Multi-fidelity optimization of a quiet propeller based on deep deterministic policy gradient and transfer learning