基于高斯过程回归的锂电池数据处理

发布时间：2018-02-01 00:01

本文关键词： 高斯过程回归锂离子电池增量学习算法智能优化健康状态估计　出处：《北京交通大学》2016年硕士论文　论文类型：学位论文

【摘要】：锂电池由于有着高能量比、低自放电速率、高功率承受力和使用寿命长等优点,逐渐成为能量存储领域的研究热点,并广泛地应用于诸多领域。然而,锂电池在使用过程中性能会逐渐衰退,意外的锂电池性能故障或失效会导致巨大损失和灾难性的后果,特别是在航空航天领域。因此,对锂电池进行健康状态估计,准确地预测出锂电池的剩余寿命,进一步指导锂电池的运行和维护,对于系统的安全性具有重要意义。本文采用数据驱动的方法,将高斯过程回归算法应用到锂电池数据处理领域,对锂电池的健康状态和剩余使用寿命进行预测分析。本文的主要工作有：(1)建立高斯过程回归模型,对电池容量数据进行预测实验,并将预测结果与基于自回归积分滑动平均模型、人工神经网络方法的预测结果进行对比分析。(2)针对当前高斯过程对核函数类型的选择方法还没有统一的理论支撑,系统地探讨了建立高斯过程回归模型中的核函数选择问题,研究了各单一核函数、组合核函数对预测分布的影响,为之后的建模提供理论参考。然后将基于组合核函数的高斯过程回归算法应用到锂电池容量数据的离线建模。(3)针对高斯过程计算量大的问题,采用一种基于增量学习的在线高斯过程回归算法,并与基础高斯过程回归算法进行计算复杂度的对比分析。然后将该在线高斯过程回归算法应用到锂电池的电压数据处理中,实验结果表明本文的高斯过程增量学习算法在保证预测精度的情况下,有效地提高了训练速度。(4)针对共轭梯度法在求取高斯过程回归最优超参数值时,存在优化效果初值依赖性强的缺陷,提出一种基于改进引力搜索算法的高斯过程回归参数优化方法,并将该算法与简单交叉验证方法相结合,以提高高斯过程回归算法的预测精度和泛化能力。将此算法应用在锂电池数据的健康状态估计和剩余寿命预测实验中,并与传统的基于共轭梯度法的高斯过程回归算法、遗传—高斯过程回归算法和粒子群—高斯过程回归算法进行比较。结果表明,基于该算法对锂电池进行健康状态和剩余使用寿命的在线预测可以取得较好的预测结果。
[Abstract]:Because of its advantages such as high energy ratio, low self-discharge rate, high power tolerance and long service life, lithium battery has gradually become a research hotspot in the field of energy storage, and has been widely used in many fields. The performance of lithium-ion battery will gradually decline in the process of use. Unexpected failure or failure of lithium battery will lead to huge losses and disastrous consequences, especially in the field of aerospace. It is very important for the safety of the system to estimate the health status of the lithium battery accurately predict the remaining life of the battery and further guide the operation and maintenance of the lithium battery. Gao Si process regression algorithm is applied to the field of lithium battery data processing to predict the health status and remaining service life of lithium battery. The prediction experiment of battery capacity data is carried out, and the prediction results are combined with the sliding average model based on autoregressive integral. The prediction results of the artificial neural network method are compared and analyzed. 2) there is no unified theoretical support for the current Gao Si process to select the kernel function type. The selection of kernel function in the establishment of Gao Si process regression model is discussed systematically. The influence of each single kernel function and combined kernel function on the prediction distribution is studied. Then the Gao Si process regression algorithm based on combinatorial kernel function is applied to the off-line modeling of lithium battery capacity data. An online Gao Si process regression algorithm based on incremental learning is adopted. Compared with the basic Gao Si process regression algorithm, the algorithm is applied to the voltage data processing of lithium battery. The experimental results show that the Gao Si incremental learning algorithm can effectively improve the training speed under the condition that the prediction accuracy is guaranteed.) according to the conjugate gradient method, we can get the optimal super-parameter value of Gao Si process. Because of the defect of strong dependence on initial value of optimization effect, this paper proposes an optimization method of Gao Si process regression parameters based on improved gravity search algorithm, and combines the algorithm with simple cross-validation method. In order to improve the prediction accuracy and generalization ability of Gao Si process regression algorithm, this algorithm is applied to the health state estimation and residual life prediction experiment of lithium battery data. And compared with the traditional Gao Si process regression algorithm based on conjugate gradient method, genetic algorithm and particle swarm optim_person2# process regression algorithm. The results show that. Based on this algorithm, good prediction results can be obtained for the health status and remaining service life of lithium battery.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TM912;O212.1

【相似文献】