Accurate capacity estimation is crucial and challenging for guaranteeing safety and durability of Li-ion batteries. This work proposes a data-fusion-model method to estimate battery capacity using the partial charging curve. First, during the charging process, two representative battery ageing features are extracted from the partial incremental capacity curve smoothed by Locally Weighted Scatterplot Smoothing (LOWESS). Second, the dual Gaussian process regressions (GPRs) are employed to establish a data-driven battery ageing state-space representation, which takes battery capacity as the state variable and takes two ageing features as the input variables. Third, combining with the GPR-based battery state-space representation, Particle Filter (PF) is introduced to suppress the measurement noises and a Gaussian Process Particle Filter (GPPF) is built to estimate battery capacity. Meanwhile, the output capacity is fed back to the GPPF to update the battery ageing state-space representation. Finally, ageing experiments are conducted to validate the effectiveness of the proposed method. Meanwhile, GPR is implemented with the cycle and two ageing features for capacity estimation as a contrast. The results show that the proposed GPPF method can provide more accurate and robustness battery estimation results than GPR.

准确的容量估算对于保证锂离子电池的安全性和耐用性至关重要且具有挑战性。这项工作提出了一种数据融合模型方法，可以使用部分充电曲线估算电池容量。首先，在充电过程中，通过局部加权散点图平滑（LOWESS）平滑的部分增量容量曲线提取了两个代表性的电池老化特征。其次，采用双重高斯过程回归（GPR）来建立数据驱动的电池老化状态空间表示，该表示将电池容量作为状态变量，并将两个老化特征作为输入变量。第三，结合基于GPR的电池状态空间表示，引入了粒子滤波器（PF）以抑制测量噪声，并建立了高斯过程粒子滤波器（GPPF）来估计电池容量。同时，将输出容量反馈到GPPF以更新电池老化状态空间表示。最后，进行了老化实验以验证该方法的有效性。同时，GPR以周期和两个老化功能来进行容量估算作为对比。结果表明，提出的GPPF方法比GPR可以提供更准确，更鲁棒的电池估计结果。对比而言，GPR是以周期和两个老化功能来进行容量估算的。结果表明，提出的GPPF方法比GPR可以提供更准确，更鲁棒的电池估计结果。GPR是通过周期和两个老化功能（用于容量估算）进行对比来实现的。结果表明，提出的GPPF方法比GPR可以提供更准确，更鲁棒的电池估计结果。