Co-Estimation of Lithium Battery States Based on BP Neural Network

Introduction: Energy storage systems and electric vehicles rely on lithium batteries as their core component. The performance and safety of these batteries are directly linked to the reliability of the equipment and the safety of the user. Consequently, real-time monitoring of lithium battery status is crucial for preventing failures and optimizing energy usage. Electric vehicles help reduce air pollution by eliminating emissions, while improved batteries offer longer range and faster charging. This paper proposes a simple, efficient, and unified framework for co-estimating the State of Charge (SOC), State of Health (SOH), and Remaining Useful Life (RUL) of lithium- ion batteries using a Backpropagation (BP) neural network. Methods: Instead of relying on complex Equivalent Circuit Models (ECMs), the BP neural network utilizes measurable features (voltage, current, temperature, and three time-domain parameters) to estimate SOC, SOH, and RUL. SOC is predicted using voltage and current; full charge capacity is inferred using temporal features derived from SOH and RUL. Results: The model is trained and validated on NASA battery datasets B0005 and B0006. SOC prediction achieves an average RMSE of 0.473, while SOH prediction achieves 0.161. Discussion: Comparisons with EKF, BiLSTM, and ELM models demonstrate that the BP network achieves comparable or better performance at a lower computational cost. Conclusion: The method provides a fast, accurate, and practical solution for real-time battery monitoring. It enables embedded system deployment with reduced computational resources. Future work will incorporate different battery chemistries an.d environmental conditions to improve generalizability.