Collaborative framework of Transformer and LSTM for enhanced state-of-charge estimation in lithium-ion batteries

Accurately estimating the State of Charge (SOC) of a battery is crucial for advancing sustainable energy technologies, particularly in optimizing energy storage solutions and enabling seamless integration with renewable energy systems. Data-driven methods for SOC estimation, powered by advancements in artificial intelligence and machine learning, effectively handle the complex nonlinear relationships in charging and discharging processes. However, current models often face challenges in estimating SOC due to battery dynamics, high variability in operating conditions, and limited diverse training data, leading to significant errors and reduced generalization capabilities. This paper introduces a collaborative framework combining Transformer and Long Short-Term Memory (LSTM) models to enhance SOC estimation for lithium-ion batteries. By leveraging the strengths of Transformers in capturing long-term dependencies and the ability of LSTMs to model short-term patterns within sequential data, the proposed method achieves significant accuracy improvements, with a minimum Mean Absolute Error (MAE) reaching 1.11 % and Root Mean Square Error (RMSE) at 1.42 %. The model's accuracy under varying temperatures and dynamic driving conditions highlights its potential for real-world applications. Beyond lithium-ion batteries, this approach offers a scalable solution for various energy storage technologies, aiding in energy management optimization and addressing challenges in smart grid integration and renewable energy utilization.