摘要
准确估计锂离子电池荷电状态(SOC)对于突破电动汽车发展瓶颈,推动电动汽车商业化至关重要。针对动力电池模型参数辨识问题,提出基于遗忘因子的递推最小二乘法(FRLS)的模型参数在线识别方法。实时测量动力电池电流和电压数据,在线辨识模型参数并实时更新,实时反映电池内部参数的变化过程,对电池动态特性进行实时模拟。针对容积卡尔曼(CKF)滤波过程中对噪声敏感的问题,提出一种基于随机加权思想的自适应容积卡尔曼滤波(ARWCKF)方法。相比于常规CKF容积点权值始终不变,通过引入随机加权因子,自适应调整容积点权值并对系统噪声、状态向量及观测向量进行预测,抑制系统噪声对状态估计的干扰,避免因容积点权重值固定所带来的误差。针对CKF算法在容积点计算过程中由于状态方差矩阵失去正定性导致的平方根分解无法使用的问题,提出基于奇异值分解的容积点计算方法,克服由于先验协方差矩阵负定性变化而导致的滤波精度下降等问题,并进行多种工况、温度下不同SOC初值的对比验证。结果表明:所提出的基于遗忘因子的递推最小二乘法的在线参数辨识及ARWCKF滤波方法具备良好的估计精度及收敛能力,最大电压估计误差不超过40 mV,SOC估计误差不超过1%。
Accurate state of charge(SOC)estimations of lithium-ion batteries are crucial for breaking the bottleneck of electric vehicle development and promoting the commercialization of electric vehicles.This study conducted the following research on the SOC estimations of lithium-ion batteries:Aiming at the parameter identification problem of battery models,an online identification method for model parameters based on the recursive least squares method of forgetting factor(FRLS)was proposed.The current and voltage data were measured online,and the model parameters were identified online and updated in real-time to realize an estimation of the dynamic characteristics of a battery.For the problem of noise sensitivity in cubature Kalman filtering,an adaptive cubature Kalman filter method based on random weighting(ARWCKF)was proposed.The random weighting factor was introduced to adaptively adjust the cubature point weight and predict the system noise,state vector,and observation vector,which restrained the disturbances of system noise on the state estimation and avoided the error caused by the fixed weight value of the cubature point.As Cholesky decomposition cannot be used due to the loss of positive definiteness of the state variance matrix in the cubature point calculation process,a cubature point calculation method based on singular value decomposition was proposed to overcome the problems of filtering accuracy caused by negative qualitative changes of the a priori covariance matrix.A comparison of different initial SOC under various working conditions and temperatures was performed to verify this approach.The results indicate that the online parameter identifications based on the recursive least squares method and ARWCKF filtering have adequate estimation accuracies and fast convergence ability.Furthermore,the voltage estimation error does not exceed 40 mV and the SOC estimation error does not exceed 1%.
作者
马建
张大禹
赵轩
张凯
MA Jian;ZHANG Da-yu;ZHAO Xuan;ZHANG Kai(School of Automobile,Chang'an University,Xi'an 710064,Shaanxi,China)
出处
《中国公路学报》
EI
CAS
CSCD
北大核心
2019年第11期234-244,共11页
China Journal of Highway and Transport
基金
国家重点研发计划项目(2018YEB1600700)
陕西省重点产业创新链(群)项目(2019ZDLGY15-01,2018ZDCXL-GY-05-03-01)
中国博士后基金特别资助项目(2018T111006)
中央高校基本科研业务费专项资金项目(310822173201)
关键词
汽车工程
SOC
容积卡尔曼
锂离子电池
随机加权
奇异值分解
automotive engineering
SOC
cubature Kalman filter
lithium-ion battery
random weighted
singular value decomposition
