摘 要

随着节能减排和绿色低碳理念的深入人心，作为最重要的交通运输工具的船舶，其绿色动力技术受到的业内外极大的关注。交通运输部近期发布的《内河航运发展纲要》中明确要加大新能源清洁能源推广应用力度，纯电力动力具有零污染零排放的优势，是探索发展的重要技术。纯电池动力的关键是锂电池储能系统，虽然锂电池随着电动汽车量产技术越来越成熟，成本越来越低，但船用动力电池系统的容量配置较电动汽车要大的多，甚至达到几十上百倍，串并联单体数量巨大。开展船舶动力锂电池组故障预测和健康管理，提高系统的可靠性，具有重要的工程应用价值。

本文以内河纯电池动力船舶锂电池系统为研究对象，针对锂电池单体的剩余寿命预测问题，选用NASA PCoE锂电池充放电循环数据集开展仿真计算，提出了基于粒子滤波算法和基于高斯过程回归模型两种容量预测方法。

（1）基于粒子滤波算法的RUL预测：以锂电池容量衰退的线性和指数组合模型为基础，通过粒子滤波代码的无噪跟踪实验确定模型参数。利用粒子滤波算法，逐步计算容量的预测值，并根据每一步的粒子分布得到容量预测序列的95%置信区间，使得输出结果具有不确定表达能力。结果表明，粒子滤波算法能很好的描述锂电池在船舶上工作的不确定因素，同时具有很好的稳定性。但由于容量衰退模型并非通用，预测方法灵活性较差。

（2）基于高斯过程回归模型的RUL预测：提取等压降放电时间作为健康因子，并建立线性回归模型和高斯过程回归模型，迭代计算出容量的预测序列及其95%置信区间。结果表明，等压降放电时间能很好表征电池的健康状态，GPR模型的输出结果具有较高的精度和良好的不确定表达能力，能为船舶电池健康管理系统提供更丰富的信息。

关键词：船舶；锂离子电池；寿命预测；粒子滤波；高斯过程回归

Abstract

With the concept of energy conservation and emission reduction and green low-carbon deeply rooted in the people's mind,as the most important means of transportation,the ship's green power technology has attracted great attention both inside and outside the industry.Recently,the Ministry of transport issued the outline for the development of inland navigation,which clearly states that the promotion and application of new energy and clean energy should be strengthened.Pure electric power has the advantage of zero pollution and zero emission,which is an important technology for exploration and development.The key of pure battery power is the lithium battery energy storage system.Although the lithium battery is becoming more and more mature with the mass production technology of electric vehicles and the cost is becoming lower and lower, the capacity configuration of marine power battery system is much larger than that of electric vehicles, even reaching tens or hundreds of times, and the number of series parallel units is huge. Fault prediction and health management,improve the reliability of the system, has important engineering application value.

In this paper,the lithium battery system of neihe pure battery powered ship is the research object.Aiming at the prediction of residual life of lithium battery,NASA pcoe lithium battery charge and discharge cycle data set is selected for simulation calculation.Two capacity prediction methods based on particle filter algorithm and Gaussian process regression model are proposed.

(1)RUL prediction based on particle filter algorithm:Based on the linear and exponential combination model of lithium battery capacity degradation,the model parameters are determined by the noiseless tracking experiment of particle filter code.Using particle filter algorithm, the prediction value of capacity is calculated step by step, and the 95% confidence interval of capacity prediction sequence is obtained according to the particle distribution of each step, which makes the output result have uncertain expression ability.The results show that the particle filter algorithm can well describe the uncertain factors of lithium battery working on the ship,and has good stability.However, because the capacity decline model is not universal,the prediction method is not flexible.

(2)RUL prediction based on Gaussian process regression model: extract the equal voltage drop discharge time as the health factor, establish linear regression model and Gaussian process regression model, and iteratively calculate the prediction sequence and 95% confidence interval of capacity. The results show that the constant voltage drop discharge time can well represent the health status of the battery, the GPR model output results have high accuracy and good uncertainty expression ability, which can provide more information for the ship battery health management system.

Key Words：ship；lithium-ion battery；life prediction；particle filter；Gaussian process regression

目 录

第1章 绪论 1

1.1课题研究背景及意义 1

1.2国内外研究现状 2

1.2.1船舶动力锂电池系统及其管理系统的研究现状 2

1.2.2锂离子电池SOH估计研究现状 2

1.2.3锂离子电池RUL预测研究现状 3

第2章 锂离子电池工作原理及衰退机理 6

2.1锂离子电池工作原理 6

2.2锂离子电池容量衰退机理 7

第3章 基于粒子滤波的锂离子电池RUL预测 9

3.1粒子滤波 9

3.1.1贝叶斯估计理论 9

3.1.2蒙特卡洛方法 11

3.1.3粒子滤波基本原理 11

3.2数据来源及分析 13

3.3锂离子电池容量退化模型 14

3.4计算结果及问题分析 16

3.4.1计算结果分析 16

3.4.2算法缺陷 18

第4章 基于高斯过程回归模型的锂离子电池RUL预测 20

4.1健康因子的构建 20

4.2高斯过程回归模型 22

4.2.1贝叶斯线性回归概述 22

4.2.2非线性系统在贝叶斯框架下的回归问题 23

4.3计算过程及结果分析 24

4.3.1预测模型的构建及更新 24

4.3.2计算结果及分析 25

第5章 结论 29

参考文献 31

致谢 33

第1章 绪论

1.1课题研究背景及意义

随着人类生产力的提高，化石能源消耗量不断增加。造成的环境污染日益严重。因此，各行各业都面临发展绿色低碳技术，实现节能减排的挑战。

我国内河柴油动力船舶运输总量大，里程长，燃料消耗量巨大。船用燃料油含有大量硫、重金属等有害物质，船舶柴油系统的大气污染物排放量占我国排放总量很大比重。我国沿海和内河流域城市多、人口密集，是限制船舶污染的重点区域。2016年5月实施的《防治船舶污染内河水域环境管理规定》为落实《大气污染防治法》，从燃油质量、防污染设备和清洁能源等方面，提高了内河船舶减少空气污染的要求。2019年，交通运输部将我国船舶排放控制区扩容，再一次鼓励发展新能源船舶。在柴油机船舶运营成本的优势逐渐瓦解的背景下，迫切需要新型清洁能源船舶，纯电池动力和混合动力船舶受到了人们的高度关注和青睐。船用动力电池组作为核心能量供给源，具有电池单体和串并联支路多、容量大、风险高等问题，监测其工作过程中的健康状态极为重要。