Abstract: To solve the problems in the Remaining Useful Life(RUL)prediction of lithium-ion batteries,such as the difficulty in measuring the actual capacity accurately and the influence of noise information on the performance of the algorithm,a prediction method based on the Denoising Autoencoder(DAE)and Broad Learning System(BLS)was proposed.Multiple Health Indicators(HI)highly correlated with battery degradation were extracted from battery charge-discharge curves,and training samples were prepared using sliding time windows.Then,the preprocessed samples were input into BLS to predict the battery RUL,and the optimal model was obtained by adjusting the window size and model parameters.Finally,the MIT-Stanford battery degradation dataset was used to verify the effectiveness of the proposed method.Compared with the existing prediction methods,this method has a better performance in the prediction accuracy.

Key words: lithium-ion batteries, remaining useful life, health indicators, denoising autoencoder, broad learning system

摘要： 为解决锂离子电池剩余使用寿命(RUL)预测中存在的实际容量难以准确测量、噪声信息影响算法性能等诸多问题,提出一种基于去噪自编码器(DAE)和宽度学习系统(BLS)相结合的预测方法。首先,从电池充放电曲线中提取多个与电池退化高度相关的健康因子(HI),并使用滑动时间窗口制备训练样本。其次,将样本输入DAE进行去噪处理。然后,将经过处理的样本输入BLS,预测电池RUL,并通过调整窗口大小和模型参数,得到最优模型。最后,利用MIT-Stanford电池退化数据集验证该方法的有效性。实验结果表明,相比于已有预测方法,所提方法在预测精度上具有更好的表现。

关键词: 锂离子电池, 剩余使用寿命, 健康因子, 去噪自编码器, 宽度学习系统