In the scenarios of traffic diagnostic analysis and prediction in optical network based on deep learning, the work of traffic data collection and storage for optical link is limited due to security and other reasons. Aiming at the problem that the amount of traffic data is too small to support deep learning training, this paper proposes an optical network traffic data synthesis algorithm based on topology link recognition. The core idea is to combine the conditional generation model based on optical network topology and the data synthesis model based on optical network traffic in the framework of generative adversarial networks. The traffic data of the given optical link is synthesized in a self-supervised way. The simulation results show that the auto-correlation coefficients of the traffic data synthesized by the proposed algorithm is close to the real data, and the accuracy of the traffic prediction model based on fully connected neural network is more than 95%.