计及漏判/误判代价的两阶段电力系统暂态稳定预测方法

基于人工智能的电力系统暂态稳定预测方法会出现漏判（将失稳样本错误分类成稳定样本）和误判（将稳定样本错误分类成失稳样本）的现象，使得该方法不易在工程实践中应用。为此，文中基于集成卷积神经网络（CNN）提出了一种计及漏判/误判代价的两阶段电力系统暂态稳定预测方法。在第1阶段，利用滑动时间窗输入特征训练得到不同响应时间层次的集成CNN模型，建立各层输出结果的可信度指标，将可信度阈值优化选择问题转化成多目标优化问题，最大限度地减少甚至消除漏判，并尽可能早地输出可信度高的样本；在第2阶段，对分层预测阶段预测的失稳样本采用多判据融合的紧急控制启动策略，尽可能减少误判所带来的实际损失。仿真算例分析表明，文中所提方法可以以最小代价最大限度地减少甚至消除漏判，以提高人工智能暂态稳定预测结果在工程上应用的可能性。

Two-stage Transient Stability Prediction Method of Power System Considering Cost of Misdetection and False Alarm

There exist misdetection (misclassification of unstable samples into stable samples) and false alarm (misclassification of stable samples into unstable samples) by the transient stability prediction method based on artificial intelligence, which is a major obstacle to practical engineering applications. In response to this deficiency, this paper proposes a two-stage power system transient stability prediction method based on the convolutional neural network (CNN) considering the cost of misdetection and false alarm. At the first stage, the corresponding sliding time window input features are trained to obtain different layers of integrated CNN models, and the credibility index for each output layer is established. Then, the credibility threshold optimization selection problem is transformed into a multi-objective optimization problem. This stage could minimize or even eliminate the misdetection and output credible samples with high credibility as soon as possible. At the second stage, an emergency control start-up strategy based on multi-criteria fusion for the credible unstable samples predicted at the hierarchical prediction stage is proposed to reduce the actual loss caused by false alarm. The analysis of a simulation system shows that the proposed method can minimize or even eliminate the misdetection at the minimum cost, and promote the probability of practical engineering application of transient stability prediction results based on artificial intelligence method.