基于代价敏感极端学习机的电力系统暂态稳定评估方法

Power system transient stability assessment based on cost-sensitive extreme learning machine

针对电力系统暂态稳定评估中稳定样本与不稳定样本误分类代价不同的特点,提出一种基于代价敏感极端学习机的电力系统暂态稳定评估方法。该方法在现有极端学习机的基础上,引入误分类代价的概念,以误分类代价最小为目标,构造代价敏感极端学习机,克服了现有极端学习机应用于暂态稳定评估时只追求高的分类准确率而忽略不稳定样本漏报率的缺点。新英格兰39节点系统和IEEE 145节点系统的仿真结果表明,所提方法的评估结果更倾向于将样本划分为误分类代价大的不稳定样本,以减小总的误分类代价。通过调整误分类代价矩阵,不仅可以使漏报率降为0,还能使稳定样本的误报率维持在较低的水平,保证了评估结果的可靠性。

Since the misclassification cost of stability sample is different from that of instability sample in the transient stability assessment,a method of transient stability assessment based on the cost-sensitive extreme learning machine is proposed,which introduces the concept of misclassification cost and takes the minimum misclassification cost as its objective to construct the cost-sensitive extreme learning machine,avoiding the demerit of existing extreme learning machine with higher classification accuracy and ignored false dismissal rate in the transient stability assessment. The simulative results of New England 39-bus system and IEEE 145-bus system show that,the proposed method inclines to classify the samples into instability case with higher misclassification cost to reduce the overall misclassification cost. By adjusting the misclassification cost matrix,the false dismissal rate can be decreased to zero and the false dismissal rate of stability samples kept at lower level,ensuring the reliability of assessment results.