Cognitive　Radio　(CR)　and　Energy　Harvesting　(EH)　techniques　have　offered　insights　to　mitigate　issues　related　to　inefficient　spectrum　utilization　and　limited　energy　storage　capacity.　In　Cognitive　Radio　Networks,　security　threats,　particularly　from　eavesdroppers,　may　result　in　information　leakage.　This　study　focuses　on　enhancing　the　Physical　Layer　Security　(PLS)　of　multi-users　with　EH　by　employing　cooperative　jamming　via　a　Autonomous　Aerial　Vehicle　(AAV)　to　maximize　the　secure　communication　rate.　In　the　AAV-assisted　EH-CR　system,　Secondary　Users　(SUs)　can　utilize　the　licensed　spectrum　band　occupied　by　a　Primary　User　(PU)　if　the　cooperative　jamming　power　from　SUs　to　the　PU　remains　below　a　certain　threshold.　SUs　can　harvest　and　use　Radio　Frequency　(RF)　energy　from　the　Primary　Transmitter　(PT).　The　AAV　jammer　disrupts　the　eavesdropper　by　transmitting　jamming　signals,　thereby　minimizing　stolen　information　to　optimize　long-term　secure　communication　performance.　The　paper　formulates　the　problem　of　maximizing　the　average　secure　communication　rate　while　considering　system　constraints　and　jointly　optimizes　the　AAV　trajectory,　transmission　power,　and　EH　coefficient.　As　the　problem　is　non-convex,　it　is　reformulated　as　a　Markov　Decision　Process　(MDP).　The　paper　employs　the　Multi-Agent　Proximal　Policy　Optimization　(MAPPO)　algorithm　to　address　the　problem,　introduces　counterfactual　baselines　to　tackle　the　credit　assignment　problem　in　centralized　learning,　and　integrates　the　Long　Short-Term　Memory　(LSTM)　network　to　enhance　the　learning　capability　of　sequential　sample　data,　thereby　improving　the　training　efficiency　and　effectiveness　of　the　algorithm.　Simulation　results　demonstrate　the　effectiveness　and　superiority　of　the　proposed　method　in　maximizing　the　system＇s　secure　communication　rate.