Document Type : Original Article

Authors

• Majid Bakhtiari ¹
• Amirhossein Panahyazdan ²
• Siavash Sabzy ²

¹ Iran University of Science and Technology

² School of Advanced Technologies, Iran University of Science and Technology, Tehran, Iran

Abstract

The Earth Orientation parameters (EOPs), such as polar motion, universal time, and length of day, play a prominent role in procedures such as monitoring the Earth's rotation, weather modeling, and disaster prevention. This paper estimates the EOPs series based on a combined series approach proposed by the International Earth Rotation and Reference System Observatory between 1962 and 2023, incorporating data from diverse space geodetic techniques, including DORIS, laser ranging (LLR and SLR), GNSS, and VLBI to create EOPs series. This paper proposes a hybrid deep-learning prediction model, combining a Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) with an attention mechanism. The CNN effectively extracts and enhances features, while the GRU facilitates medium- to long-term predictions based on historical time series data. The attention mechanism prioritizes relevant data aspects, enhancing the Model's ability to discern intricate patterns, particularly for Length of Day (LOD) variations, where some covariants affect its pattern and should be considered. One of the most practical applications of these parameters is mapping the points in the terrestrial and celestial reference systems to each other. These predicted EOPs are used to create a high-accuracy coordinate transformation matrix from ECEF to ECI for applications such as high-precision navigation.

Keywords

• Neural Network: Earth Orientation Parameters: Space geodetic techniques
• Coordinate transformation
• Deep learning

Main Subjects

• Aerospace Science and Technology