State of health estimation based on inconsistent evolution for lithium-ion battery module

Estimating state of health for battery module is one of the most significant and challenging techniques to promote the commercialization of electric vehicles. Based on the relationship changes of branch current and its estimation error during aging, a state of health estimation general framework is presented for battery module. Firstly, the parallel battery module aging experiment is designed. In addition, the consistency changes of branches were analyzed. A neural network model utilizing dual back-propagation for estimating branch current errors was developed by employing the experimental data of battery module. Through estimation error of branch current under five working conditions, two aging characteristics are extracted, one is the slope of compensation value and current, the other is the slope of compensation value and current change rate. These features are fed into gaussian process regression training to obtain a state of health estimation model for the battery module. Furthermore, the model is validated with new european driving cycle working condition. Finally, a dual bidirectional long short-term memory neural network is utilized to illustrate the versatility of the presented universal framework, which can effectively estimate state of health of battery module with the maximum relative error of 2.1226 %.