基于PCA与LS-SVM的故障预测应用研究

Research and Application of PCA and LS-SVM Based Fault Prediction

基于数据驱动的故障预测通过对故障特征的历史数据进行建模,对新的数据进行分类,达到故障识别的目的。由于故障数据的样本空间小,而故障特征的维度非常高,故障预测的建模呈现高维度建模困难的特点。针对这些问题,本文利用主成分分析结合最小二乘支持向量机对故障数据进行建模,并构建故障预测的分类模型。首先利用主成分分析方法对高维的故障特征数据进行降维处理,保留故障信息贡献较大的特征(维度),再利用最小二乘支持向量机对降维的样本故障数据进行分类训练,构造故障的分类模型。在不同特征的故障数据集上的测试表明,基于主成分分析故障预测预处理结合最小二乘支持向量机可以更为准确地对故障进行预测,且模型的构建时间较少。

Fault prediction is a model analysis technology through fault features. To deal with fault sample insufficiency and high di-mension fault features, a fault prediction model based on principal component analysis(PCA) and least square support vector ma-chine(LS-SVM) is proposed. PCA is used to project high dimension data onto a lower dimension feature space so that LS-SVMcan be utilized to model the fault data space and efficiently classify the fault data. Experimental tests on most popular fault predic-tion benchmark datasets show that the proposed method can reduce fault prediction training time and is efficient and effective topredict faults.