基于主动学习和加权支持向量机的工业故障识别

发布时间：2018-10-05 07:59

【摘要】：针对流程工业过程中有标签故障样本少,样本标注代价昂贵,样本集存在类不平衡以及样本孤点问题,研究基于最优次优标号(BvSB)和加权支持向量机(WSVM)的工业故障分类方法.通过综合考虑样本的信息度和代表性以及样本中可能存在的孤立点,提出改进的主动学习算法,用于挖掘那些对当前分类器模型最有价值的样本进行标注.在支持向量机训练学习中,对不同样本采用不同的权重系数,不同类别赋予不同的惩罚因子,减少了样本分布不平衡时对主动学习和分类精度的影响,充分考虑样本点在特征空间的分布情况,提出新的惩罚系数选取方法.以TE过程为例,实验结果证明,提出的方法能够在获得较高故障分类准确率的情况下减少标注负担.
[Abstract]:In view of the problem that there are few labeled fault samples, the cost of sample labeling is expensive, the class imbalance exists in the sample set and the sample is isolated, an industrial fault classification method based on the optimal sub-optimal label (BvSB) and weighted support vector machine (WSVM) is studied. An improved active learning algorithm is proposed by considering the degree of information and representativeness of the samples and the possible outliers in the samples, which can be used to extract the most valuable samples from the current classifier model. In support vector machine training learning, different samples are given different weight coefficients, and different types are given different penalty factors, which reduces the influence on active learning and classification accuracy when the distribution of samples is unbalanced. Considering the distribution of sample points in the feature space, a new penalty coefficient selection method is proposed. Taking the TE process as an example, the experimental results show that the proposed method can reduce the tagging burden under the condition of higher fault classification accuracy.
【作者单位】：浙江大学控制科学与工程学院;
【基金】：国家自然科学基金资助项目(61174114,U1509203) 教育部高校博士点专项科研基金资助项目(20120101130016) 浙江省公益性技术应用研究计划资助项目(2014C31019)
【分类号】：TP181;TP277

【相似文献】