State of health prognostics for series battery packs: A universal deep learning method

Abstract Prognostic and health management for battery packs depend greatly on the accurate and efficient state of health prognosis. This paper proposed a strategy for generating universal health indicators and model fusion method. Several health indicators that reflect the integral characteristic and information distribution are generated and proved with high correlation with capacities. The generation method can be adapted for various battery packs regardless of the battery types, connected cell numbers, and aging statuses. Then the generation method is extended to dynamic working conditions by combining the mean plus difference model and recurrent least-square online parameter identification. Thanks to the uniform feature input and state of health output, a universal prognostic model is constructed with deep learning frameworks. The prognostic performance is further improved by the model migration and fusion, which extend the application area to predict the state of health for different battery packs and/or under different working conditions. Experimental results show that the prognostic model can be implemented among various battery packs with satisfactory accuracy and reliability. The mean absolute errors and root mean square errors are less than 2.5% and 3.1%, respectively, under various application occasions.