Big field data-driven battery pack health estimation for electric vehicles: A deep-fusion transfer learning approach

Accurate estimation of battery state of health (SOH) is significant to guide optimal electric vehicles (EVs) operating and maintenance. Regrettably, the SOH estimation based on real -world EVs' data is still not deep enough, and some methods developed in the laboratory are quite difficult to apply in practice. To bridge the gap, this paper proposes a novel battery SOH estimation framework for on -road EVs. Firstly, based on the open circuit voltage correction and working condition filtering, the labeled capacities of 707 real -world EVs in 3 years are extracted through the historical operating data. The mean absolute percentage error (MAPE) of the label capacity is validated to be less than 2.97 % with full charge and discharge tests. Subsequently, 22 health indicator (HI)points and 4 HI sequences are generated using partial charging data. Using the points -to -point and sequences -to -point methods, two frameworks are then established, which are a global SOH estimation framework targeting all EVs and a deep fusion transfer learning network (DFTN) to improve the SOH estimation of the single EV. The global framework adopts the Gaussian process regression (GPR) method and achieves an impressive MAPE of 2.07 % through 100 times repeated validation. In the DFTN framework, the average MAPE of SOH estimation for 30 testing vehicles is only 1.42 %, with the best -testing vehicle achieving a MAPE of 0.23 %.