Aiming at the inaccuracy of global positioning system（ GPS） coordinate time series prediction

a prediction model of GPS coordinate time series based on multivariable long short-term memory（ LSTM） is proposed. The grey correlation method is used to analyze the correlation degree of GPS coordinate time series data from different observation stations in the same area

and the data to be predicted（ U-direction historical data of SHAO station are found.） The data with strong correlation are used as the input of the multivariable prediction model. The multivariable LSTM model is established by combining the existing input information with the historical input information by LSTM. By comparing the prediction results with ARIMA and single variable LSTM models

it is proved that LSTM model can be used as the input of the multivariable prediction model. This method has better prediction effect.