基于GRU软测量与卡尔曼滤波的电池SOC快速估计  

Fast estimation of battery SOC based on GRU soft sensing and Kalman filter

作　　者：陈志宣 王浩 陆玲霞[1] 华思聪 和嘉睿 于淼[1] CHEN Zhixuan;WANG Hao;LU Lingxia;HUA Sicong;HE Jiarui;YU Miao(College of Electrical Engineering,Zhejiang University,Hangzhou Zhejiang 310027,China;Hangzhou Gold Electronic Equipment Inc.,Hangzhou Zhejiang 310023,China)

机构地区：[1]浙江大学电气工程学院,浙江杭州310027 [2]杭州高特电子设备股份有限公司,浙江杭州310023

出　　处：《电源技术》2025年第4期740-749,共10页Chinese Journal of Power Sources

基　　金：浙江省2024年度“尖兵”“领雁”研发计划项目[2024C01237(SD2)]。

摘　　要：锂离子电池的荷电状态(state of charge,SOC)在电池均衡、优化能量使用等方面具有重要作用。针对基于模型的SOC估计方法中状态空间方程非线性导致计算量大的问题,提出了使用门控循环单元(gated recurrent units,GRU)软测量SOC,并以此为观测量构建线性状态空间方程,进而使用卡尔曼滤波(Kalman filter,KF)估计SOC的方法。在随机驾驶循环工况下,所提出方法的SOC估计最大绝对误差为0.017,同时具有较快的估计速度。进一步研究发现,不同充放电倍率下电池模型的参数具有很大差异,导致基于模型的SOC估计方法在复杂情况下的估计精度较低,而所提出的GRU-KF方法因为不需要精确的电池模型,更能适应复杂的工况。The state of charge(SOC)of lithium-ion battery plays a crucial role in battery balancing,energy usage optimizing,etc.To address the issue of heavy computational burden caused by the nonlinearity of state-space equations in model-based SOC estimation methods,gated recurrent units(GRU)was used for soft sensing of SOC.Then linear state-space equations were constructed by using the measured SOC as an output variable and the Kalman filter(KF)was applied to estimate SOC.Under random driving cycles,the proposed method achieves a maximum absolute error of 0.017 in SOC estimation,while also offering fast estimation speed.Further study indicates that there are significant differences in the parameters of battery models under different charging and discharging rates,leading to lower SOC estimation accuracy of model-based methods in complex conditions.In contrast,the proposed GRU-KF method,due to its independence from precise battery models,demonstrates better adaptability to complex conditions.

关 键 词：锂离子电池 SOC估计 门控循环单元 软测量 卡尔曼滤波 

分 类 号：TM912.9[电气工程—电力电子与电力传动]