Satellite Fault Diagnosis Method Based on RBFNN and DS Theory

To overcome long training time and low-level precision rate of traditional satellite defect detection methods based on neural network, a novel fault diagnosis algorithm based on radial basis function neural network (RBFNN) and Dempster-Sharfer (DS) theory is put forward. Principal component analysis (PCA) algorithm is firstly adopted to decrease the high-dimension remote metering data’s quantity. Then RBFNN would be implemented offline training and adjustment. During the process of satellite fault diagnosis, RBFNN is applied to carry out preliminary detection and evaluation. And DS evidence theory is used to locate the error finally. To decrease calculation time, a matrix factorization algorithm is proposed to implement the matrix's parallel arithmetic during RBFNN’s training process, which could distribute remote metering data to different computing core. The experimental findings suggest that the proposed satellite fault diagnosis algorithm base on RBFNN and DS theory can achieve satisfactory fault prediction effects.