Stage of Charge Estimation of Lithium-Ion Battery Packs Based on Improved Cubature Kalman Filter With Long Short-Term Memory Model

Xing Shu,Zheng Chen,Yuanjian Zhang,Yonggang Liu,Guang Li,Shiquan Shen

doi:10.1109/tte.2020.3041757

Abstract

Accurate estimation of the state of charge (SOC) of lithium-ion battery packs remains challenging due to inconsistencies among battery cells. To achieve precise SOC estimation of battery packs, first, a long short-term memory (LSTM) recurrent neural network (RNN)-based model is constructed to characterize the battery electrical performance, and a rolling learning method is proposed to update the model parameters for improving the model accuracy. Then, an improved square root-cubature Kalman filter (SRCKF) is designed together with the multi-innovation technique to estimate the battery cell's SOC. Next, to cope with inconsistencies among battery cells, the SOC estimation values from the maximum and minimum cells are combined with a smoothing method to estimate the pack SOC. The robustness and accuracy of the proposed battery model and the cell SOC estimation method are verified by exerting the experimental validation under time-varying temperature conditions. Finally, real operation data are collected from an electric-scooter (ES) monitoring platform to further validate the generalization of the designed pack SOC estimation algorithm. The experimental results manifest that the SOC estimation error can be limited to 2% after convergence.

Highlights

Transportation electrification represents a promising solution to mitigate greenhouse gas (GHG) emission and environmental pollution; and this provides favorable opportunities for prompting development of electric vehicles (EVs) and electric-scooters (ESs) [1]
This paper presents a data-driven battery model based on long short-term memory (LSTM) recurrent neural network (RNN) to accurately simulate battery characteristics, and a rolling learning method is developed to improve the simulation precision and adapt to environment variation
An improved SRCKF based state of charge (SOC) prediction algorithm is proposed, which is fully substantiated in the case of time-varying temperature environment by comparing with the commonly used SOC estimation algorithms

Summary

INTRODUCTION

Transportation electrification represents a promising solution to mitigate greenhouse gas (GHG) emission and environmental pollution; and this provides favorable opportunities for prompting development of electric vehicles (EVs) and electric-scooters (ESs) [1]. In [30], the second-order ECM and simple resistance model are respectively established as the cell mean model and difference model, and the EKF is introduced to estimate the mean and difference SOC This approach needs to build two battery models, and the number of parameters waiting to be identified increase. Three original contributions of this study are briefly summarized as follows: 1) A fusion method combining the LSTM network and the improved SRCKF algorithm is developed to effectively model the battery’s electrical characteristics and estimate the cell SOC. When the LSTM is applied online, the battery operation data will be continuously collected by the central monitoring platform, and the rolling learning method is employed to learn and update the model parameters under different working conditions. After the battery model is constructed by the LSTM network, the SOC estimation will be conducted based on reliable simulation of battery electrical characteristics

Cell SOC Estimation

PERFORMANCE EVALUATION OF BATTERY CELL

Parameter Selection

Data Acquisition

Findings

CONCLUSION