A method for estimating and predicting the health state of lithium battery

The health status of lithium-ion batteries serves as a crucial indicator in battery management systems, reflecting the aging characteristics of batteries. When the health status falls below 80%, regulations mandate decommissioning. To monitor this, it is essential to identify sensitive indicators from parameters such as voltage, current, and temperature. Using the Stanford-MIT-Toyota Research Institute battery dataset, this study examines batteries with over 1000 cycles and includes five unique charge-discharge strategies. Health indicators are extracted from voltage, current, and temperature profiles within a state of charge range of 20% to 80% during routine operation and maintenance of the energy storage station. Ridge regression is applied to develop a model for estimating health status. To expand the training dataset, both the actual and estimated health status values are combined. A GRU model is then employed to predict the health state. Simulation results indicate that the proposed indicators provide an effective measure of lithium battery health, with ridge regression errors below 2.5% and GRU prediction errors under 4%. This methodology demonstrates stability, high accuracy, and strong generalization potential.