针对低地球轨道（LEO）卫星上行通信环境高动态导致信道误差统计先验信息获取难度大的问题，提出一种上行鲁棒传输优化方法。该方法以总发射功率最小化为目标，以用户服务质量（QoS）和最大发射功率为约束，构建用户角度信息不准确场景下的上行鲁棒传输优化问题。通过将此非凸问题建模为马尔可夫决策过程，设计一种基于柔性演员-评论家（SAC）的上行多用户鲁棒波束成形（BF）与功率控制算法，使卫星无需依赖误差统计先验信息，通过与环境交互自主学习并调整波束成形权矢量与功率控制策略。仿真结果表明，相比于非鲁棒方案和近端策略优化基准方案，所提方法在不同误差场景下所需发射功率能分别平均节省59%和24%，表明所提方法的鲁棒性和优越性。

To address the challenge of obtaining prior statistical information on channel errors caused by the highly dynamic environment of Low Earth Orbit (LEO) satellite uplink communications, this paper proposes a robust uplink transmission optimization method. Aiming to minimize the total transmit power subject to User Quality of Service (QoS) and maximum transmit power constraints, a robust transmission optimization problem is formulated for scenarios characterized by inaccurate user angle information. By modeling this non-convex problem as a Markov Decision Process (MDP), a multi-user robust beamforming (BF) and power control algorithm based on Soft Actor-Critic (SAC) is designed. This approach enables the satellite to autonomously learn and adjust beamforming weight vectors and power control strategies through interaction with the environment, without relying on prior information regarding error statistics. Simulation results demonstrate that, compared with the non-robust scheme and the Proximal Policy Optimization (PPO) baseline scheme, the proposed method reduces the required transmit power by an average of 59% and 24%, respectively, under different error scenarios, thereby verifying the robustness and superiority of the proposed method.