Abstract: There are only a few equipments that can be installed with thermistors because of the limited sources on satellite. A BP neural network which can predict the temperature of equipment without thermistor was built based on the excellent fitting ability of BP neural network for complex nonlinear system. The temperature data acquired in thermal test through thermal couple of equipments on satellite either with or without thermistors were used to train and test the neural network. The test result shows that the temperature prediction accuracy of BP neural network is smaller than 1℃, and the temperature prediction neural network can be used to accurately predict the temperature of equipment without thermistor. Additionally, the relationship between samples and estimation errors were also studied, showing that sample diversity and large data can reduce the estimation error significantly. 

Key words: BP neural network, without thermistor, temperature, prediction, machine learning