The state of health (SOH) of batteries is an important but unmeasurable parameter closely related to battery safety and durability. However, most existing SOH estimation strategies rely on a specific load profile (e.g., constant current). To tackle this issue, we here report a method that first converts the dynamic voltage trajectories to the curves corresponding to the constant current profiles using a neural network model. Then, the aging characteristics are selected and the battery SOH is estimated accordingly from the converted voltage data using a Gaussian process regression model. Batteries with different aging degrees are tested with different working conditions to verify the proposed method. Numerically, the errors of the voltage reconstruction are bounded within 2 mV, while the SOH estimation errors under four dynamic working conditions remain within 2%. Our technical approach reduces the dependency of traditional SOH estimation methods on specific working conditions and shows strong potential for practical applications.
Bibliographical noteFunding Information:
This research is supported National Natural Science Foundation of China (NSFC) under Grant numbers 51977131 , 52277223 , and 52277222 , Hong Kong RGC Postdoctoral Fellowship Scheme ( PDFS2122-6S06 ), Shanghai Science and Technology Development Fund 22ZR1444500 and 19ZR1435800 .
© 2023 Elsevier Ltd
- Electric vehicles
- Lithium-ion batteries
- Neural network
- State of health
- Voltage reconstruction