The Remaining Useful Life Estimation of Lithium-ion Batteries Based on the HKA -ML-ELM Algorithm
The Remaining Useful Life Estimation of Lithium-ion Batteries Based on the HKA -ML-ELM Algorithm
复制标题
基于HKA-ML-ELM算法的锂离子电池剩余使用寿命估算
DOI：
10.20964/2019.08.44
复制
发表时间：
2019-08
影响因子：
1.5
通讯作者：
Hong Zhu
中科院分区：
文献类型：
--
作者：
Yanying Ma;Dongxu Shen;Lifeng Wu;Yong Guan;Hong Zhu
锂离子电池已成为许多电子设备的核心能源供应部件。准确预测锂离子电池的剩余使用寿命（RUL）对于电池管理和确保电子设备的可靠性具有重要意义。将极限学习机（ELM）算法应用于锂离子电池RUL的预测；然而，这种方法也存在一些缺点：(1)ELM的单隐层结构必然限制了其在高维数据中捕获有效特征的能力。（ii） ELM的输入权值和偏置是随机产生的，影响其预测精度。为了克服这些问题，本文提出了一种预测锂离子电池RUL的HKA-ML-ELM方法。首先，构建了一种新的多层ELM （ML-ELM）网络。通过在ML-ELM的最后一个单独的ELM中增加一个输入层，并随机选择这些输入节点与隐藏层部分连接，网络具有更高的鲁棒性，可以有效地防止过拟合。其次，采用启发式卡尔曼算法（HKA）对ML-ELM的输入权值和偏置参数进行优化，提高了预测精度；最后，对不同额定容量和不同放电电流的电池组进行了RUL预测实验。实验结果验证了该方法的有效性。与其他算法的比较表明，该方法具有更好的预测精度。
Lithium-ion batteries have become the core energy supply component for many electronic devices. An accurate prediction of the remaining useful life (RUL) of lithium-ion batteries is of great significance for battery management and ensuring the reliability of electronic devices. The extreme learning machine (ELM) algorithm has been applied to predict the RUL of lithium-ion batteries; however, there are some disadvantages in this method: (i). the single hidden layer structure of the ELM necessarily restricts its ability to capture effective features in high-dimensional data. (ii). the input weights and biases of the ELM are generated randomly, which affects its prediction accuracy. To overcome these problems, this paper proposes an HKA-ML-ELM method for predicting the RUL of lithium-ion batteries. First, a new multi-layer ELM (ML-ELM) network is constructed. By adding an input layer into the last individual ELM of the ML-ELM and implementing the random selection of these input nodes to partially connect with the hidden layer, the network has higher robustness and can effectively prevent over-fitting. Second, the heuristic Kalman algorithm (HKA) is used to optimize the input weights and biases parameters of the ML-ELM, which improves the prediction accuracy. Finally, RUL prediction experiments are carried out for battery packs with different rated capacities and different discharge currents. The experimental results verify the effectiveness of the proposed method. The comparisons with other algorithms show that the proposed method has better prediction accuracy.
DOI：
10.1007/978-1-4471-5188-3
发表时间：
2013-05
期刊：
--
影响因子：
--
作者：
R. Toscano
通讯作者：
R. Toscano