A Deep Learning Method with Ensemble Learning for Capacity Estimation of Lithium-ion Battery

Abstract

The accurate estimation of the capacity of lithium-ion batteries can help us better understand its working state. However, there are some intractable problems including poor adaptability and robustness in current data-driven models which affect their estimation accuracy. The advance in deep learning and ensemble learning introduces brand-new data-driven methods to solve this problem. This paper proposes a comprehensive model of lithium-ion capacity estimation involving deep learning and ensemble learning structure. In this model, based on the effectiveness of deep learning in feature extraction, an autoencoder (AE) is used to extract features, and a deep neural network (DNN) is adopted for the estimation of lithium-ion capacity. On the other side, we take a random forest (RF) method to ensure the stability and robustness of the capacity estimation model. These two models were integrated to build an ensemble model called EADNN-RF. The ensemble model is applied to a dataset of lithium-ion batteries via the accelerated test taken by NASA, and the prediction effect of the model is compared with ADNN and RF. The results of the controlled trial demonstrate that the proposed ADNN-ERF method which is more accurate and robust than the other similar data-driven methods.