The current lithium-ion battery remaining useful life (RUL) prediction techniques are mainly developed dependent on offline training data. The loaded current, temperature, and state of charge of lithium-ion batteries used for electric vehicles (EVs) change dramatically under the working conditions. Therefore, it is difficult to design acceleration aging tests of lithium-ion batteries under similar working conditions as those for EVs and to collect effective offline training data. To address this problem, this paper developed an RUL prediction method based on the Box-Cox transformation (BCT) and Monte Carlo (MC) simulation. This method can be implemented independent of offline training data. In the method, the BCT was used to transform the available capacity data and to construct a linear model between the transformed capacities and cycles. The constructed linear model using the BCT was extrapolated to predict the battery RUL, and the RUL prediction uncertainties were generated using the MC simulation. Experimental results showed that accurate and precise RULs were predicted with errors and standard deviations within, respectively, [-20, 10] cycles and [1.8, 7] cycles. If some offline training data are available, the method can reduce the required online training data and, thus, the acceleration aging test time of lithium-ion batteries. Experimental results showed that the acceleration time of the tested cells can be reduced by 70%-85% based on the developed method, which saved one to three months' acceleration test time compared to the particle filter method.

中文翻译：

---

利用 Box__ox 变换和蒙特卡罗模拟预测锂离子电池剩余使用寿命


目前的锂离子电池剩余使用寿命（RUL）预测技术主要是依赖离线训练数据开发的。电动汽车（EV）用锂离子电池在工作条件下的负载电流、温度和荷电状态会发生巨大变化。因此，设计与电动汽车类似工况下的锂离子电池加速老化测试并收集有效的离线训练数据是很困难的。针对这一问题，本文提出了一种基于Box-Cox变换（BCT）和蒙特卡罗（MC）模拟的RUL预测方法。该方法可以独立于离线训练数据来实现。该方法利用BCT对可用容量数据进行变换，并在变换后的容量和周期之间构建线性模型。使用 BCT 构建的线性模型外推来预测电池 RUL，并使用 MC 模拟生成 RUL 预测不确定性。实验结果表明，预测的 RUL 准确且精确，误差和标准偏差分别在 [-20, 10] 个周期和 [1.8, 7] 个周期内。如果有一些离线训练数据，该方法可以减少所需的在线训练数据，从而减少锂离子电池的加速老化测试时间。实验结果表明，基于所开发的方法，测试电池的加速时间可以减少70%-85%，与粒子过滤方法相比，节省了1到3个月的加速测试时间。