A Data Augmentation Method for Distributed Photovoltaic Electricity Theft Using Wasserstein Generative Adversarial Network

Because of the concealment of distributed photovoltaic (PV) electricity theft, the number of electricity theft samples held by the power sector is insufficient, which results in low accuracy of electricity theft detection. Therefore, this paper proposes a data augmentation method about electricity theft samples of distributed PV using Wasserstein generative adversarial network (WGAN). First, through the confrontation training of WGAN's generator network and discriminator network, the neural network can learn the time correlation of PV electricity theft data sequence that is difficult to explicitly model, and generate new electricity theft samples with similar distributions to the real electricity theft samples. Then, three typical photovoltaic power stealing models are proposed, and a convolutional neural network (CNN) is constructed based on the data features of the electricity theft samples. Finally, a practical example verifies the effectiveness and adaptability of the method. The experimental results show that WGAN can consider the shape and distribution of the samples, and the generated electricity theft samples can significantly improve the performance of different classifiers.