基于优化高斯过程回归算法的锂离子电池可用容量估算

为了解决应用数据驱动算法估算锂离子电池可用容量时存在的电池老化特征提取不准确、可用容量衰退趋势跟踪精度低及模型要求训练数据量大等问题,提出一种基于优化高斯过程回归算法的锂离子电池可用容量估算方法,实现锂离子电池强非线性全衰退过程可用容量精确估算。首先,提取电池表面平均温度、容量增量曲线峰值及峰值对应电压作为表征电池老化状态的健康因子,通过灰色关联度分析法和熵权值法对所选健康因子进行合理性评估;然后,用2个单一核函数构造高斯过程回归算法复合核函数,并利用鲸鱼优化算法完成复合核函数的参数寻优,基于优化后的高斯过程回归模型实现锂离子电池可用容量估算;最后,通过对比不同核参数寻优算法,证明鲸鱼优化算法在参数寻优方面的先进性,并通过与传统的高斯过程回归、支持向量机、径向基神经网络等机器学习算法进行可用容量估算对比,证明模型的有效性。研究结果表明:基于复合核函数和鲸鱼优化算法参数寻优可以有效改善高斯过程回归模型性能,所建立的优化高斯过程回归模型能够基于较少训练数据实现电池容量的精确估算,并能够有效追踪锂离子电池非线性长周期衰退趋势;对不同电池数据也具备较好的自适应能力,可用容量估算最大误差低于1.56%。

Available Capacity Estimation of Lithium-ion Batteries Based on Optimized Gaussian Process Regression

To address the problems including inaccurate feature extraction,low-accuracy tracking of the available capacity degradation trend,and the large amount of training data required by the model during the current data-driven capacity estimation process,an optimized Gaussian process regression (GPR) algorithm is developed to estimate the available capacity with high efficiency and accurate estimation performance throughout the nonlinear degradation process.First,the average temperature measured on the battery surface,peak value of the capacity increment curve,and peak corresponding voltage are extracted as health factors characterizing the aging state.Moreover,the rationality of the selected health factors is evaluated using grey relation analysis (GRA) and the entropy weighting method.Then,a composite kernel function of the GPR algorithm is constructed using two single kernel functions;the whale optimization algorithm (WOA) is applied to complete the parameter search optimization of the composite kernel function,and the available capacity estimation of the lithium-ion battery is attained based on the optimized GPR model.Finally,the superiority of the WOA in terms of optimization efficiency is demonstrated by comparing different kernel parameter-seeking algorithms,and the validity of the model is demonstrated by comparing it with the capacity estimation performance based on traditional GPR,supported vector machine (SVM),and radial bias function (RBF) neural networks.The results show that the performance of the GPR model can be effectively improved based on the composite kernel function and WOA algorithm parameter search and that the optimized GPR model developed in this study can achieve accurate battery capacity estimation based on less training data and efficiently track the long-term nonlinear degradation trend of lithium-ion batteries.Meanwhile,the algorithm shows preferable adaptability to different battery data with a maximum error of less than 1.56% in capacity estimation.