Application of DE-ELM in analog circuit fault diagnosis

The extreme learning machine (ELM) possesses the advantageous features of the fast learning speed, great generalization performance and high precision. However, the randomness of the parameters will influence its generalization performance and precision greatly. This paper proposes a learning algorithm which is based on the differential evolution extreme learning machine (DE-ELM) for parameter optimization of ELM. It can optimize two parameters, input weights and threshold value, which are random-generated in the network. The experiment selects the elliptic filter circuit to build the fault model. We extract the information of the fault samples using the wavelet packet transformation, then compress the data with the method of principal component analysis. Finally, the DE is applied to optimize the parameters of ELM. The results verified that the proposed method significantly enhances the accuracy of the diagnosis.