The invention discloses a constant-temperature
crystal oscillator time keeping method and
system based on an LSTM neural network. The method comprises the steps that
clock correction, environment temperature,
time sequence and frequency control quantity data are collected in real time to construct a training sample set; taking the historical temperature and the
time sequence as input and the filtered frequency control quantity as output, and training an LSTM neural network to obtain a
frequency drift prediction model; updating the model online by using new data when
satellite signals are available; when the
signal is interrupted, switching to a time-keeping forecasting mode, and inputting the real-time temperature and the
time sequence into a model to predict a frequency control quantity; and generating a
voltage-controlled
voltage according to the predicted value to perform
frequency compensation on the constant-temperature
crystal oscillator, and maintaining the local time reference precision. According to the method, the nonlinear time-varying characteristic of temperature-aging
coupling is captured through the LSTM, model calculation and real-
time control decoupling are realized in combination with a dual-core architecture,
nanosecond-level long-term
punctuality precision is realized in a
satellite rejection environment, and the method has online self-learning capability and
mode switching smoothness.