基于生成对抗网络的小样本数据生成技术研究

基于数据驱动的深度学习技术成为新一代智能电网的应用趋势，该技术对电网中有标注训练数据的量级提出更高的要求。为了获取更多有标注的智能电网样本数据，文章提出了一种基于改进的生成对抗网络（generative adversarial network，GAN）的训练样本生成算法。该方法通过交替训练改进GAN的生成模型与判别模型，无需先验知识的指导，自主学习原始样本的分布规律，生成新的数据样本。然后采用人工神经网络作为基础分类器，计算样本分类的准确率，检验生成样本的有效性。实验表明，改进GAN模型可以有效学习样本的分布规律，提升谐波分类的准确率，该方法同时具有良好的抗噪性和泛化性，对深度学习技术在智能电网中的深入发展具有重要意义。

Research on Generation Technology of Small Sample Data Based on Generative Adversarial Network

Data-driven deep learning technology has become the application trend of the new generation of smart grids， which puts higher demands on the magnitude of the labeled training data in the grid. In order to obtain more labeled smart grid sample data， this paper proposes a training sample generation algorithm based on improved generative adversarial network. The method improves the generation model and discriminant model of GAN through alternate training， and does not need the guidance of prior knowledge to autonomously learn the distribution law of the original samples and generate new data samples. Then the artificial neural network is used as the basic classifier to calculate the accuracy of the sample classification and verify the validity of the generated samples. Experiments have shown that the improved GAN model can effectively learn the distribution law of samples and improve the accuracy of harmonic classification. This method has good anti-noise ability and generalization， which is of great significance for development of deep learning technology in smart grid.