Indirect health state prognosis of lithium-ion batteries based on VMD decomposition and neural network model

Monitoring internal changes in lithium-ion batteries during operation remains a challenge, as direct measurement of their State of Health (SOH) is infeasible, potentially delaying preventive maintenance or replacement. This paper presents a data-driven framework for indirect SOH prediction, offering a comprehensive methodology for feature extraction, evaluation, and model construction. Initially, potential indirect data is extracted from relevant datasets, and feature reconstruction techniques are employed to construct indirect health indicators (HIs). Pearson and Spearman correlation analyses are then applied to select optimal HIs, while the Variational Mode Decomposition (VMD) algorithm is utilised to decompose these indicators into distinct phases. To address practical prediction scenarios and the characteristics of feature data under indirect prediction, a hybrid neural network model, VMD-CNN-BiLSTM-AM, is proposed. The model integrates the advantages of various neural network architectures. Experimental validation is conducted using reconstructed CALCE and NASA PCoE datasets to evaluate overall SOH prediction accuracy and End-of-Life (EOL) prediction accuracy. Results demonstrate that the proposed model achieves high prediction accuracy with only 50% of the training set, highlighting its effectiveness and robustness. This paper validates the feasibility of the indirect SOH prediction approach, offering a reliable solution for practical applications.