Charge Diagnostics and State Estimation of Battery Energy Storage Systems Through Transformer Models

With the continuous development of Artificial Intelligence (AI), designing accurate algorithms that provide diagnostics and maintenance of energy technologies is a challenging task in the energy transition domain. This research work focuses on the implementation of Transformer models for charge diagnostics and algorithm design of Battery Energy Storage Systems (BESSs). Experimentally, two Lithium-ion (Li-ion) battery cells were tested using a programmable DC electronic load to evaluate charge indicators, and 20 battery tests were performed for each cell. Filter, Wrapper, and Embedded methods are the techniques implemented to achieve Feature Selection and illustrate Key Performance Indicators (KPIs) in battery testing. Time series and state estimation are the Supervised Learning techniques executed for charge diagnostics and State of Charge (SOC) predictions. The results show remarkable performance metrics of the Transformer models, achieving over 94% accuracy in Model evaluation compared to traditional Deep Learning algorithms.