Feature extraction method of analog circuit based on deep learning

Abstract

The signals of early fault analog circuits are weak, and it is difficult to extract the depth and essential features by using the traditional extraction method. In addition, the extracted features are easy to overlap. To solve these problems, a novel optimization deep belief network (DBN) for extracting circuit features is proposed in this paper. which uses particle swarm optimization to optimize the number of neurons in the hidden layer, so as to determine the optimal network structure. At the same time, it can overcome the deficiency that many attempts should be paid to set the number of neurons according to experience, and improve the effectiveness of feature extraction furtherly. This method digs the feature information from the original data directly, and it has strong characterization ability, which overcomes the dependence of the traditional feature extraction method on the knowledge of signal processing. So the project is easy to realize. The validity of this method was verified by using the simulation of the Sallen-Key filter circuit. The results confirm that the characteristics of the different fault categories extracted by this method have better separation. The redundancy of features was reduced, besides, the accuracy of subsequent health state monitoring and residual life prediction was improved.