Comparison of state of charge estimation methods based on Artificial Intelligence algorithms for lithium-ion batteries used in automotive applications.

In recent years, laws and regulations on vehicle emissions have become increasingly stringent in various countries, leading to the gradual replacement of internal combustion engine vehicles by electric vehicles (EVs) and hybrid electric vehicles (HEVs) as the mainstream. To ensure optimal management and safe operation of the battery, a battery management system (BMS) is introduced to estimate the state of the battery through two basic parameters: State of Charge (SOC) and State of Health (SOH). Due to the complex chemical reaction inside the battery and the aging phenomenon caused by frequent use, it is difficult to clearly observe the status of the battery, so estimating SOC and SOH in real time becomes a challenge. This thesis presents a novel methodology based on Artificial Intelligence. Feed forward neural network (FNN), Long short-term memory (LSTM) and Nonlinear autoregressive neural network with external input (NARX) are taken into SOC estimation performance comparison, then the best neural network is chosen and applied to a specific Li-ion battery module. Through the aging characteristics (specific SOH decreasing range) of this module, three different network classification options are constructed (1-Class,3-Class and 5-Class) and incorporated into SOC subsystems. These SOC subsystems are combined with SOH subsystem respectively, for estimating both SOC and SOH. Finally, comparison between these three combined options is implemented. The developed method is able to achieve on-board and real-time functionality. The results and experiments prove that this methodology is not only feasible, but also has excellent performance of fast and accurate estimation with compared to other state-of-the-art methodologies.