Research on Predicting the Frequency Variation of 5G Terminal Time Standard Based on LSTM Deep Neural Network

In this paper, we introduce a method of using long short-term memory (LSTM) neural network to predict the time-frequency variation of atomic clocks which are used as time primary standards for 5G communication terminals. We designed a LSTM networks model and trained it using the relative frequency data records of 10 different atomic clocks in NIM during 291 days. Additionally, we discuss the influence of LSTM neural network parameters on its predictive ability. It can be seen from experiments that the LSTM method has well prediction accuracy and repeatability. In particular, it has an advantage in capturing partial frequency hopping.