基于Mixup数据增强的可再生能源场景生成方法

Renewable Energy Scenario Generation Method Based on Mixup Data Enhancement Technology

摘要: 随着可再生能源渗透率的持续提升，其出力不确定性所带来的挑战也日益凸显。如何精准刻画可再生能源的不确定性，为电力系统安全运行、调度以及规划等相关决策提供科学的数据支撑已成为当前研究的热点之一。生成式对抗网络(Generative Adversarial Networks, GAN)因其强大的特征提取能力和生成能力被广泛应用于可再生能源不确定性建模任务中。然而研究表明GAN的训练稳定性、在小样本数据集上的生成效果、对风光出力数据的特征提取能力等存在不足。本文针对以上存在的问题，深入研究了改进GAN模型的搭建以及Mixup数据增强策略在生成模型中的植入方法采用了Wasserstein条件生成式对抗网络(Wasserstein Conditional Generative Adversarial Networks, WCGAN)提升模型的训练稳定性，并且可以生成特定标签的可再生能源场景。并且在生成模型数据处理阶段引入Mixup数据增强策略，通过对输入的原始样本数据以及其标签信息进行线性插值，混合形成新的样本，增扩模型的输入数据，使得生成模型能够在只有小样本数据量作为输入时依然能够训练达到拟合。本文提出的基于WCGAN-GP与Mixup的可再生能源场景生成模型相比于传统GAN模型，其训练过程更加稳定，并且能够在原始输入数据量较小的极端情况下，也能充分捕捉可再生能源出力的时空特性，生成高质量的可再生能源场景。

Abstract: The penetration rate of renewable energy, mainly composed of clean energy such as wind and photovoltaic, in the power system is constantly increasing. The renewable energy output in the power system exhibits highly intermittent, stochastic, and fluctuating characteristics, which will have a significant impact on the stable operation of the power system. At the same time, the output of renewable energy under high penetration rates will also pose higher requirements for the optimization and control of power system scheduling. How to accurately characterize the output characteristics of renewable energy has become a hot topic in current research. This article will combine Wasserstein Conditional Generative Adversarial Network (WCGAN) with self-attention mechanisms data augmentation technology, utilizing the powerful generation ability of WCGAN and the excellent performance of self-attention mechanisms data augmentation in small data training to learn the unknown distribution of renewable energy historical data and generate new samples that conform to the distribution rules of observed data. Accurately characterize the uncertainty of renewable energy output. After comparative experiments, the method proposed in this article has better generation quality on small sample data.