An interactive prognostics framework for lithium-ion battery remaining useful life based on neural networks and statistical processes

Accurately predicting the remaining useful life of lithium-ion batteries is crucial for ensuring the operational safety and reliability of energy systems. However, conventional models assume constant degradation patterns under varying operational conditions and lack an interactive mechanism to update model structure or parameters using newly acquired battery data. To address this issue, this paper proposes an interactive prognostics framework integrating a deep learning model, a statistical process, and an interactive mechanism to jointly predict the remaining useful life of lithium-ion batteries. A deep learning model is developed based on a multi-scale cross-attention network, which employs a dual-channel network integrated through a cross-attention mechanism to estimate the capacity of the lithium-ion batteries. Meanwhile, the capacity degradation of lithium-ion batteries is modeled as a statistical process governed by a switching state-space model, whose parameters and capacity states are dynamically updated via an expectation-maximization algorithm. Afterward, the interactive prognostics mechanism iteratively integrates capacity estimation from both the multi-scale cross-attention network and switching state-space model models. Each model’s capacity estimates are used to update the parameters of the other, to recursively update the remaining useful life prognostics of lithium-ion batteries. The proposed approach is validated by two case studies, which demonstrate the effectiveness and superiority of the presented model.