An LSTM neural network combined with a binary search algorithm improved battery state-of-power estimation accuracy and robustness by at least 3x compared to ECM and EM approaches.
A novel machine learning framework combining LSTM and binary search improves battery state-of-power estimation accuracy by at least 3x compared to traditional models, offering a strong candidate for next-generation battery management systems.
Accurate battery State-of-Power (SOP) estimation is essential for ensuring the safety, reliability, and performance of lithium-ion battery-powered systems. Traditional estimation approaches based on lookup tables or equivalent circuit models (ECMs) face challenges under dynamic conditions, while electrochemical models (EMs) offer insight into underlying chemical processes but are computationally intensive and tricky to parameterize. This work proposes a novel SOP estimation framework that combines a long short-term memory (LSTM) neural network with a numerical binary search algorithm. The LSTM-based battery model is trained using multi-temperature drive cycle data and optimized through cross-validation to achieve low voltage prediction error. The binary search algorithm utilizes the LSTM model to apply power pulses iteratively, estimating the maximum power that can be delivered without violating voltage and current constraints. The proposed method is extensively validated across a wide range of temperatures, state-of-charge (SOC) levels, and power durations, demonstrating at least a 3x improvement in accuracy and robustness compared to ECM and EM approaches. Finally, real-time deployment of the algorithm on a Jetson Nano platform confirms its computational efficiency and practical viability via hardware-in-the-loop (HIL) testing. The results position this framework as a strong candidate for next-generation battery management systems (BMS) in electric vehicles and other applications.
Chen et al. (Mon,) reported a other. LSTM neural network with numerical binary search algorithm vs. Equivalent circuit models (ECMs) and electrochemical models (EMs) was evaluated on State-of-Power (SOP) estimation accuracy and robustness. An LSTM neural network combined with a binary search algorithm improved battery state-of-power estimation accuracy and robustness by at least 3x compared to ECM and EM approaches.