基于动态贝叶斯网络的锂离子电池soh估计-soh estimation of lithium ion batteries based on dynamic bayesian network.docxVIP

杭州电子科技大学硕士学位论文 I 摘 要 随着电子科学技术的发展和人们生活水平的日益提高，锂离子电池越来越多地出现在各 种移动设备和电动汽车上。锂离子电池经常由于过充、过放而发生爆炸，因此电池管理系统 (BMS)显得至关重要。电池管理主要包括电池荷电状况(SOC)估计、电池健康状况(SOH)估计 和电池均衡等。SOH 是 BMS 需要监控的电池关键参数之一，准确的 SOH 估算能够帮助人们 在系统安全性和经济效益之间取得更好的平衡。 SOH 不能通过直接测量的方式得到，目前主要使用的间接估计方法包括完全放电法、基 于神经网络的算法和基于卡尔曼滤波器的算法等。本文首先详细介绍了几种常用的 SOH 估计 方法，并对比分析了它们在电池 SOH 估算方面的优缺点。接着，结合实际应用情况，本文将 SOH 归纳为“很新”、“新”、“一般”，“旧”和“很旧”五个阶段，并在此基础上提出 了基于动态贝叶斯网络(DBN)的电池 SOH 估算方法。DBN 模型的结构是根据以往的经验人工 构建的，而模型的参数则是通过电池老化实验中获得的数据学习得到的；本文使用前向算法 进行 DBN 的推理，降低了计算量和计算复杂度，使得通过普通单片机实时估计 SOH 成为可 能。 最后，为了验证本文所提出的算法的可行性，我们设计并实现了基于 STM32F103VCT6 的 电池 SOH 估算系统，实验结果表明，本文所提出的锂离子电池 SOH 估计算法是可行的。 关键字：锂离子电池，电池管理系统，电池健康状况，动态贝叶斯网络 II ABSTRACT With the development of electronic science and improvement of people’s living standards, more and more lithium-ion batteries are used in mobile devices and electro mobiles. Safety issues must be concerned when we use Li-ion batteries, so battery management system (BMS) is of great importance. The state of health (SOH) is one of the key parameters of batteries that a BMS should monitor. Accurate estimation of SOH can help us to get better balance between safety of system and economic benefits. The SOH cannot be measured by any direct method, at present; the main estimation methods invented by researchers all over the world include completely discharge method, the method based on neural network and the method based on Kalman filter. Firstly, these three methods are introduced in detail and compared in this thesis, and some deficiencies of them are pointed out. Then, according to the actual situation, SOH is originally discretized using five intervals (brand-new, new, ok, old and very-old). On this basis, this thesis presents a novel online method for the estimation of the SOH of Li-ion. The proposed method is based on Dynamic Bayesian Network (DBN). The structure of the model is created according to the experience of experts, and the parameters of the model is learned based on data collected in a Li-ion battery agi