State of health estimation for lithium-ion batteries based on hybrid attention and deep learning
State of health estimation for lithium-ion batteries based on hybrid attention and deep learning
复制标题
DOI：
10.1016/j.ress.2022.109066
复制
发表时间：
2022-12
期刊：
影响因子：
--
通讯作者：
Hongqian Zhao;Zheng Chen;Xing Shu;Jiangwei Shen;Z. Lei;Yuanjian Zhang
中科院分区：
文献类型：
--
作者：
Hongqian Zhao;Zheng Chen;Xing Shu;Jiangwei Shen;Z. Lei;Yuanjian Zhang
准确评估锂离子电池的健康状态是电动汽车可靠、安全运行的必要条件。本研究提出了一种用于锂离子电池健康状态预测的混合注意力和深度学习方法。首先，根据充电数据计算出温差曲线，然后用卡尔曼滤波进行平滑。接下来，从温差曲线中提取与容量退化相关的健康特征，以表征温度与老化之间的关系。然后，结合卷积神经网络、门控递归单元递归神经网络和注意机制的优点，建立了一种混合注意力和深度学习模型来预测电池的健康状态。通过与11种主流预测方法的比较，验证了该方法的预测性能。在不提取高度相关的健康特征的情况下，所有的估计误差都可以保持在1.3%以内，说明了所开发的健康状态估计方法具有良好的准确性和可靠性。另外，实验结果验证了该算法对电池不一致性具有较好的鲁棒性。
Accurate state of health estimation of lithium-ion batteries is imperative for reliable and safe operations of electric vehicles. This study presents a hybrid attention and deep learning method for state of health prediction of lithium-ion batteries. First, the temperature difference curves are calculated from the charging data and subsequently smoothed by the Kalman filter. Next, the health features related to capacity degradation are extracted from the differential temperature curves to characterize the relationship between temperature and aging. Then, a hybrid attention and deep learning model integrating the strengths of convolutional neural network, gated recurrent unit recurrent neural network and attention mechanism is developed to forecast the battery's state of health. The superior prediction performance of the proposed method is verified by comparing with eleven mainstream methods. All the estimation errors can be maintained within 1.3% without extracting highly correlated health features, illustrating the promising accuracy and reliability of the developed state of health estimation method. In addition, the results validate that the proposed algorithm can achieve satisfied robustness to battery inconsistency.