基于双重注意力机制的电池SOH估计和RUL预测编解码模型  

Encoding and Decoding Model of State of Health Estimation and Remaining Useful Life Prediction for Batteries Based on Dual-stage Attention Mechanism

作　　者：戴俊彦 夏明超[1] 陈奇芳 DAI Junyan;XIA Mingchao;CHEN Qifang(School of Electrical Engineering,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]北京交通大学电气工程学院,北京市100044

出　　处：《电力系统自动化》2023年第6期168-177,共10页Automation of Electric Power Systems

基　　金：中央高校基本科研业务费专项资金资助项目(2018JBZ004);国家自然科学基金资助项目(52007004)。

摘　　要：锂电池的健康状态(SOH)和剩余使用寿命(RUL)精确评估对电池安全稳定运行极为重要,而现有预测模型内部运行机制透明性低,导致评估可靠性较差。文中提出一种基于双重注意力机制的双向长短期记忆网络编解码模型进行SOH估计和RUL预测。编码侧的特征注意力机制和解码侧的时序注意力机制不仅通过动态分配特征和时序信息的权重提升了模型预测精度,还通过可视化权重的方法实现了模型可解释性。最后,利用NASA和CALCE公开的电池数据集进行实验测试,验证了所提方法具有较高的精度和可靠性。Accurate assessment of the state of health(SOH) and remaining useful life(RUL) of lithium-ion batteries is extremely important for the safe and stable operation, and the low transparency of the internal operation mechanism of existing prediction models leads to poor assessment reliability. In this paper, an encoding and decoding model of bi-directional long short-term memory network based on dual-stage attention mechanism is proposed for SOH estimation and RUL prediction. The feature attention mechanism on the encoding side and the timing attention mechanism on the decoding side not only improve the model prediction accuracy by dynamically assigning the weights of feature and timing information, but also realize the model interpretability by visualizing the weights. Finally, experimental tests using National Aeronautics and Space Administration(NASA) and Center for Advanced Life Cycle Engineering(CALCE) public battery datasets verify the high accuracy and reliability of the proposed method in this paper.

关 键 词：锂电池 长短期记忆网络 注意力机制 健康状态 剩余使用寿命 

分 类 号：TM912[电气工程—电力电子与电力传动] TP183[自动化与计算机技术—控制理论与控制工程]