基于SVDD的发动机冷试多参数控制限设计

Normal Domain Design of Multi-parameter in Engine Cold Test Based on SVDD

为了提高发动机冷试检测的准确率,将支持向量数据描述应用于正常域设计。基于尾部关联性选择关联参数,针对数据的特点改进了SVDD参数优化方法,实现了SVDD边界形成算法,提出增量学习策略应对学习过程样本量巨大的问题。实例测试结果表明新型正常域降低了漏报率,并指出误报率将随着样本的引入降低并趋于稳定。

In order to improve the accuracy of engine cold test, apply Support Vector Data Description （SVDD） to design of the normal domain. Choose relevant parameters based on the tail relevance, improve parameters optimiza- tion method of SVDD according to the data characteristic, achieve the SVDD boundary formation algorithm, put forward incremental learning strategies to deal with the big sample size. Test results show that the new normal domain reduces Type 1I error , and point out that the Type I error will be reduced with the introduction of the sample and tends to the stability.