Lithium-Ion Battery SOH Estimation Based on a Long Short-Term Memory Model Using Short History Data

Accurate estimation of the state-of-health (SOH) is essential in prognostics and health management for Battery management system in vehicle application. Algorithms using short-term data from flexible voltage ranges are gaining significant attention since partial cycling is a common case in real-world applications. To this end, a data-driven model using short-term charging history is proposed. The model integrates Incremental capacity analysis curve classification with time series forecasting based on long–short term memory for SOH estimation with a short state of charge (SOC) variation. Three datasets with different cell chemistries and degradation trajectories are used for validation. Results show that the proposed model achieves accurate SOH estimation with the mean absolute error and root mean squared error between 1% and 2%. The model is highlighted by its ability to process inputs from flexible voltage ranges and to provide accurate SOH estimation with SOC changes of less than 10%. The model can also be adapted to different cell chemistries and aging behaviors.