基于边界强化混合采样的两阶段电力系统暂态稳定评估

受制于样本固有的不平衡性，基于数据挖掘的暂态稳定预测方法不易用于工程实践，为此，提出一种基于边界强化混合采样的两阶段暂态稳定评估模型。在第1阶段，利用预训练的级联卷积神经网络模型确定边界和非边界样本集，利用条件生成对抗网络合成边界集失稳样本，并对非边界集稳定样本进行欠采样，以实现边界强化；在第2阶段，利用混合采样后的重构样本集再训练卷积神经网络模型，以更好地挖掘失稳样本的隐含特征，并采用改进后的焦点损失函数加强模型对边界集样本的学习能力。新英格兰39节点系统与南方某省级电网的仿真结果表明，所建模型有效降低了对失稳样本的漏判率，提高了整体预测精度，在样本极不平衡的情况下仍有良好的评估性能。

Two-stage transient stability assessment of power system based on boundary enhanced hybrid sampling

Due to the inherent imbalance of samples, the transient stability prediction method based on data mining is tough to be used in engineering practice, for which, a two-stage transient stability evaluation model based on boundary enhanced hybrid sampling is proposed. In the first stage, the pre-trained cascaded convolutional neural network model is used to determine the boundary and non-boundary sample sets, the conditional generative adversarial network is used to synthesize the unstable samples of boundary set, and the stable samples of non-boundary set are under sampled to achieve boundary reinforcement. In the second stage, the convolutional neural network model is retrained with the reconstructed sample set after hybrid sampling to better mine the hidden features of unstable samples, and the improved focal loss function is used for enhancing the learning ability of the model to the samples of boundary set. The simulative results of New England 39-bus system and a provincial power grid in Southern China show that the proposed model can effectively reduce the omission rate of unstable samples, improve the overall prediction accuracy, and still have good evaluation performance in the case of extremely unbalanced samples.