1
LEOSTP: A Spatio-Temporal Traffic Prediction Framework for LEO Satellite Networks
arXiv:2606.29856v1 [cs.IT] 29 Jun 2026
Shaoyou Ao, Graduate Student Member, IEEE, Yong Niu, Senior Member, IEEE, Zhu Han, Fellow, IEEE, Cheng Li, Member, IEEE, and Bo Ai, Fellow, IEEE
Abstract—With the evolution of next-generation mobile communication networks and the commercial boom of Low Earth Orbit (LEO) satellites, globally covered satellite networks are gradually becoming a crucial infrastructure for massive user access and seamless connectivity. Accurate traffic prediction is crucial for maintaining the quality of service (QoS) and resource allocation efficiency in satellite networks. However, existing methods struggle to effectively address the three major challenges of LEO networks: highly complex temporal dynamics caused by satellite cross-regional movement, multivariate dependencies in multi-satellite collaboration, and strong spatial heterogeneity driven by user distribution, human activity intensity, and local geographic environments. In this article, we propose a LEO Satellite Traffic Predictor (LEOSTP) framework, a diffusion model-based end-to-end model that forecasts future satellite traffic by jointly leveraging historical traffic patterns and contextual characteristics of the corresponding service regions. The framework consists of two core modules: 1) The general traffic feature extractor module combines the diffusion process with a Transformer architecture to model the multi-scale temporal features of the traffic itself. 2) The external condition encoder module integrates geographic semantic information such as population distribution, point-of-interest (POI) distribution, and local time into the prediction process through a Transformer-based encoder. In this way, the model captures the deep correlation between the external environment and traffic dynamics. Experimental results based on large-scale simulated constellation data show that LEOSTP significantly outperforms traditional statistical models such as ARIMA and SVR, and classical sequence models including LSTM and Transformer, in prediction accuracy. Index Terms—Mobile traffic prediction, Non-terrestrial networks, Spatio-Temporal Learning.
I. I NTRODUCTION Ver the past several decades, with the continuous evolution of global mobile communication networks, people have placed higher expectations on ubiquitous broadband access. Enabled by advances such as low-cost satellite manufacturing, reusable launch vehicles, and rapid progress in spacebased communication technologies, Low Earth Orbit (LEO) satellite networks are emerging as a pivotal infrastructure for future ubiquitous wireless access. These developments position LEO constellations not only as a complementary layer to terrestrial networks, but also as a critical component in delivering seamless, wide-area, and resilient broadband
O
S. Ao, Y. Niu (Corresponding author), C. Li, B. Ai are with the School of electronic and Information Engineering and the Beijing Engineering Research Center of High-speed Railway Broadband Mobile Communications, Beijing Jiaotong University, Beijing 100044, China. Z. Han is with the Department of Electrical and Computer Engineering at the University of Houston, Houston, TX 77004 USA, and also with the Department of Computer Science and Engineering, Kyung Hee University, Seoul, South Korea, 446-701.
services worldwide. Accurately forecasting traffic demand on satellite links is becoming increasingly vital for delivering stable and high-quality satellite Internet services. For example, if the network scheduling system can anticipate variations in the service load of each satellite over a future period, it can proactively plan inter-satellite forwarding and dynamically allocate network resources. In addition, advance traffic awareness enables more effective scheduling of ground gateway stations, thereby avoiding congestion in hotspot areas and improving overall quality of service (QoS) and resource utilization efficiency [1], [2]. More importantly, accurate traffic and load prediction has direct implications for practical satellite control and scheduling. Since LEO satellites spend much of their orbital period serving low-demand remote regions and only briefly traverse high-activity areas, advance knowledge of traffic demand enables region-aware control strategies, such as conservative resource use in sparse areas and proactive preparation for peak loads [3]. Similar to how traffic prediction supports energy-efficient operations in terrestrial networks, accurate satellite traffic forecasting serves as an upstream decision input for routing, resource allocation, and network scheduling, making it essential for efficient LEO network operation. Existing research typically treats network traffic prediction as a general time-series forecasting problem and analyzes it using statistical or deep learning methods. To improve the accuracy of traffic forecasting, extensive work has been conducted along this line. Traditional methods such as Seasonal Autoregressive grated Moving Average (SARIMA) and Support Vector Regression (SVR) exhibit a certain degree of nonlinear fitting capability, but they struggle to capture spatial correlations, which are both common and crucial in mobile networks [4]. Many researchers have also employed machine learning techniques to model traffic prediction problems. To overcome the limitations of traditional methods, recurrent neural network (RNN) and long short-term memory network (LSTM) have been introduced to capture the distribution and trends of cellular traffic sequences [5]–[7]. Meanwhile, researchers have recognized the spatial correlation in satellite traffic and adopted methods such as graph convolutional networks (GCNs) to incorporate graph-structured data and exploit the spatial dependencies between nodes, thereby enhancing the predictive performance of the model [8]–[11]. In [12], a spatiotemporal graph attention and gated convolutional network prediction model based on spatio-temporal feature extraction (ST-GAGCN) is proposed, which extracts the most critical features in LEO satellite traffic prediction, thereby improving its performance. These models integrate temporal and spatial
2