A deep learning-based power battery capacity estimation model assisted with feature vector construction

This study aims to answer the key question on how to detect the capacity degradation of the power batteries in electric vehicles (EVS) with emerging intelligent methods. Focusing on the random and incomplete charging process in actual EVs usage, the new estimation method only needs a specific charging current segment rather than data from the whole charging process for model training and estimation. Given that, this paper proposes a novel feature vector construction method to extract the key features from the current segment firstly. On this basis, a fully connected neural network is established to recognize the constructed feature vectors and obtain the mapping between the charging information and capacity degradation. Battery cycling data measured from 718650Li-ion cells was utilized for model training and performance verification. Compared with the traditional data-driven capacity estimation methods proposed previously, the proposed method demonstrates the great superiority in terms of the method applicability and estimation accuracy for online capacity estimation in actual battery usage.

I. Introduction

Electrified transportation is developing rapidly worldwide, which greatly reduces greenhouse gas emissions and alleviates global warming. As the main energy storage device, Lithium-ion batteries are widely used to store electricity for electric vehicles (EVs). For the power batteries in EVs, capacity degradation is inevitable during operation and storage due to the growth of solid electrolyte interface (SEI) layer and the loss of available lithium inventory [1], which is one of the most significant reasons for battery failure. However, it is difficult to measure or calculate the capacity directly during operation since only a few measurable signal information (i.e. current, voltage, temperature, etc.) can be obtained and cannot be used for capacity estimation easily. For this issue, a variety of capacity degradation estimation methods are proposed, which can be roughly divided into model-based methods and data-driven methods.