动力电池SOH在线实时估计算法研究

发布时间：2018-06-29 22:15

本文选题：递推最小二乘法 + 遗传粒子滤波　；参考：《河南师范大学》2017年硕士论文

【摘要】：随着油价的上升、PM2.5不断地爆表给传统汽车生产企业带来不小的压力,也迫使这些企业朝着更具发展前景的新能源汽车方向转型。这些新能源汽车大多以电池组供能为主,结合电池管理系统来满足人们的日常行驶需求。在汽车领域,新能源汽车目前还处于刚起步阶段,人们将大部分研究重心放在对电池剩余电量(SOC)的精确研究上,而对电池的健康状况涉及的较少。目前,电池自燃、爆炸等现象频频出现,电池的健康成为了人们关注的对象,促使着人们将重心开始朝向电池健康状况(SOH)偏移。电池的健康状况已经成为电池管理中重要的一个环节,如何准确的预测SOH对新能源汽车的发展具有重要的意义。本文就是以此为背景对电池的健康状态进行研究,具体工作如下:首先对锂电池寿命衰减分析并参照现有的相关文献,发现电池中欧姆内阻阻值的大小可以作为SOH的评判准则。然后对现有的三类模型进行对比分析,最终选择了二阶RC电池等效电路为模型。最后,选取一节电池并测取该节的数据,通过递推最小二乘法进行参数辨识,将辨识的值代入求解来确定该电池相对应的模型。通过对电池等效电路分析得出该模型是一个非线性系统,因此提出了适合处理非线性、非高斯系统的粒子滤波算法。本文通过序贯重要性采样(SIS)在高维函数采样的思想,将时间函数替代高维函数引出粒子滤波这一算法。然而粒子滤波在处理问题时会随着迭代次数的增加而出现粒子消失的退化现象,会严重影响到预测的精度。为此本文在原有粒子滤波的基础上又引入了重采样的概念。其思想是当有效粒子数低于设定的阈值时,将所有粒子进行等权值重新分配。将分配的粒子继续代入循环中,直到运行到规定的迭代次数后结束。根据等效电路模型建立动态方程与观测方程,并列出具体的操作步骤。根据所列的操作步骤进行编程并通过MATLAB对其仿真。通过对仿真图形的观察,发现重采样粒子滤波对预测电池内阻具有较高的精确度。为了更加精确的估算电池内阻,本文提出了将遗传算法中进化思想来代替重采样过程。该算法不仅解决了粒子退化的问题而且运用基因重组、基因突变的方式来丰富了粒子的种类。其思想是当有效粒子数低于设定的阈值时,将所有粒子采用遗传算法的方式进行处理。将处理过的粒子继续代入循环中,直到运行到规定的迭代次数后停止。最后,建立动态方程与观测方程,列出关于遗传粒子滤波具体的操作步骤。根据所列的操作步骤进行编程并通过MATLAB对其仿真。将上述两种算法仿真出来的图形进行观察比较,发现采用遗传粒子滤波预测的曲线更为平缓且抖动更小。因此表明遗传粒子滤波对估算内阻时具有更优越的性能。
[Abstract]:As oil prices rise, PM2.5 keeps popping up, putting pressure on traditional auto makers and forcing them to shift towards more promising new energy vehicles. Most of these new energy vehicles are powered by battery pack and combined with battery management system to meet people's daily driving needs. In the automotive field, the new energy vehicle is still in its infancy. Most of the researches focus on the accurate study of the battery residual charge (SOC), but less on the battery health. At present, the phenomena of battery spontaneous combustion and explosion appear frequently, and the health of battery becomes the object of concern, which urges people to shift the center of gravity towards the battery health condition (SOH). Battery health has become an important link in battery management. How to accurately predict SOH is of great significance to the development of new energy vehicles. In this paper, the health status of the battery is studied in this paper. The main work is as follows: firstly, the life attenuation of lithium battery is analyzed, and referring to the existing literature, it is found that the ohmic resistance in the battery can be regarded as the criterion of SOH. Then the three kinds of models are compared and the second order RC battery equivalent circuit is selected as the model. Finally, a battery is selected and the data of the section is measured. The parameters are identified by recursive least square method, and the corresponding model of the battery is determined by substituting the identified value into the solution. Through the analysis of the equivalent circuit of the battery, the model is a nonlinear system. Therefore, a particle filter algorithm suitable for dealing with nonlinear and non-Gao Si systems is proposed. Based on the idea of sequential importance sampling (SIS) sampling in high dimensional function, the particle filter algorithm is derived by replacing the time function with the high dimensional function. However, particle filter will degenerate with the increase of iteration times, which will seriously affect the accuracy of prediction. In this paper, the concept of resampling is introduced based on the original particle filter. The idea is to redistribute all particles with equal weights when the number of effective particles is lower than the set threshold. Continue the allocated particles into the loop until the end of the specified number of iterations. According to the equivalent circuit model, the dynamic equation and the observation equation are established, and the concrete operation steps are listed. According to the listed operation steps to program and through MATLAB to its simulation. By observing the simulation figure, it is found that the resampling particle filter has a high accuracy in predicting the internal resistance of the battery. In order to estimate the internal resistance of the battery more accurately, the evolutionary idea of genetic algorithm is proposed to replace the resampling process. The algorithm not only solves the problem of particle degradation, but also enriches the species of particles by gene recombination and gene mutation. The idea is that all particles are processed by genetic algorithm when the number of effective particles is lower than the set threshold. The processed particles continue to be inserted into the loop until they have stopped running at the specified number of iterations. Finally, the dynamic equation and observation equation are established, and the operation steps of genetic particle filter are listed. According to the listed operation steps to program and through MATLAB to its simulation. By observing and comparing the figures simulated by the above two algorithms, it is found that the curve predicted by genetic particle filter is more gentle and the jitter is smaller. Therefore, genetic particle filter has better performance in estimating internal resistance.
【学位授予单位】：河南师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM912

【参考文献】