基于深度学习的暂态稳定评估与严重度分级

提出一种安全域概念下的堆叠降噪自动编码器和支持向量机集成模型相结合的暂态稳定评估方法。将故障前的潮流量作为输入,利用堆叠降噪自动编码器对输入量进行多层抽象表达,使用提取的各层特征训练支持向量机;建立支持向量机集成分类模型进行暂态稳定评估,对评估结果进行可信度分析,将输入空间划分为稳定区、边界区和失稳区;利用效用理论结合所提出的暂态稳定裕度指标对运行方式进行严重度分级。算例结果表明,所提暂态稳定评估方法具有更高的评估准确率和一定的泛化能力;所提严重度分级方法能够直观表现不同运行方式的危险程度。

Deep learning based transient stability assessment and severity grading

A transient stability assessment method combining SDAE(Stacked Denoising AutoEncoder) and an ensemble model based on SVMs(Support Vector Machines) is proposed under the concept of security region. The power flow data before the fault are set as inputs, from which the multi-level features are abstracted by SDAE for training SVMs. An ensemble classification model based on SVMs is established to assess the transient stability, and the confidence analysis of the results is carried out to divide the input space into stable area, boundary area and unstable area. The utility theory combined with the proposed transient stability margin index is used to grade the severity of operation modes. Case results show that, the proposed transient stability assessment method has higher accuracy and better generalization ability, and the proposed severity grading method can directly show the risk degree of different operation modes.