基于SVR的旋转机械转子早期故障预示方法研究-机械制造及其自动化专业论文.docxVIP

摘 要 支持向量机主要包括支持向量回归机和支持向量分类机。近年来支持向量机以其 优越的性能得到了广泛的应用，它的理论基础是统计学习理论，一种适用于小样本条 件下的学习理论。传统的机器学习理论过分注重减小机器的学习误差而忽略了学习机 器的推广能力，支持向量机不仅关注算法的学习误差同时更注重算法的推广能力。因 此支持向量机的性能要优于传统的学习算法。 论文主要研究基于支持向量机的旋转机械转子早期故障预示方法，具体研究内容 如下： 在比较几种支持向量回归机算法的基础上提出了一种改进的 SVR 算法，并研究 了该算法的收敛性。改进的 SVR 算法对论文中仿真数据和实验数据的预测能力优于 传统的几种 SVR 算法和 ARMA 预测模型，表明了改进的 SVR 算法可以有效地应用 于转子运行状态趋势预示。 研究了时域特征量及其频率分量在时间域内的组合预示，以及幅值域特征量的预 示方法，进一步提出了基于 SVR 的转子混合域状态趋势预示方法。仿真分析验证了 基于 SVR 的转子混合域状态趋势预示方法的可行性。 设计了基于 DAGSVM 的转子状态分类器并与 BP 神经网络模型进行了对比研究。 同时开展了基于 SVR 的转子早期故障预示方法的实验验证研究。通过理论和实验分 析表明：基于 DAGSVM 的转子状态分类器是有效的，其中对转子运行状态的正确识 别率优于 BP 神经网络模型。 关键词：支持向量回归机；支持向量分类机；旋转机械；转子；早期故障预示. Abstract Support vector machine(SVM) includes support vector regression(SVR) and support vector classification(SVC). Support vector machine has been applied broadly for advantageous function in recent years. Its theory basis is statistical learning theory that applies to minor sample analysis. Traditional machine study theory pays so much attention to diminish the studying error of machine but ignores its ability for extension. Support vector machine not only pays attention to studying error but also considers the ability for extension of algorithm more. That is why the function of SVM is better than traditional machine study theory. The thesis studies the method of early fault prediction about rotating machine’s rotor based on SVM, specific study content as follows: An improved SVR algorithm was presented on the basis of comparing several kinds of SVR algorithms in the paper. It studied the converge of the improved SVR algorithm. The prediction ability of improved SVR algorithm to the simulated data and the experiment data is better than other SVR algorithms introduced in the paper and ARMA model. That shows the improved SVR algorithm can be applied to the prediction of rotor’s state trend feasibly. The combination prediction of time domain characteristic and its frequency domain characteristic was studied in the paper. Meanwhile , the method of p