A Method for Smart Meter Fault Data Augmentation Based on Improved Wasserstein Generative Adversarial Network

Fault data of smart meters is difficult to obtain in large quantities, which often leads to problems such as overfitting during the training of neural network models. Generative Adversarial Networks (GANs) have been widely applied in the field of data augmentation. Wasserstein Generative Adversarial Networks (WGANs) have been developed to improve upon traditional GAN. However, WGAN still suffers from challenges such as the reliance on empirical tuning of the penalty coefficient. This paper introduces the structure and typical fault of smart meter, proposes an improvement to WGAN by dynamically adjusting the penalty coefficient based on layer-wise penalty and generator-feedback penalty, to achieve data augmentation for smart meter fault data. A fault rate prediction model for smart meters based on the LightGBM model is employed to compare the training performance before and after data augmentation. The results demonstrate that the accuracy of smart meter fault prediction is significantly improved. This validates the effectiveness of the proposed method for smart meter fault data augmentation.