面向梯次利用的锂离子电池健康状态估计与剩余寿命预测

China is in the process of facing the peak period of the retired batteries. Echelon-use of the aged batteries is consistent with the national advocate of the low-carbon economy, green economy and circular economy. It is imminent to develop a new, low-cost method for battery state evaluation. This project aims to develop methods only relying on the data of the electric vehicle operating conditions, based on the physics-based battery model, multi-dimensional state of health estimation and fault diagnosis, the remaining useful life prediction methods will be developed..Firstly, the characteristic of the operating conditions and the identifiability of battery physical model parameters will be analyzed. A database of battery simulation and experimental data will be set up. A physical model parameter identification method based on Convolutional Neural Network will be developed. The mathematical relationship between battery performance degradation and battery model parameters will be analyzed. Based on the parameters’ values and changing trends, the multi-dimensional state of health will be estimated, and the battery fault will be diagnosed. A battery aging mechanism model will be developed, the multi-dimensional state of health will be regard as the observation, with an adaptive state-dual extended Kalman filter, and the battery remaining useful life will be predicted in close-loop. The research results will provide new approaches for battery testing and state evaluation in the battery echelon-use applications.

我国即将面临电动汽车锂离子电池的退役高峰，对退役电池进行梯次利用，符合国家低碳经济、绿色经济以及循环经济的倡导，研究一种新型、低成本的锂离子电池状态评估方法迫在眉睫。为了避免传统电池状态评估方法测试时间长、鲁棒性差的问题，本项目以电池物理模型为基础，利用电动汽车工况数据，研究电池多维健康状态估计与故障诊断、剩余寿命预测的新方法。.首先，分析工况和参数可辨识性特征，采用仿真与实验相结合的方法建立电池数据样本库，设计基于卷积神经网络的电池物理模型参数辨识方法；随后，分析电池性能衰退与电池模型参数之间的数学联系，根据参数值及其变化趋势，实现电池可循环锂离子损失、电极活性材料损失的多维健康状态估计和故障诊断；最后，构建电池老化机理模型，将多维健康状态信息作为系统观测量，利用自适应状态-双扩展卡尔曼滤波算法，实现电池剩余寿命的闭环预测。研究成果将为电池梯次利用实践中的电池状态评估提供新的技术手段。

我国即将面临电动汽车锂离子电池的退役高峰，对退役电池进行梯次利用，符合国家低碳经济、绿色经济以及循环经济的倡导。本项目的研究目标是针对车用锂离子电池，研究其健康状态评估方法和剩余可用寿命的预测方法。为了避免传统电池状态评估方法测试时间长、鲁棒性差的问题，本项目以电池物理模型为基础，利用电动汽车工况数据，研究了物理模型降阶方法和参数提取方法。搭建了针对动力电池的多物理场融合测量系统，实施了电池老化实验和特性分析实验，实现了电池多维健康状态估计与故障诊断。利用实车数据和迁移学习方法，实现了动力电池的闭环剩余可用寿命预测。设计了电池管理系统软硬件，基于电池物理模型对退役电池模组进行评价与筛选。研究成果将为电池梯次利用实践中的电池状态评估提供新的技术手段。