An Improved GNSS Vertical Time Series Prediction Model Using EWT

GNSS vertical time series are non-stationary, non-linear, noisy, etc. Based on the in-depth study of the Prophet prediction model and Empirical wavelet transform (EWT), Aiming at the poor effect of the decomposed trend item and the single cycle item in the Prophet prediction process, an improved Prophet prediction method using EWT is proposed. First, the original time series is decomposed by EWT. Then, prophet prediction is performed on every component and the predicted time series signal is reconstructed. Finally, the accuracy and reliability of the prediction result are verified. This paper uses the measured GNSS vertical time series data from BJFS, WUHN and URUM stations provided by China Earthquake Administration to conduct four short-term prediction experiments with different time spans. The results show that the improved model can better represent the change trend of the original time series. Compared with the single model, its prediction effect is increased by 31.5%, 35.03%, 19.32%, 10.76% in the root mean square error, respectively. The average percentage error increased by 32.76%, 43.61%, 29.28%, 14%, respectively. It shows that the improved model has better short-term prediction effect and better applicability.