Abstract: In order to solve the problem of insufficient prediction accuracy of state of health (SOH) of lithium batteries, a model based on variational mode decomposition and the hippopotamus algorithm optimized long short-term memory (VMD-HO-LSTM) neural network was proposed to predict the remaining life of lithium batteries. Firstly, in order to eliminate the false capacity signal of lithium battery, the variational mode decomposition (VMD) method was used to decompose the capacity of lithium battery, and the intrinsic modal component (IMF) was obtained and reconstructed. Secondly, the logistic mapping and adaptive learning rate were integrated into the hippopotamus optimization (HO) algorithm to avoid the iterative process falling into the local optimum. And the improved hippopotamus algorithm was used to optimize the network parameters of long short-term memory (LSTM), then an improved HO-LSTM model was established. Finally, based on the improved HO-LSTM model, the SOH prediction of lithium battery was carried out to improve the prediction accuracy. Based on the verification of lithium battery capacity data, the results indicate that compared with the single LSTM prediction model, the VMD-HO-LSTM model based on dual-strategy optimization improves the prediction accuracy by 49.6%~81.9%. Compared with the VMD-LSTM model, the battery prediction accuracy of the proposed model is improved by 23.4%~59.0%, and the prediction accuracy is 0.976~0.998. The established model and analysis method have better prediction effect on SOH of lithium battery.

Key words: vehicle and mechatronic engineering; lithium battery; remaining life; dual-strategy optimization; long short-term memory neural networks

摘要： 针对锂电池健康状态(SOH)预测精度不足的问题，提出一种基于变分模态分解与河马算法，优化长短期记忆神经网络(VMD-HO-LSTM)的模型预测锂电池剩余寿命。首先，为消除锂电池的虚假容量信号，采用变分模态分解（VMD）方法对锂电池容量进行分解，得到本征模态分量（IMF）并进行重构；其次，将Logistic映射和自适应学习率融入河马算法（HO），避免迭代过程陷入局部最优，并采用改进河马算法优化长短时记忆（LSTM）网络参数，建立改进的HO-LSTM模型；最后，基于改进的HO-LSTM模型开展锂电池SOH预测，提升预测准确度。基于锂电池容量数据验证，结果表明：相较于单一LSTM预测模型，基于双策略优化的VMD-HO-LSTM模型预测精度提升了49.6%~81.9%；相较于VMD-LSTM模型，电池预测精度提升23.4%~59.0%，该模型预测精度在0.976~0.998，建立的模型和分析方法对锂电池SOH具有较好的预测效果。

关键词: 车辆与机电工程；锂电池；剩余寿命；双策略优化；长短期记忆神经网络