期刊名称：IAENG International Journal of Computer Science
印刷版ISSN：1819-656X
电子版ISSN：1819-9224
出版年度：2021
卷号：48
期号：4
语种：English
出版社：IAENG - International Association of Engineers
摘要：The existing intelligent condition monitoring model needs a large number of historical data and corresponding tags under different health states to train the model. It is difficult to collect abnormal samples in some actual system. A new system anomaly detection method trained with no abnormal sample is proposed. The proposed method combines the reward function model with maximum entropy and generative adversarial networks (GAN). Firstly, the GAN is trained with expert samples to generate virtual expert samples. The non-expert samples are generated by using random strategy on this basis. The mixed sample set with expert and non-expert samples is constructed. Combined with the maximum entropy probability model, the reward function is computed, and the gradient descent method is used to solve the optimal reward function. Secondly, the proposed model is trained by the normal samples collected in the early stage, and then used to detect the unknown state. Finally, the system is monitored by observing the change of the difference index generated by GAN with maximum entropy. The experimental analysis results verify the effectiveness of the method. Compared with the traditional algorithm, the proposed method detects the system anomaly earlier, and the difference index increases more rapidly when the anomaly occurs.
