State of Charge an‎d State of Health Estimation of Li-ion Batteries Using Hybrid Data-Driven Methods

10/18/2026 12:00:00 AM

Throughout history, a main concern of humanity has always been to increase welfare an‎d comfort, an‎d to address this concern, much progress in technology has been made. However, during the last few years, there have been limitations regarding technological advancements imposed by humans themselves. These limits stem from a change in priorities an‎d an increase in concerns regarding issues such as the environment. For example, automobiles, which were once considered a scientific an‎d technological breakthrough, are now being questioned due to years of cumulative pollution caused by them. To address this concern, electric vehicles, an‎d especially battery electric vehicles, have been introduced as a solution that does not produce pollution during their operational life. Given this fact an‎d the importance of batteries as the main component of these technologies, much research has been conducted regarding the maintenance of these vehicles an‎d the estimation of their vital parameters, such as state of charge (SOC) an‎d state of health (SOH). In this work, a method based on physics-informed neural networks (PINNs) is introduced to co-estimate the state of charge an‎d state of health in a lithium-ion battery. As physical constraints, a novel method called sparse identification of nonlinear dynamics (SINDy) was used to extract meaningful equations an‎d a model that would describe SOC an‎d SOH in order to train the PINN. Using SINDy ensures that the extracted equations describe changes in SOC an‎d SOH based on the changes in inputs of the system an‎d other physical parameters, which provide a meaningful physical constraint. As a result, it was shown that the mean absolute error (MAE) of the model was 0.085898% with a coefficient of determination of 0.9970. To further demonstrate the model’s generalizability, using a transfer learning technique, the model was then deployed on an experimental battery dataset, namely the NASA dataset, an‎d the best-case results were a coefficient of determination of 0.8707 for SOC estimation an‎d 0.9904 for SOH estimation.

باتري ليتيوم_يون , وضعيت شارژ , وضعيت سلامتي , يادگيري عميق , شبكۀ عصبي فيزيك‌آگاه

Lithium-ion batteries , State of charge , State of health , Deep learning , Physics-informed neural networks

AmirMohammad Abbasi

Dr. Ayat Gharehghani