In this paper, a scenario generation method based on deep transfer learning is proposed, wherein knowledge transfer is performed from nearby data-rich power plants to help generate scenarios of newly-built PV plant. Within this approach, scenarios are extracted with attention modules and generated in an adversarial way. Subsequently, a transfer learning method called unsupervised adversarial domain adaptation is utilized for transferring knowledge from data-rich plant to the newly-built plant. Experimental results demonstrate that the application of knowledge transfer is effective in this case.