Physics-Informed Conditional GAN for Transmission Line Sag Image Generation under Varying Weather Conditions

The monitoring of sage condition in transmission line plays an important role in the stable operation of the power grid. However, it is difficult to collect sufficient transmission line images that cover different sag conditions, which reduces the generality of dataset for sag monitoring. Aiming at the data problem, this paper proposes a physics-informed conditional generative adversarial network (PICGAN) for transmission line sag monitoring. Firstly, the edge detection is utilized to detect and extract the transmission line to isolate the transmission line layer and the background layer. Secondly, the effects of multiple adverse weather conditions on transmission line sag are considered to evaluate the slackness index. Then, the PICGAN is introduced to enhance the controllability of the model by taking the original images as inputs. Further, the images and labels processed by the physical information module are inputted into the discriminator to guide the generator, in order to optimize the images of transmission lines with different labels. Finally, the generated transmission line layer is fused with the background layer to reconstruct the image. Experiments verified that the proposed method can generate transmission line images with selected sag value under different weather conditions, providing a new idea for augmentation the transmission line fault diagnosis dataset.