摘要
在工业系统中普遍存在样本数据不平衡现象,正常样本数量远远大于异常样本数量。而传统的机器学习算法和深度学习方法,例如朴素贝叶斯和支持向量机,在处理类不平衡问题时,很难获得较高的识别分类准确率,因为它们往往会偏向保证多数类的准确率。为此,本文提出了一种基于生成对抗网络(GAN)的异常检测方法。这个方法中的生成器结构是“编码器–解码器–编码器”的三子网,并且训练该生成器只需要从正常样本中提取特征,所以训练数据集中就不需要异常样本。此系统的异常检测结果由样本的最终得分来判别,其中异常分数由表观损失和潜在损失组成。本文方法的亮点在于可以实现在无异常样本训练的情况下对异常数据样本做检测,通过系统生成更高的异常分数来诊断故障。本项目在凯斯西储大学(CWRU)获得的基准滚动轴承数据集上验证了该方法的可行性和有效性。本文提出的方法在数据集中区分异常样本与正常样本的准确率达到了100%。
The imbalance of sample data is common in industrial system, and the number of normal samples is much larger than the number of abnormal samples. However, machine learning algorithms and deep learning methods, such as Naive Bayes (NB) and Convolutional Neural Networks (CNN), are difficult to obtain high recognition and classification accuracy when dealing with class imbalance problems, because they tend to guarantee the accuracy of majority classes. This paper proposes an anomaly detection method based on Generative Adversarial Networks (GAN). The generator structure in this method is a triple subnet of “encoder-decoder-encoder”, and the training generator only needs to extract features from normal samples, so there is no need for abnormal samples in the training data set. The anomaly detection of the system is determined by the final score of the sample, in which abnormal scores are composed of apparent loss and potential loss. The highlight of the method in this paper is that it can detect abnormal data samples without training abnormal sample and diagnose faults by generating higher abnormal scores in the system. This project has verified the feasibility and effectiveness of this method on the rolling bearing data set obtained from Case Western Reserve University (CWRU). The accuracy of the method proposed in this paper in distinguishing abnormal samples from normal samples in the data set is 100%.
出处
《人工智能与机器人研究》
2019年第4期208-218,共11页
Artificial Intelligence and Robotics Research
