Application of Attention-Based CNN-BiLSTM Model in Lithium-Ion Battery SOH Prediction

Lithium-ion batteries have broad applications in power operation and maintenance scenarios such as inspection drones and robots owing to their high energy density and low weight, and environmental friendliness. However, as usage time increases, battery performance gradually degrades, posing safety risks and placing higher demands on the battery management system (BMS). State of Health (SOH) estimation is a key technology for real-time condition monitoring and early warning. This paper proposes an attention-enhanced CNN-BiLSTM hybrid framework for assessing battery SOH. A CNN is utilized for extracting representations, and a BiLSTM is adopted to model temporal dependencies, and integrates an attention mechanism to enhance responsiveness to local fluctuations. The evaluation results highlight that the proposed model achieves estimation errors below 0.01 even when trained with only the first 40% of the battery cycle data, demonstrating excellent prediction accuracy and generalization ability, making it suitable for engineering scenarios with high battery health requirements.