快速准确地获取锂离子电池的剩余使用寿命，对提高设备的可靠性有着重要意义。针对传统高斯过程回归（gaussian process regression, GPR）超参数寻优效果差，寻优困难，利用麻雀搜索算法（sparrow search algorithm, SSA）对高斯过程回归进行超参数优化，同时利用小波包分解（wavelet packet decomposition, WPD）降低数据集复杂度，提取相关信息，增加预测精度，提出了将小波包分解和高斯过程回归以及麻雀搜索算法相结合，建立剩余使用寿命（remaining useful life, RUL）预测模型。首先，等压降放电时间曲线作为间接健康因子，电池容量作为直接健康因子，利用Pearson系数验证二者的相关性。其次，利用小波包分解对直接健康因子与间接健康因子进行分解，提取出高频信号和低频信号并将这些信号分为训练集与测试集。然后，建立高斯过程回归模型，利用SSA对该模型进行超参数优化，分别对不同信号进行预测、叠加，实现剩余使用寿命的准确预测。最后，利用公开数据集进行验证。结果表明，本文提出的模型平均绝对误差不超过0.006 5、平均绝对百分比误差不超过0.005 2，均方根误差不超过0.007 8，拥有良好的预测精度和泛化性。

Quick and precise assessment of the remaining service life of lithium-ion batteries is considered to be of substantial importance for enhancing the reliability of equipment. Traditional Gaussian process regression (GPR) is found to exhibit a poor hyperparameter optimization effect and is deemed difficult to optimize. The sparrow search algorithm (SSA) was employed to enhance the optimization of hyperparameters within Gaussian process regression. Additionally, wavelet packet decomposition (WPD) was utilized to reduce the complexity of the dataset, extract relevant information, and improve prediction accuracy. A model for predicting the remaining useful life (RUL) was proposed, which combined wavelet packet decomposition, Gaussian process regression, and the sparrow search algorithm (SSA). Firstly, the equal voltage drop discharge time curve was used as the indirect health factor, while the battery capacity was used as the direct health factor. The correlation between the two was verified using the Pearson coefficient. Second, the wavelet packet decomposition was utilized to decompose both the direct and indirect health factors. High-frequency and low-frequency signals were extracted and divided into the training set and the test set. Then, a Gaussian process regression model was established. SSA was used to optimize the hyperparameters of this model. Different signals were predicted and superimposed respectively to achieve accurate prediction of the remaining service life. Finally, validation was carried out using the public dataset. The results demonstrate that the average absolute error of the model proposed in this paper is not exceeded by 0.006 5, the average absolute percentage error is not exceeded by 0.005 2, and the root-mean-square error is not exceeded by 0.007 8, being characterized by good prediction accuracy and generalizability.