RUL Prediction Based on MBGD-WGAN-GRU for Lithium-Ion Batteries

To address the challenges associated with acquiring complete charge-discharge cycle data and extracting health indicator factors (IHFs) from fragmented datasets in current automotive lithium-ion batteries (LIBs), this study proposes a novel online remaining useful life (RUL) prediction method. First, the IHF, which captures battery aging characteristics, is extracted from raw LIBs data, and the dataset is partitioned into training (70%) and testing (30%) subsets. Subsequently, the mini-batch stochastic gradient descent algorithm is employed to optimize a Wasserstein generative adversarial network, thereby augmenting the training set and enhancing data diversity. The extracted IHF and expanded training set are then used to train a gated recurrent unit (GRU) model, leveraging GRU’s strengths in sequential data processing to improve the characterization of battery aging trends. Finally, the proposed GRU-based model is validated for RUL prediction using an open-source test dataset. Experimental results demonstrate that the mean absolute error of the proposed method on B0005, B0006, and B0007 batteries is 0.0023 Ah, 0.0030 Ah, and 0.0014 Ah, respectively, confirming its effectiveness and practical applicability. This approach provides robust technical support for LIBs health management.