基于BiGRU和PF的锂电池SOC估计

Estimation of Lithium Battery SOC Based on BiGRU and PF

提出了一种基于双向门控循环神经网络(Bidirectional Gate Recurrent Unit,BiGRU)和粒子滤波(Particle Filter,PF)相结合的方法,对电池的荷电状态(State of Charge,SOC)进行估计。利用BiGRU网络,根据可测量的电压(V)、电流(I)、温度(T)等信息学习锂电池内部的动态特性。由于电池的平台期特性导致BiGRU网络在捕捉测量数据与电池内部特性时会有一些波动,因此采用PF对BiGRU网络的输出进行滤波,让最终的估计更加稳定。对所提出的方法在FUDS和US06工况下进行验证,实验结果显示BiGRU能很好地捕捉到电池的动态特性,从而避免了繁杂的建模过程,结合PF,进一步提升了模型估计精度和鲁棒性。

A method based on a combination of Bidirectional Gate Recurrent Unit(BiGRU)and Particle Filter(PF)is proposed to estimate the state of charge(SOC)of the battery.The Bi-GRU network is used to learn the dynamic characteristics inside the lithium battery based on measurable voltage(V),current(I),temperature(T),and other information.Due to the platform period characteristics of the battery,the BiGRU network will have some fluctuations when capturing the measurement data and the internal characteristics of the battery,so the output of the BiGRU network is filtered by PF to make the final estimate more stable.The proposed method is verified under the conditions of FUDS and US06,and the experimental results show that BiGRU could well capture the dynamic characteristics of the battery,thus avoiding the complicated modeling process.Combined with PF,the model estimation accuracy and robustness are further improved.