【作者】 金辉；

【导师】 葛红娟；

【作者基本信息】 南京航空航天大学 ， 交通运输工程（适航技术与管理）， 2022， 硕士

【摘要】 机载蓄电池是飞机辅助电源和应急电源的重要组成部分,能够在主电源故障后保证飞机安全着陆,对其开展故障预测与健康评估研究具有工程应用价值。论文以作为机载蓄电池未来发展趋势的锂离子电池为研究对象,基于电池循环充放电实验数据,采用数据挖掘算法开展基于数据的机载蓄电池故障预测和健康评估研究,并提出了一种优化的支持向量机算法。本文首先基于数据分析了环境温度和放电深度对电池性能退化的影响,对充电电压数据微分分析提取容量增量曲线,使用SG滤波和卡尔曼滤波对曲线平滑降噪;提取特征并依据相关系数和灰色关联度进行筛选,使用结合多项式变化的岭回归算法构建间接健康指标,实现基于充电数据评估电池健康状态。将锂电池样本划分为性能良好、临近故障和故障三个健康等级。基于间接健康指标,借助灰色预测算法和使用灰狼算法优化后的长短期记忆网络,建立了能够根据电池健康等级切换算法的机载蓄电池故障预测模型;使用不同工作模式下采集的电池数据训练模型并验证,证明模型有着良好的预测精度。基于支持向量机(Support Vector Machine,SVM)算法建立锂电池健康评估模型,解决不同健康等级的锂电池样本分类识别问题。分析锂电池样本不平衡引起的SVM分离超平面偏移问题,提出SVM惩罚参数分段调整方法,结合决策函数修正系数,对SVM算法进行优化得到分段惩罚参数支持向量机。基于不同条件下采集的电池数据,使用优化前后的算法训练模型并测试评估效果,验证了本文提出的优化思路对SVM分类性能提升。更多还原

【Abstract】 As an important part of the aircraft’s auxiliary power system and emergency power system,the airborne battery can ensure the safe landing of the aircraft after the main power supply fails.It has engineering application value for its failure prediction and health assessment research.The thesis takes lithium-ion battery which is the future development trend of airborne batteries as the research object.Based on the experimental data of battery cycle charging and discharging,data mining algorithms are used to carry out data-based airborne battery fault prediction and health assessment research,and an optimized Support Vector Machine(SVM)algorithm is proposed.This article first analyzes the impact of ambient temperature and depth of discharge on battery performance degradation based on the data,extracts the capacity increment curve by differential analysis of the charging voltage data,and uses SG filter and Kalman filter to smooth the curve and reduce noise;Features are extracted and screened based on correlation coefficients and gray correlations.Ridge regression algorithm combined with polynomial changes is used to construct indirect health indicators to evaluate battery health based on charging data.The lithium-ion battery samples are divided into three health levels: good performance,near failure and failure.Based on indirect health indicators,with the help of the gray prediction algorithm and the long short-term memory network optimized by the gray wolf algorithm,an airborne battery failure prediction model that can switch the algorithm according to the battery health level is established;Use battery data collected in different working modes to train and verify the fault prediction model,which proves that the model has good prediction accuracy.Based on SVM algorithm,a lithium-ion battery health assessment model is established to solve the classification and identification of lithium-ion battery samples with different health levels.Analyze the problem of SVM separation hyperplane offset caused by the imbalance of lithium-ion battery samples,propose a segmented adjustment method of SVM penalty parameters,and combine the decision function correction coefficients to optimize the SVM algorithm to obtain a segmented penalty parameter support vector machine.Based on the battery data collected under different conditions,using the algorithm before and after optimization to train the model and test the evaluation effect,it is verified that the optimization ideas proposed in this paper can improve the performance of SVM classification.更多还原