电动汽车动力电池在线状态估计与均衡控制研究

This project focus on the study of lithium-ion battery online state estimation and equalization. Power battery is a quite complex and nonlinear system comprised of interacting physical and chemical processes. State-of-Charge (SOC)，State of Health (SOH) and Remaining Useful Life (RUL) are three key states in battery management and their estimation is an important and challenging task. . This project focus on developing a class of online estimation algorithms to achieve the goal. First, an online battery model parameter identification algorithm based on impulse response is proposed. Then we will develop a Neural Network-Unscented Kalman Filter (NN-UKF) for SOC online estimation. It’s a more robust learning algorithm used for nonlinear state estimation. UKF is simultaneously used for both neural network online training and state estimation simultaneously. The neural network in this algorithm is used to online approximate the uncertainty of the system model due to miss-modeling, extreme nonlinearities, etc. Compared with EKF, UKF gives a more accurate estimate of the connecting weights, so the convergence performance is also improved. . This project also addresses the problem of SOH and RUL estimation used in resource-constrained lithium-ion battery management systems(BMS). SOH estimation includes the maximum available capacity and internal resistance estimation .In this project, same voltage drop interval time sequential is tested online and used to estimate the maximum available capacity. Adaptive UKF method will be used to co-estimate Ohmic internal resistance with SOC.. These estimation algorithms are proposed, along with the analysis of their computing complexity performances and prediction accuracy, as they are designed to run online, being a part of an embedded battery management system.. The proposed battery equalization scheme is a bidirectional dc-dc converter with soft switching that can be used to design the bidirectional nondissipative equalization module for a battery balancing system. The equalization module prescribe the cells equalizing behavior within a safe equalizing region for rapid cell voltage balancing.. The research results in this project will be helpful to improve the reliability of power battery and its service life, thus promoting the industrialization of electric vehicles.

锂离子动力电池是包含复杂物理、化学反应的非线性系统，电池组剩余电量(State of Charge, SOC)、电池组健康状态(State of Health, SOH)和剩余寿命(Remaining Useful Life, RUL)等状态估计以及均衡控制是困扰其应用的主要问题。本项目主要研究锂离子动力电池的在线状态估计算法以及均衡技术：1、基于脉冲响应的电池等效电路模型参数在线辨识算法；2、基于神经网络-无迹卡尔曼滤波的SOC在线状态估计算法；3、基于自适应无迹卡尔曼滤波（AUKF）的SOH在线估计和RUL预测算法研究；4、具有软开关功能的双向主动均衡技术研究。以上算法能在线更新模型参数，提高状态估计精度而且计算量小，便于在嵌入式系统中实现，有效克服了现有算法的不足。本项目针对锂离子动力电池的热点问题深入研究，研究成果将提高动力电池组的使用安全和寿命，推动产业化。

锂离子动力电池是包含复杂物理、化学反应的非线性系统，电池组剩余电量( SOC)、电池组健康状态( SOH)和剩余寿命(RUL)等状态估计以及均衡控制是困扰其应用的主要问题。本项目针对电动汽车电池管理系统开发过程中迫切需要解决的关键科学问题，主要研究锂离子动力电池的在线状态估计算法以及均衡技术：（1）基于脉冲响应的电池等效电路模型参数在线辨识算法：针对戴维南电池模型，建立了一个简单易行的状态空间模型，然后采用基于最小二乘法来识别电池模型的参数。（2）锂离子动力电池的SOC、SOH、RUL在线状态估计算法：针对电动汽车中锂离子动力电池的SOC、SOH、RUL估计问题，研究了三种不同的估计算法，即扩展型卡尔曼滤波（EKF）、粒子滤波（PF）、基于无迹的卡尔曼滤波（UKF）。同时，针对粒子滤波算法，提出了基于顺序蒙特卡罗的SOC估计方法；采用无迹卡尔曼滤波算法实现SOC在线估计，可以有效地提高SOC的估计精度，通过实验数据仿真显示可将估计误差由10%减小到5%；基于不同的机器学习方法，提出了一种基于减法聚类的神经模糊结构，用于电动汽车在不同行驶周期下的SOC估计。（3）具有软开关功能的双向隔离型均衡技术的研究：采用改进的双向Cûk变换器，这种电路结构是对称的，且均衡电路本身不会损耗电池能量，实现了能量双向传递的要求，最大均衡电流可达1A；不仅提高单体电池间均衡的速度，而且比于传统电阻式均衡技术可降低90%能源损耗，使得整个系统相当节能。同时能够延长电池的使用寿命，可实现续航里程最大化。（4）本项目开发了一套BMS样机采用集中式结构、可对各种电动汽车锂离子电池组进行管理，标配安时法估计电池SOC，估计误差5%之内，可根据客户需求定制非线性卡尔曼滤波方法估计SOC，估计误差控制在4%之内。