A Dual-task Prediction Model for Starlink Maritime Performance

While LEO satellite systems have revolutionized broadband access for maritime industries, the dual challenges of high orbital mobility and dynamic weather lead to unpredictable signal instability. This prevents the maritime industry from effectively managing bandwidth. To address this issue, this work developed a short-term, dual-task prediction model to predict future downlink throughput and latency simultaneously using a 15-minute prediction horizon. The model is trained using real-world experimental data, which includes recent throughput history, network-layer indicators, and environmental variables. By utilizing a shared backbone for simultaneous downlink throughput and latency forecasting, this dual-task approach outperforms single-task models. Furthermore, its computational efficiency makes it ideal for edge deployment and real-time, application-aware decision-making.