低轨多波束卫星下行业务强度时空预测方法OA

LEO mega-constellations are revolutionizing global broadband connectivity landscape.However,constrained by the multidi-mensional scarcity of on-board power,spectrum bandwidth,and beam hopping time slots,efficient resource scheduling critically de-pends on prior knowledge of future traffic demands.Consequently,achieving robust multi-step forecasting of downlink traffic intensity at the logical beam level is indispensable for enabling proactive resource provisioning.To address the long-term periodicity and local burstiness of satellite traffic,a decoupled spatio-temporal forecasting framework is proposed.Specifically,this method utilizes data-adaptive orthogonal complex-valued bases to extract dominant temporal evolution patterns,achieving trend prediction via low-dimensional subspace projection and basis extension.Subsequently,a sparse vector autoregression model is introduced to model predic-tion residuals,explicitly capturing transient spatial couplings among beams.Experimental results demonstrate that,leveraging its com-pact representation and sparse structure,the proposed framework achieves superior prediction accuracy with significantly reduced com-putational overhead compared to existing baselines algorithms.