Joint estimation of state-of-charge and state-of-energy of lithium-ion batteries at different ambient temperatures based on domain adaptation and unscented Kalman filter

Accurate estimation of the state-of-charge (SOC) and state-of-energy (SOE) of lithium-ion batteries (LIBs) is fundamental for the battery management system. This paper proposes a method based on the combination of domain adaptation (DA) and unscented Kalman filter (UKF) (DA-UKF) to achieve joint estimation of SOC and SOE at distinct temperatures. A data-driven network consisting of source domain (SD) and target domain (TD) parts is adopted. A gated recurrent unit network and linear layer are used to extract features of the SD and TD datasets, while maximum mean difference and adversarial DA are adopted to align the features. The linear layer outputs SOC and SOE joint estimation results, and the UKF smooths the outputs to obtain accurate and stable joint estimation. Experimental results show that, regardless of whether performing in supervised or unsupervised mode, the DA-UKF can achieve highly robust and accurate joint estimation of SOC and SOE at various temperatures. Compared with other advanced methods, the root mean square error and the mean absolute error of the DA-UKF, at different temperatures, reduce, on average, between 49.760% and 84.150%, and 53.579% and 84.787%, respectively. Moreover, the DA-UKF does not require complex adjustments to the hyperparameters of the network.