Digital-Twin-Based Satellite Orbit Prediction for Internet of Things Systems

Satellites play a crucial role in Internet of Things (IoT) applications that require precise positioning. Satellite orbit prediction serves as the foundation for providing accurate terminal location services. However, traditional satellite orbit prediction faces challenges like measurement errors, estimation errors, and unmodeled orbit disturbances, leading to low prediction accuracy. To address this issue, this article introduces a groundbreaking satellite digital twin (DT) system based on container technology. This system facilitates real-time mirroring, monitoring, optimization, and control of satellite orbit prediction with low power consumption. Leveraging the advantages of container technology allows for convenient and efficient model updating. Furthermore, a new satellite orbit error prediction model is explored within this system. This model utilizes the seasonal-trend decomposition using locally weighted regression (STL) method and the temporal convolutional network (TCN) algorithm. By decomposing satellite orbit data into multiple components, the proposed model achieves enhanced future orbit Prediction by combining predicted values from each component. Different from existing machine learning (ML) orbit prediction models, our proposed model explores the variation patterns of satellite orbit data from a trend and cycle perspective, rather than relying solely on collecting more data and training larger models to improve prediction accuracy, which makes the novel prediction scheme get good performance while keeping low prediction complexity. Extensive experiments validate the effectiveness of the proposed method using two publicly available satellite orbit datasets (ILRS catalogue and TLE catalogue). The experimental results show that compared with traditional orbit prediction models, the novel DT system has less model update time and occupies less memory. The mean absolute error (MAE) value of the new model is lower than the five ML models in existing researches, which proves that the proposed STL-TCN model has higher prediction accuracy than existing ML orbit prediction models. In addition, we discussed the impact of atmospheric pressure density on the STL-TCN model, and experiments have shown that the correction of different atmospheric pressure density models has a very small impact on the prediction accuracy of the STL-TCN model. Finally, we further investigate the generalization ability of the STL-TCN model for other satellite orbits and future time orbits, and the results show that the novel model has satisfactory generalization ability.