This letter investigates a reconfigurable intelligent surface (RIS)-assisted overlay cognitive radio network (CRN) that incorporates rate splitting multiple access (RSMA) and energy harvesting (EH) schemes to provide resilience during transmission power outages. In particular, we focus on maximizing the CRN’s throughput using the joint allocation of cooperation slot, power-splitting factor, and beamforming design at the cognitive base station (CBS) and RIS subject to the CBS’s power budget, primary user (PU) cooperation rate, and quality of service for the secondary users (SUs). To tackle the non-convexity of the formulated time-variant resource allocation problem, we adopt an algorithm based on modified proximal policy optimization (MPPO) that dynamically adjusts the penalty coefficient to ensure better control of the optimization process. Simulation results demonstrate that the MPPO algorithm outperforms conventional deep learning algorithms regarding CRN’s sum-throughput maximization. Moreover, the proposed RIS-aided RSMA framework offers a relatively high data rate compared to traditional schemes.