GNSS-RTK time series denoising based on deep learning and mode decomposition techniques for offshore platform

Global navigation satellite systems (GNSS) in real-time kinematic (RTK) mode play a crucial role in the dynamic displacement monitoring of offshore platforms. This provides valuable data for assessing the impacts of environmental conditions and operational activities on platform stability. However, its positioning accuracy is limited by multipath and satellite shielding effects, system errors, and other factors. To address these challenges, this study develops a denoising strategy based on deep learning and mode decomposition techniques to enhance the accuracy of GNSS-RTK dynamic monitoring. First, an autoencoder-based network is established and trained for the anomaly detection of the GNSS time series, primarily focusing on tasks such as missing data imputation and outlier removal. Subsequently, the data reconstructed via the network undergo decomposition into a series of independent modes known as the intrinsic mode function (IMF) using an improved variational mode decomposition (IVMD) approach. IVMD can remove the impact of noise-dominant components (i.e., spurious IMF components) by introducing a cross-correlation filtering mechanism. Finally, a high-precision GNSS dynamic displacement monitoring result is derived by summing the remaining IMF components. The developed approach is verified though field tests on an offshore platform under ambient excitation. The results illustrate that the developed method is capable of effectively reconstructing lost data and removing unnecessary outliers, achieving a low root mean square error (RMSE). This approach can eliminate the influence of spurious components induced by noise while retaining the components related to the structural dynamic vibration.