摘 要

受能源与环保理念影响，新能源汽车的崛起成为汽车发展的必然。锂离子电池以其优良的性能在电动汽车上得到广泛应用。本文以锂离子电池为研究对象，根据电池循环实验数据分析实验过程中电池的性能变化，并由在此基础上建立基于电池放电过程容量增量曲线特征参数的SOH估算模型。

本文首先通过文献整理，对锂离子电池的工作原理和老化机理做了初步分析。并根据NASA实验室的公开数据说明了放电深度对锂离子电池老化的影响。之后，由电池的放电数据获取锂离子容量增量曲线分析。利用小波分析原理对IC（Incremental Capacity）曲线过滤，再提取和峰值相关的参数：峰高、峰位、峰半高宽、峰面积、左斜率、右斜率共六组参数。接着利用灰色关联分析得到与锂离子电池容量关联度最大的三组参数：峰高、峰位、峰面积。

分别建立基于GA-BP（Genetic Algorithm- Back Propagation）神经网络和基于高斯过程回归的锂离子电池SOH（State of Health）估算模型，模型输入选择关联度最高的三组特征参数，而SOH作为模型输出。并将公开数据中的两组电池数据分别作为训练集和预测集进行验证。最终验证了所建立模型的正确性，且精度满足要求。其中高斯过程回归模型在预测结果上提出了一定置信水平下的置信区间。

最后，对比分析了两种模型。结果证明两模型在基于容量增量曲线特征参数的锂离子电池SOH估算上，效果相近，都能满足较高的预测精度要求，但高斯过程回归模型由于其给出的置信区间能过一定程度反映电池老化过程中的不确定因素，因此适用前景更广。

关键词：锂离子电池；容量增量分析法；小波分析；GA-BP神经网络；高斯过程回归

Abstract

The concept of energy crisis and environmental protection has deeply influenced the automobile industry. It has been inevitable for the rise of new energy vehicles. Because of their excellent performance, electric vehicles widely use the lithium-ion batteries. This paper takes lithium-ion battery as the research object, analyzes the performance changes of the battery during the experiment according to the battery cycle experimental data, then establishes the SOH value estimation model based on the characteristic parameters of the capacity increment curve of the battery discharge process.

This paper has simply analyzed the principle and aging mechanism of lithium-ion battery by reviewing literatures. The effect of discharge depth on the aging of Lithium-ion batteries is illustrated by public data from NASA laboratories. After that, Incremental Capacity Analysis (ICA) was obtained from the battery discharge data. The IC curve was filtered by using the wavelet analysis principle, and then the parameters related to the peak value were extracted: peak height, peak position, peak half height and width, peak area, left slope and right slope. In order to obtain the correlation degree with the capacity of lithium-ion battery, I used the grey correlation analysis. And the top three of correlation degree are peak area, peak height and peak position.

GA-BP neural network and Gaussian process regression were used respectively to build the two different kinds of SOH value estimation model of lithium ion battery. Three groups of characteristic parameters with the highest correlation degree were selected for model input, and SOH value was used as model output. In order to verify the accuracy of the model established, two sets of battery data in the public data were respectively used as training set and prediction set. Finally, according to predict outcome, the correctness and accuracy of the model are verified. However, Gaussian Process Regression model can propose the confidence interval at 95% confidence level on the prediction results.

Finally, the two models are compared and analyzed. Result shows that it is similar for the prediction effect of two models which based on the characteristic parameters of Lithium-ion battery capacity increment curve SOH value estimation, both can reach the request of high degree of accuracy, but the Gaussian process regression model because of its confidence interval can give a certain degree reflect the uncertain factors in the process of cell aging, so more widely applicable prospects.

Key Words: Lithium-ion battery; Incremental Capacity Analysis; Wavelet filtering; GA-BP neural network; Gaussian process regression

目录

第1章 绪论 1

1.1研究背景 1

1.2 电池健康状态预测模型研究现状 2

1.2.1经验模型 2

1.2.2电化学模型 3

1.2.3数据驱动模型 4

1.3本文主要的研究内容 5

第2章 锂离子电池性能分析 7

2.1锂离子电池结构及工作原理 7

2.2锂离子电池老化机理 8

2.3锂离子电池循环性能变化 9

2.3.1电池容量变化 9

2.3.2电池放电电压曲线变化 10

2.4 本章小结 10

第3章 锂离子电池容量增量分析 11

3.1容量增量分析法及容量增量曲线 11

3.2基于小波的容量增量曲线去噪处理 12

3.2.1连续小波变换和离散小波变换 13

3.2.2小波分解和重构 13

3.3锂离子电池IC曲线特征参数 15

3.3.1 特征参数提取 15

3.3.2灰色关联分析 16

3.4 本章小结 19

第4章 GA-BP神经网络的SOH建模 20

4.1基本原理 20

4.1.1 BP神经网络 20

4.1.2遗传算法 22

4.2 GA-BP估算模型建立 23

4.2.1遗传算法优化神经网络 23

4.2.2模型建立及参数选择 24

4.3 SOH估算模型仿真结果及分析 25

4.4本章小结 30

第5章 基于高斯过程回归的区间SOH估算模型 31

5.1高斯过程回归 31

5.1.1基本原理 31

5.1.2核函数和超参数初值设定 32

5.1.3估算值的置信区间 33

5.2建立模型 33

5.3仿真结果及评价 35

5.4两种模型对比分析 36

5.4.1预测效果分析 36

5.4.2优化空间分析 37

5.5本章小结 37

第6章 结论与展望 38

6.1结论 38

6.2展望 38

参考文献 39

附录 41

致谢 44

第1章 绪论

1.1研究背景

在资源日益匮乏的今天各国都在寻找方法，采取各种措施来应对能源问题。发展可再生能源是其中一项重要举措，交通领域中，电动汽车作为新能源交通工具正受到广泛关注。不同于传统燃油车，电动汽车直接将电能作为供给来源，没有传统燃油车排放尾气，燃料转化不完全等污染物问题，是更加高效绿色的交通工具。正因如此，各国都投身于电动汽车的研究热潮。

全球各国政府都相当关注电动汽车产业，欧美日等发达国家都制定了相关政策支持电动汽车产业的发展。美国早在2009年投入24亿元用于研究开发电动汽车，并且在2015年成为全球第一个电动汽车数量过百万的国家。日本同样拨款大量资金，推广电动汽车，提升其市场占比率。欧盟更是在经济振兴计划中投资50亿欧元用于电动汽车项目，并计划大规模生产^[1]。