36 / 2025-03-28 13:55:30

Adaptive Mamba-LSTM Hybrid Network with Joint State Estimation and Fault Diagnosis for Lithium Batteries

Mamba network, Adaptive architecture search, Lithium battery, State of charge, Fault diagnosis

Accurate State of Charge (SOC) estimation and fault diagnosis are crucial for the safe operation of lithium-ion battery management systems (BMS). However, traditional methods struggle to achieve joint estimation of SOC and fault states under faulty conditions, with additional challenges in parameter optimization. To address these issues, this paper proposes a Mamba-LSTM multi-task learning model based on Adaptive Architecture Search (AAS). The model first initializes the search space of network structural parameters, then employs differentiable gradient propagation to achieve automatic parameter optimization, ultimately constructing a hybrid architecture that integrates Mamba's selective feature extraction with LSTM's temporal modeling capabilities. Specifically, the Mamba module captures global dependencies and filters out noise interference through the state space model, while the LSTM network focuses on extracting local temporal features. The model adopts a dual-branch output structure to simultaneously generate SOC estimates and fault classification probabilities. Finally, the feasibility of the proposed method is validated using a lithium-ion battery dataset containing various fault states.