Abstract: This paper addresses the challenges of measuring the state-of-health (SOH) of lithium-ion batteries through direct features, which often suffer from poor prediction accuracy and insufficient generalization. To overcome these issues, we propose a integrating network prediction model suitable for indirect feature data. Our approach involves a comprehensive process for predicting the state-of-health from data to model using indirect methods. First, we extract potential indirect data from relevant data sets and employ feature reconstruction techniques to construct indirect health indicator (HI). We then apply the variational mode decomposition(VMD) algorithm to decompose these indirect HI, such as time and temperature, and verify their effectiveness through correlation analysis. To accommodate the computational power of prediction devices, the scale and characteristics of feature data in the indirect prediction mode, and the individual performance of the model, we construct the VMD-CNN-AttBiGRU integrating neural network for state-of-health prediction. Finally, the validated feature data is used to verify the lithium-ion battery health state-of-health prediction in the indirect mode. Through a two-dimensional comparative analysis, our model achieves high prediction accuracy, demonstrating the effectiveness of the proposed indirect prediction model.