Abstract: Accurate and efficient battery health estimation is very important for vehicle battery management. The data-driven method, which does not depend on the complex reaction mechanism and underlying mechanism inside the battery, is widely used in the field of SOC and state of health estimation. Due to the high computational complexity caused by training and testing a large amount of data in the previous methods, a new data mining technology of fuzzy information granulation is proposed. Firstly, the asymmetric Gaussian membership function is proposed to further improve the performance of data mining at three granular levels: upper boundary, mean and lower boundary. Then, a lithium-ion battery cycle test is conducted in combination with the Gaussian process regression method to verify the performance of the state of charge estimation. Finally, a battery capacity estimation model is established and the performance of the health state prediction is verified using a public dataset. The results show that by combining the fuzzy information granulation technology, the state of charge estimation accuracy of the Gaussian process regression method is improved by 15.38% and 31.25%, respectively, under the two current conditions, and the computation time can be reduced from 26.9 s to 3.3 s. The state of charge estimation by fuzzy information granulation can achieve high accuracy among the five machine learning methods and require only the lowest computational cost, thus providing ideas and guidance for highly efficient state of charge estimation of lithium-ion batteries based on data-driven methods.

Key words: lithium-ion batteries, fuzzy information granulation, Gaussian process regression, state of charge, state of health