基于生成对抗网络与深度学习的少数据云资源预测

精确的云资源预测对计算平台实现安全运行具有十分重要的意义，针对新公司的云计算资源缺乏足够数据样本而造成预测模型精度降低的问题，本文提出一种基于WasserStein生成对抗网络和双向门控循环单元网络的少样本云计算资源预测模型。通过生成对抗网络去学习原始少样本数据的分布规律，以高斯噪声作为输入生成与原始数据具有相同分布的新样本数据，实现数据增强的行为；由于传统门控单元网络无法完全利用数据的时间信息，本文采用双向门控循环单元网络对数据的前向、反向时间信息进行双向提取并预测。以Google公开数据集进行仿真，对无增强数据和增强数据后的不同机器算法模型的预测结果进行对比。实验结果表明，使用WasserStein生成对抗网络数据增强后的双向门控循环单元网络模型精度的达到98.3%，所提方法适用于少样本数据的云计算资源预测。

Few data cloud resource prediction based on generative adversarial network and deep learning

Accurate cloud resources prediction for computing platform to realize the safe operation is of great significance. For the cloud computing resources lack of enough data of the new company and reduce the precision of forecasting model problem, this paper proposes a cloud computing resources prediction model based on WasserStein generative adversarial network with gradient penalty (WGAN-GP) and bidirectional gate recurrent unit (BiGRU). The WGAN-GP is used to learn the distribution of the original sample, and Guassian noise is used as input to generate new sample data with the same distribution as the original data, so as to realize the behavior of data enhancement. Because the traditional GRU cannot fully utilize the time information of the data, the bidirectional GRU is adopted to extract and predict the forward and reverse time information of the data. Based on the cloud resource data of a certain company, the prediction results of different machine algorithm models without enhanced data and with enhanced data are compared. The experimental results show that the accuracy of the BiGRU model enhanced by WGAN-GP data reaches 98.3%, and the proposed method is suitable for cloud computing resource prediction with few sample data.