Data-Driven Beam-to-Gateway Matching and Adaptive Load Balancing for Multi-Beam GEO Satellite Systems

Resource allocation, traffic steering, and load balancing in multi-gateway, multi-beam geostationary (GEO) satellite communication (SatCom) networks, particularly in the extremely high frequency (EHF) bands (Ka, Q, and V), face critical challenges due to dynamic traffic variations, unpredictable rain attenuation, and stringent Quality of Experience (QoE) requirements. This paper presents a novel dynamic beam-to-gateway matching and load balancing framework designed to optimize system capacity utilization by adaptively responding to real-time forecasts of traffic demand and rain attenuation. To achieve this, we first formulate a statistical optimization problem for adaptive gateway load balancing. A stacked, multi-horizon, multi-output Long Short-Term Memory (LSTM) model predicts both traffic demand and rain attenuation to inform the optimization. Given the stochastic nature of QoE, blocking probabilities, and binary beam-gateway assignments, the resulting optimization problem is non-convex. To address this, we apply queuing approximation to model packet queue dynamics, making the problem more tractable. The optimization is then solved in two stages: a linear optimization for capacity allocation, followed by a linear integer program (LIP) for adaptive beam-to-gateway matching to achieve load balancing. To circumvent the exponential complexity of the LIP, we design a low-complexity heuristic based on an Adaptive Genetic Algorithm (AGA). The simulation results demonstrate that the proposed approach effectively balances network load and optimizes gateway capacity utilization, outperforming benchmark techniques.