基于自动编码器的锂离子电池状态评估方法  

Autoencoder-based State Evaluation Method for Lithium-ion Battery

作　　者：韩云飞[1] 谢佳[1] 蔡涛[1] 段善旭[1] 程时杰[1] HAN Yunfei;XIE Jia;CAI Tao;DUAN Shanxu;CHENG Shijie(State Key Laboratory of Advanced Electromagnetic Engineering and Technology(Huazhong University of Science and Technology),Wuhan 430074,China)

机构地区：[1]强电磁工程与新技术国家重点实验室(华中科技大学),湖北省武汉市430074

出　　处：《电力系统自动化》2021年第24期41-48,共8页Automation of Electric Power Systems

基　　金：国家自然科学基金委员会-国家电网公司智能电网联合基金资助项目(U196620053)。

摘　　要：准确的电池状态估计对于确保电池储能系统的安全可靠运行至关重要。电池的健康状态(SOH)虽然能反映电池的老化状态,但SOH估计模型的建立受到实际标签数据难以获得或是测试代价高昂的限制。文中基于无监督机器学习模型,建立了一种新的健康指标对电池进行状态评估。首先,从电池的电压-放电容量曲线选择特征,根据锂离子电池的老化机制将电池状态划分为健康和异常,使用健康的数据对基于卷积神经网络的自动编码器模型进行训练,根据自动编码器的输入、输出计算重构误差,最后将重构误差输入逻辑回归模型对电池状态进行判别。在开源的MIT-Stanford数据集上进行实验,验证了所提方法的有效性。Accurate battery state evaluation is critical to ensure the safe and reliable operation of the battery energy storage system.Although the state of health(SOH)can reflect the aging state of the battery,the establishment of the SOH evaluation model is impeded because the actual label data is hard to obtain or the testing is expensive.Based on an unsupervised machine learning model,a new health indicator is established to evaluate the battery state.Firstly,the characteristics are selected from the voltagedischarge capacity curve of the battery,and the batteries are divided into healthy and abnormal states according to the aging mechanism of the lithium-ion battery.The healthy data is used to train the autoencoder model based on convolution neural network,and the reconstruction error is calculated according to the input and output of the autoencoder.Finally,the reconstruction error is input into the logical regression model to assess the battery state.The experiments on the open-source MIT-Stanford dataset show the effectiveness of the proposed method.

关 键 词：锂离子电池 状态评估 深度学习 卷积神经网络 自动编码器 

分 类 号：TM912[电气工程—电力电子与电力传动]