Amid the escalating demand for accessible users and security insurance in satellite aerial terrestrial integrated networks (SATINs), security and energy efficiency emerge as pivotal indicators. This paper proposes a secure beamforming scheme in reconfigurable intelligent surface (RIS) aided SATINs, in presence with multiple eavesdroppers, where rate splitting multiple access (RSMA) and RIS are adopted at the secondary UAV networks for achieving multiuser diversity and antijamming. To optimize the secrecy energy efficiency (SEE) for secondary networks while adhering to constraints on ground earth station (GES) secrecy rate, a deep reinforcement learning (DRL) framework is proposed to address the coupling between optimization variables through the improved proximal policy optimization (PPO) method, of which from existing DRL scheme is that the proposed one builds a unified learning framework. Simulation results indicate that the SEE derived by the proposed DRL scheme is superior to that of benchmark schemes, which validate the advantage of this work.