Research on Capacity Modeling and Prediction of Power Batteries Based on Embedded Sensors

This paper investigates capacity modeling and life prediction for lithium-ion power batteries based on embedded sensors. The specific research work includes: Firstly, to meet the demand for multidimensional battery parameter sensing, intelligent sensing devices were selected and integrated. Sensors were implanted within the battery to monitor distributed internal data. Secondly, a series of cyclic aging experiments were designed to systematically investigate the effects of temperature and charge/discharge rates on battery aging characteristics, yielding five sets of battery data. Subsequently, a recurrent neural network (RNN)-based approach was employed for lithium-ion power battery capacity modeling. Prediction results were compared with other machine learning algorithms through error analysis, validating its accuracy. However, since neural networks cannot quantify prediction accuracy, Gaussian process regression (GPR) was adopted for remaining life prediction. The adaptability and accuracy of the GPR model were verified under various experimental conditions, providing support for further optimization of battery management systems.