Capacity estimation for lithium-ion batteries based on heterogeneous stacking model with feature fusion

Accurately estimating battery capacity plays a crucial role in determining the State of Health (SOH) of lithium-ion batteries, which is essential for ensuring their safe operation and protection. This paper proposes a Stacking ensemble model based on feature fusion using Principal Component Analysis (PCA) for battery capacity estimation. Multiple health factors are extracted from the battery testing data, and use PCA to fuse the health factors to reduce the computational complexity of the model. In view of the performance difference of a single model on different datasets data sets, this paper proposes a new Stacking ensemble model that utilizes different model stacks to complement each other's strengths. The Stacking model improves the generalization ability and stability of the model on different datasets by using ridge regression to fuse three heterogeneous base models. This method is validated on the NASA battery dataset, and by comparing the errors of the base models and other ensemble methods across different training data ratios, the Stacking model has the smallest error across all datasets, it is demonstrated that the Stacking ensemble model has significant advantages in terms of accuracy and generalization ability. The capacity estimation results for the four datasets show that the Stacking model achieved error of less than 0.015Ah. The average absolute error and root mean square error of the dataset with the smallest error are 0.0025Ah and 0.0031Ah, respectively. The estimation results indicate that the Stacking ensemble model has higher accuracy and robustness in estimating battery capacity compared to other data-driven and ensemble methods.