关键词: 支持向量机, 故障诊断, 分类效率

Abstract: The extreme sample in training sample set usually make the separation hyper-surface of support vector machines unnecessarily over-convoluted，thus affecting both the classification efficiency and the generalization ability of classifier.After analyzing the reason for this problem，an adaptive adjust method for penalization parameter of support vector machines is proposed.The experimental result shows that this method not only is simple and feasible but also has better anti-jamming ability and higher generalization ability.And it will have a better application foreground in practical work.

Key words: support vector machines, fault diagnosis, classify efficiency