Real-Time SOC Estimation Using Task-Specific Deep Neural Networks for Battery Management

Accurate state of charge (SOC) estimation is essential to optimize lithium-ion battery performance, safety, and longevity in behind-the-meter storage (BTMS). Conventional methods like coulomb counting and Kalman filter face issues such as error accumulation and high computational complexity. A hybrid convolution neural network (CNN)-long short-term memory (LSTM) based SOC estimation is proposed in this paper to increase battery performance and efficiency. The spatial features are extracted by the CNN while long-term charge/discharge patterns are captured by the LSTM layers. The model achieves high accuracy by utilizing real-world battery data, which outperforms conventional methods. Explicit battery models are required for conventional methods, while the data-driven CNN-LSTM estimator generalizes the behavior well with minimal tuning. Its computational efficiency makes it suitable for real-time battery management systems (BMS) applications and BTMS implementations.