基于支持向量机的发动机性能衰指标分类和预测

摘要

The engine performance degradation index was established based on geometrical distance of support vector machine （SVM） and predicted by multi-step method based on phase space reconstruction theory. The result of support vector regression （SVR） is superior to neural network. Performance characteristic parameters were simplified by principal component analysis （PCA） and kernel PCA （KPCA）. The principal component was extracted and classification of KPCA was better. Parameters of support vector classification （SVC） and kernel function were optimized by cross validation. Curve of engine performance degradation index was presented. Variation of engine performance was analyzed by establishing statistics to determine the key point of performance variation. The results provide reference for engine maintenance and life extension.%基于支持向量机几何距离建立表征发动机性能衰退程度的指标，并基于相空间重构理论对该指标进行多步预测，表明回归支持向量机结果优于神经网络预测结果。利用主元分析、核主元分析方法对发动机性能特征量约简并提取其主元，得到核主元分析的分类效果更好。利用交叉验证的方法优化分类支持向量机和核函数中相关参数，给出发动机性能衰退指标曲线。通过建立统计量的方法分析发动机性能变化，确定性能变化关键点。所得结论对做好发动机维护保养工作，延长发动机使用寿命具有一定的指导意义。