels and diesel-powered vehicles has encouraged the development of electric vehicles and associated battery storage systems. Traditional data driven machine learning techniques are resource hungry, to estimate or predict a SOC. Thanks to Edge computing and Tiny ML, which aided in offering a unique opportunity to employ machine learning techniques on the edge to compute SOC. The Tiny ML paradigm - Embedded Machine Learning - seeks to shift a substantial chunk of users from conventional High-End to Low-End devices. Maintaining learning model correctness, offering training for deployment capabilities in Micro edge devices, optimize processing capacity, and enhancing resilience are just a few of the issues that arise during such a move. Therefore, this review presents an up-to-date data-driven state of charge estimation techniques, with algorithm, input features, execution method, strength, weakness, and estimation error. This review critically explores the various implementation factors of the data-driven algorithms in terms of computational cost in terms of error percentage and footprint of the memory. In addition, the review explores the deficiencies of existing infrastructure such as experimental setups, datasets, computing platforms and software frameworks to identify the gaps for future research. The paper concludes with several useful future directions that will help research in the automotive domain to develop a reliable and accurate SOC estimate method for use in future applications involving sustainable electric vehicle technology.