Optional SVM for Fault Diagnosis of Blast Furnace with Imbalanced Data

• DOI: 10.2355/isijinternational.51.1474
• Publication Date: 2011-09-15T10:42:28Z
• Authors: Limei Liu; Anna Wang; Mo Sha; Xiyan Sun; Yunlu Li
• Keywords: Pruning

Fault diagnosis for blast furnace is actually a multi-class classification problem because the may appear usually many kinds of abnormal states. Moreover, those states should be monitored and diagnosed timely what can help workers take effective measures. Support vector machine (SVM) state-of-the-art problems currently. But tasks involve imbalanced training examples in practice. Imbalanced dataset learning an important practical issue learning, especially support (SVM). such data problem. A novel algorithm named optional proposed to solve this by pruning sets adding unlabeled applying edited nearest neighbor (ENN) rules. Firstly, majority class are pruned order reduce time. Secondly, selects some useful unlabelled adds them sets. Those samples used replenish lack so that representative. However, they contain noisy examples. Finally, rule removed The (testing) balance number between minority one. real-time producing running experiment. In more accurately diagnose which fault happened, binary tree method adopted based on characteristics. Simulation results show feasible effective.