锂离子电池在储能电站中为消纳可再生能源作出了重要贡献，其运行的稳定性和可靠性受到了研究人员的持续关注。为了解决锂离子电池容量及剩余寿命的预测和抑制测量过程中因各种外界因素引起的噪声，提出了一种基于改进的集成经验模态分解MEEMD（Modified Ensemble Empirical Mode Decomposition）去噪和经贝叶斯优化的高斯过程回归BO-GPR（Gaussian process regression optimized by Bayesian optimization algorithm）的锂离子电池容量及剩余寿命预测方法。利用MEEMD方法识别并去除原始测量数据中的噪声分量。利用BO-GPR方法预测锂离子电池容量及剩余寿命，其中贝叶斯优化方法对高斯过程回归的部分超参数进行了进一步寻优。文章基于美国国家航空航天局研究中心提供的锂离子电池测量数据进行了预测实验，结果表明，该方法能够有效去除噪声信号，选取的协方差函数和超参数组合达成的预测效果优于初始GPR模型，证明了其有效性。
英文摘要:
Lithium-ion battery, whose operational robustness and reliability have increasingly drawn attention of researchers, plays a crucial role in consuming renewable energies for energy storage power station. Consequently, a prediction approach, based on modified ensemble empirical mode decomposition (MEEMD) and gaussian process regression optimized by Bayesian algorithm (BO-GPR), is presented in the hope of prognosticating the capacity and remaining useful life (RUL) of battery and suppressing the noise invited by some external causes. Firstly, MEEMD is employed to recognize and filter the noise components in the raw measured data. Then, BO-GPR predicts the capacity and RUL of lithium-ion battery, where the Bayesian optimization method further searches for the optimal values of some hyperparameter. Based on the measured dataset of lithium-ion battery capacity provided by NASA Research Center, the prediction result is conducted. The results show that the proposed method could effectively erase noise signals, and the forecast results founded on the chosen covariance function and hyperparameters are better than those of the original GPR model, which verifies its superiority.