Accurately predicting the remaining useful life of lithium-ion batteries is critical to battery health management systems. Aiming at the problems of low long-term prediction accuracy, unstable model output, and difficult key parameter selection, this paper proposes a self-adaptive differential evolution optimized monotonic echo state network prediction method. First, we analyze the life decay characteristics of Li-ion batteries and select appropriate indirect health indicators to replace the capacity based on the partial correlation coefficient analysis. Then use the self-adaptive differential evolution algorithm to optimize the free parameters of the monotonic echo state network to maintain the monotonic relationship between input and output. Finally, the remaining useful life indirect prediction model is established. This paper uses NASA Li-ion battery experimental data and independent experimental data to verify the feasibility, followed by the different starting points experiments and cut-off voltage experiments. The accuracy of the proposed method is compared with other commonly used artificial intelligence prediction algorithms. Experimental results prove that this method has high prediction accuracy and stable output.

准确预测锂离子电池的剩余使用寿命对于电池健康管理系统至关重要。针对长期预测精度低，模型输出不稳定，关键参数选择困难的问题，提出了一种自适应差分演化优化的单调回波状态网络预测方法。首先，我们通过部分相关系数分析，分析了锂离子电池的寿命衰减特性，并选择适当的间接健康指标来替代容量。然后使用自适应微分进化算法对单调回波状态网络的自由参数进行优化，以保持输入输出之间的单调关系。最后，建立剩余使用寿命间接预测模型。本文利用NASA锂离子电池的实验数据和独立的实验数据来验证可行性，然后进行不同的起点实验和截止电压实验。将该方法的准确性与其他常用的人工智能预测算法进行了比较。实验结果表明，该方法具有较高的预测精度和稳定的输出。