基于数据增强和深度残差网络的电力系统暂态稳定预测

针对传统数据驱动的电力系统暂态稳定分析方法中，较少考虑输入数据存在噪声和信息缺失后对预测模型性能的影响问题，提出一种基于数据增强和深度残差网络的暂态稳定预测方法。首先，考虑噪声和信息缺失情况，对原始训练数据进行扩充；然后，采用发电机受扰后动态数据作为输入特征；考虑到输入的高维时序数据具有图像的特点，利用图像处理中一种特殊的卷积神经网络—深度残差网络构建用于暂态稳定评估的深层模型。算例分析表明，所提出的方法能够提高模型的泛化能力，在含噪声以及部分发电机信息缺失情况下具有更好的鲁棒性。

Transient Stability Prediction of Power Systems Based on Deep Residual Network and Data Augmentation

In traditional data-driven power system transient stability assessment methods, the impact of noise in the collected data and the information missing problems are rarely considered for the transient stability prediction. To deal with these problems, this paper presents a method for transient stability prediction based on data augmentation and deep residual network (ResNet). Firstly, the original training data is extended with consideration of the noise data and data-missing conditions. Then, the real-time data of the disturbed generator is used as input features. Considering the similarity between high-dimensional time series data and images, the deep residual network, which is an improved algorithm based on convolutional neural networks, is used to build transient stability assessment model. The case studies show that the proposed method can improve the generalization ability of the model, and has better robustness in dealing with noise data or data missing problems.