Abstract: When Bayesian method updates the model parameters offline,the historical degradation data isn't well applied for parameter estimation.For this problem,a new method based on sequential Bayesian was proposed to update parameters online.A nonlinear Wiener process degradation model was constructed for the lithium-ion battery capacity degradation path under variable working conditions,and the Maximum Likelihood Estimation(MLE)was used to estimate the model parameters at the initial time.Followed by it,the drift coefficients in the degradation model were updated online based on the sequence Bayesian updating method.Then,the probability density function of the lithium-ion battery's Remaining Useful Life(RUL)was derived for prediction.The proposed model was applied and demonstrated by the lithium-ion battery dataset in various conditions.Moreover,the results showed the prediction accuracy of RUL based on the proposed model was higher than those obtained from the degradation models based on the power exponent or linear function for the sequential Bayesian method realized the real-time update of parameter estimation.

Key words: Wiener process, maximum likelihood estimation, sequential Bayesian updating, remaining useful life, non-linear degradation

摘要： 针对贝叶斯方法在更新模型参数时无法充分利用历史退化数据的问题,提出基于序列贝叶斯的在线更新方法实时估计锂电池退化模型参数。构建基于指数函数的非线性维纳退化模型描述变工况下锂电池容量的退化路径,并采用最大似然估计法估计初始时刻的模型参数;利用实时容量监测数据,基于序列贝叶斯更新方法在线更新退化模型中的漂移系数;推导锂电池剩余寿命的概率密度函数并预测剩余寿命。通过对不同工况下的锂电池退化数据进行实例验证表明,与基于幂指数和线性函数的退化模型相比,由于序列贝叶斯方法能够实时更新锂电池非线性退化模型参数,采用所提模型预测的剩余寿命精度更高。

关键词: 维纳过程, 最大似然估计, 序列贝叶斯更新, 剩余寿命, 非线性退化