Lithium-ion Battery State of Health Estimation with Recurrent Convolution Neural Networks

Artikel i vetenskaplig tidskrift, 2022

State of health (SOH) estimation of lithium-ion batteries is one of the major functions conducted by battery management systems. Machine learning and neural networks become popular for SOH estimation in recent years, especially when huge amounts of battery data from laboratories and end-users are available. This paper proposed a new SOH estimation structure, which combines recurrent neural network and convolution neural network together. Measurements not only from the studied cycle but also from the previous two cycles is used together to estimate the studied cycle SOH. By this way, hidden information of aging trends among cycles is utilized. Furthermore, two-dimensional filters in convolution layers are applied to study features of voltage, current, and temperature during the charging process. Validation results show that the proposed SOH estimation structure can reduce the validation set loss by 31.5% and 18.8% respectively, compared to the long-short-term model and another proposed reference structure that combines recurrent neural network and feedforward neural network together.