LSTM and State Space Modeling for Predicting Lithium-Ion Battery Health and Ensuring Safety in Real-World Applications
Palavras-chave:
State of Health, State of Charge, Remaining Useful Life, State Space Model, Long Short-Term MemoryResumo
Accurate estimation of battery State of Charge (SOC), State of Health (SOH), and Remaining Useful Life (RUL) is essential for the safe and efficient operation of lithium-ion energy storage systems. Understanding and predicting the capacity and degradation behavior of individual battery cells and packs in real time is critical for reducing operational risk and extending service life. In real-world applications, such as electric vehicles, mobile electronics, and aerospace systems, undetected battery aging can lead to performance loss, sudden shutdowns, or even catastrophic thermal events.
The ability to anticipate failure modes through accurate SOH and RUL estimation enables manufacturers and users to implement timely maintenance, load balancing, and safe charging strategies. This not only increases energy efficiency and reliability but also plays a pivotal role in preventing accidents and ensuring user safety.
This study proposes a hybrid evaluation framework using two advanced sequence modeling approaches—Long Short-Term Memory (LSTM) networks and the recently introduced Mamba framework, a selective state space model—to predict key battery health indicators. The dataset consists of voltage, current, and temperature profiles collected from lithium-ion battery cycling experiments under variable load conditions.
LSTM networks are applied for their ability to learn short- and mid-range dependencies using gated memory units, while the Mamba model integrates state-space representations to encode long-range sequence information effectively. Both models are trained and evaluated using metrics such as Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) to assess predictive accuracy.
This work underscores the importance of integrating advanced machine learning models into battery management systems and demonstrates how intelligent diagnostics can support safer, longer-lasting energy storage technologies.
