Fault Prediction for Satellite Communication Equipment Based on Deep Neural Network

Aiming at the problem of fault prediction for satellite communication system, a prediction model based on deep learning is proposed in this paper. Firstly, the equipment parameters are summed up, and then two kinds of states covering normal and abnormal situations are determined. After feature learning, self-encoding network is used to obtain new features which can characterize the deep feature of the data. Then the tagged data extracted from monitoring equipment are applied to train the prediction classifier which is the combination of deep belief network and softmax classifier. The deep belief network is composed of limited Boltzmann machine as well as BP network. BP network is used for parameters adjustment. Finally, the effects of fault prediction including the performance of model and average prediction accuracy are tested through simulation.