基于生成对抗网络的电能质量扰动数据增强方法

Power Quality Disturbance Data Enhancement Method Basedon Generative Countermeasure Network

针对当前电能质量扰动自动识别受样本集的规模和质量影响较大及扰动数据匮乏的问题,提出一种在二维尺度上结合深度卷积生成对抗网络(deep convolutional generative adversarail networks,DCGAN)对电能质量扰动数据进行增强的方法。将典型扰动二维图像数据作为输入,以提高数据特征提取能力,再通过深度卷积生成对抗网络不断生成优化扰动数据,并选择验证集上取得最高AUC值的增强数据集进行电能质量扰动的识别测试。在某电网公司提供的真实数据集上进行测试,结果表明:基于DCGAN数据增强方法能生成较大规模、高质量的数据,在网络训练速度及电能质量扰动识别的准确率上有明显提升。

Aiming at the problem that the automatic identification of power quality disturbance was greatly affected by the scale and quality of sample set and the lack of disturbance data,a method to enhance power quality disturbance data on two-dimensional scale combined with deep convolutional generative adversarial networks(DCGAN)was proposed.The typical disturbed two-dimensional image data was taken as the input to improve the data feature extraction ability,then the anti-network was generated through deep convolution,the optimized disturbed data was continuously generated,and the enhanced data set with the highest AUC value on the verification set was selected for the identification test of power quality disturbance.The results show that the data enhancement method based on DCGAN can generate large-scale and high-quality data,and significantly improve the network training speed and the accuracy of power quality disturbance identification.