State-of-health estimation of lithium-ion batteries based on semi-supervised transfer component analysis

Accurate state-of-health estimation can ensure the safe and reliable operation of Lithium-ion batteries in any given application. Nevertheless, most of the state-of-health estimation methods require a large amount of laboratory aging data to offer precise results. As obtaining battery aging data under laboratory conditions requires a considerable amount of time and incurs high economic costs, in this paper, a method based on transfer learning is proposed to monitor state-of-health of batteries. A novel data processing method based on maximum mean discrepancy is considered to eliminate redundant information and minimize the difference between different data distributions. Then, mutual information is used to prove that the correlation between processed data is not decreased. To validate the developed transfer learning method, the data sets of four batteries in different working conditions are considered. Different error-detection methods, maximum average error, mean squared error and root mean squared error, which are utilized to evaluate the proposed model. The state of health is estimated effectively with less than 2.5% error considering the aforementioned errors after processed by using semi-supervised transfer component analysis algorithm, although the training set only accounts for about 35% of the entire set. The results indicate that transfer learning plays an important role in improving the estimation accuracy of a battery state-of-health.