Please contact：6720230184@mail.jxust.edu.cn (Yu Zhou) and 2452481248@qq.com (Jun Li).
- He, X., Zhou, Y., Li, J., Kermarrec, G., Fernandes, R., Montillet, J-P., Zhang, S., Hu, S., TS_Predictor: A Deep Learning Toolbox for GNSS Time Series Prediction with Signal Decomposition and Nonlinear Modelling, Advances in Space Research (2026), doi: https://doi.org/10.1016/j.asr.2026.03.083
- Zhang, S., Li, J., Zhao, L., Zeng, A., Ming, F., Liu, N., Chen, X., Feng, Z. and Wang, H., 2025. gCMEbox: A MATLAB toolbox for extracting and analyzing common-mode errors from GNSS time series. Advances in Space Research, 75(1), pp.497-514.
- Zhou, Y., He, X., Montillet, J.P., Wang, S., Hu, S., Sun, X., Huang, J. and Ma, X., 2025. An improved ICEEMDAN-MPA-GRU model for GNSS height time series prediction with weighted quality evaluation index. GPS Solutions, 29(3), p.113.
- Hu, S., Chen, K., He, X., Zhu, H., & Wang, T. (2025). Research on the impact of environmental loading on nonlinear variations of 3D coordinate time series of GNSS stations in Sichuan and Yunnan region. Acta Geodaetica et Cartographica Sinica, 54, 805-818. (胡顺强,陈克杰,贺小星,等. 环境负载对川滇地区GNSS观测站三维坐标时间序列非线性变化的影响 [J]. 测绘学报, 2025, 54 (05): 805-818.)
- He, X., Bos, M. S., Montillet, J. P., & Fernandes, R. M. S. (2019). Investigation of the noise properties at low frequencies in long GNSS time series. Journal of Geodesy, 93(9), 1271-1282.
- He, X., Montillet, J.P., Fernandes, R., Bos, M., Yu, K., Hua, X. and Jiang, W., 2017. Review of current GPS methodologies for producing accurate time series and their error sources. Journal of Geodynamics, 106, pp.12-29.
Predicting Global Navigation Satellite System coordinate time series is essential for early warning systems, infrastructure monitoring, and geophysical modeling, yet no dedicated open-source tool currently exists for this task. We developed TS_Predictor, a MATLAB toolbox that brings together deep learning, signal decomposition and nonlinear modelling in a single, accessible package. Providing a complete workflow from raw data preprocessing to prediction accuracy assessment. The software supports multiple data formats from major data centers, implements comprehensive preprocessing functions including missing data interpolation, outlier detection, offset correction, and common mode error removal, and offers six decomposition methods combined with six prediction models. We introduce a novel Weighted Quality Evaluation index (WQE) that combines root mean square error, mean absolute error, symmetric mean absolute percentage error, and coefficient of determination into a unified metric for model comparison. Validation using data from ten stations in Yunnan, China (2010 to 2025) demonstrates that the decomposition-prediction approach consistently outperforms direct prediction.
- Data Import: Supports multiple formats from NGL, PBO, and CEA data centers
- Preprocessing: Missing data interpolation, outlier detection, offset correction, CME removal
- Signal Decomposition: EMD, EEMD, CEEMD, CEEMDAN, ICEEMDAN, VMD
- Prediction Models: SVM, RBF, ELM, BP, LSTM, GRU
- Accuracy Evaluation: RMSE, MAE, sMAPE, R², and unified WQE index
- Visualization: Interactive plots and exportable results
- MATLAB 2024b or later
- Deep Learning Toolbox
- Signal Processing Toolbox
- Statistics and Machine Learning Toolbox
-
Download or clone this repository:
git clone https://github.com/SpaceGeodesyLab/TS_Predictor.git
-
Open MATLAB and navigate to the TS_Predictor directory
-
Right-click on
TS_predictor.mlappinstalland select Install -
After installation, TS_Predictor will appear in the Apps toolbar
-
Click the TS_Predictor icon to launch the program
TS_Predictor accepts GNSS time series from:
- Nevada Geodetic Laboratory (NGL):
.txtformat - Plate Boundary Observatory (PBO):
.csvand.posformats - China Earthquake Administration (CEA):
.txtformat
Use the Data Conversion module to convert your data to the unified CSV format.
The preprocessing module (based on gCMEbox) provides:
- Interpolation: Kriging-Kalman filtering for missing data
- Outlier detection: IQR and MAD methods
- Offset correction: Weighted least-squares fitting
- CME removal: PCA, ICA, FastICA, vbICA
Select a decomposition method to separate your time series into components:
| Method | Description | Best for |
|---|---|---|
| EMD | Empirical Mode Decomposition | Basic separation |
| EEMD | Ensemble EMD | Reducing mode mixing |
| CEEMDAN | Complete Ensemble EMD with Adaptive Noise | Cleaner separation |
| ICEEMDAN | Improved CEEMDAN | Best overall performance |
| VMD | Variational Mode Decomposition | Known number of modes |
Choose a prediction model:
| Model | Type | Characteristics |
|---|---|---|
| SVM | Machine Learning | Good for small datasets |
| RBF | Machine Learning | Fast training |
| ELM | Machine Learning | Very fast, less accurate |
| BP | Neural Network | Classic approach |
| LSTM | Deep Learning | Best for long sequences |
| GRU | Deep Learning | Good balance of speed/accuracy |
Recommended: ICEEMDAN + GRU for most GNSS applications.
Compare models using multiple metrics:
- RMSE: Root Mean Square Error
- MAE: Mean Absolute Error
- sMAPE: Symmetric Mean Absolute Percentage Error
- R²: Coefficient of determination
- WQE: Weighted Quality Evaluation (unified metric) Results are exported to Excel for further analysis.
- gCMEbox - GNSS Common Mode Error extraction toolbox
- Hector - GNSS time series noise analysis
- TSAnalyzer - GNSS time series analysis
- ......
Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.