Prognostics of battery cycle life in the early-cycle stage based on hybrid model

Accurately predicting the remaining useful life (RUL) of lithium-ion batteries in early-cycle stage can speed up the battery improvement and optimization. However, slowly varying and weak predictability of the characteristic quantities in early-cycle stage make it challenging to predict RUL. To overcome this problem, the paper proposes a hybrid prediction model, which integrates random forest (RF), Artificial Bee Colony (ABC) and general regression neural network (GRNN), called RF-ABC-GRNN. First, a new feature space is obtained by performing linear and non-linear transformations on the original features that are slowly changed in early-cycle stage, such as discharge capacity, terminal voltage, discharge current, and internal resistance. Second, RF is used to measure and rank the importance of these new features so as to screen out the high-importance feature combination. Third, the prediction model based on GRNN is constructed. Considering that the smoothing parameter of the model has great influence on the prediction performance, ABC is used for parameter optimization. Finally, in order to verify perfor-mance of the model, initial cycles data that have yet to exhibit apparent degradation is used. Comparison results show that the proposed model could effectively screen out the high-importance features and make accurate prediction much earlier. (c) 2021 Elsevier Ltd. All rights reserved.

Automated summary

This paper proposes a hybrid prediction model RF-ABC-GRNN, integrating RF, ABC, and GRNN, to accurately predict the remaining useful life of lithium-ion batteries in the early-cycle stage. By screening high-importance feature combinations and optimizing parameters, the model can make accurate predictions earlier.