锂离子电池健康状态(SOH)的在线估计对电池管理系统的安全稳定运行至关重要.为克服传统基于数据驱动的锂离子电池SOH估计方法训练时间长、计算量大、调试过程复杂的问题,提出一种基于融合健康因子和集成极限学习机的锂离子电池SOH估计方法.该方法通过 dQ/dV 和dT/dV曲线分析,筛选出与电池SOH相关性较高的数据区间进行多维健康特征提取,并对其进行主成分分析降维处理得到间接健康因子;利用极限学习机的随机学习算法建立间接健康因子和SOH之间的非线性映射关系.在此基础上,针对单一模型输出不稳定的特点,提出一种集成极限学习机模型,通过对估计结果设置可信度评价规则剔除单一极限学习机不可靠的输出,从而提高锂离子电池SOH的估计精度.使用NASA和牛津大学的锂离子电池老化数据集对该方法进行验证,结果表明该方法的平均绝对百分比误差小于1%,具有较高的准确性和可靠性.

Online estimation of the state of health (SOH) of lithium-ion batteries (LIB) is crucial for the security and stability operation of battery management systems. In order to overcome the problem such as long training time, large amount of computation, and complex debugging process of the LIB SOH estimation methods based on traditional data-driven, an LIB SOH estimation method based on fusion health factor (HF) and integrated extreme learning machine is proposed. The interval data with a high correlation with the SOH was found by analyzing the dQ/dV and dT/dV curves of the battery. Multi-dimensional HFs are extracted from the interval data, and the indirect HF are obtained by principal component analysis. The stochastic learning algorithm of extreme learning machine is used to establish the nonlinear mapping relationship between indirect HF and SOH. Considering the unstable output of a single model, an integrated extreme learning machine model is proposed. The unreliable output is eliminated by setting credibility evaluation rules for the estimation results, and the estimation accuracy of the model is improved. Finally, the method proposed in this paper is validated using the NASA LIB aging dataset and the LIB aging dataset of Oxford University. The results show that the average absolute percentage error of SOH estimation method proposed is less than 1%, and it has a high accuracy and reliability.