Abstract The critical clearing time (CCT) is one of the most important indexes for large-disturbance rotor angle stability margin evaluation. In practice, model-driven methods are usually realized based on simplified models to ease the computational burden, but the accuracy is sacrificed. To solve this problem, a data-driven method is adopted in this paper for fast error correction of a model-driven method, creating an integrated method. Both a reliable accuracy and an acceptable computation speed can be achieved with this integrated method. Meanwhile, involvement of model-driven method helps enhance robustness of the integrated method to training sample insufficiency, measurement error and power system scale. In addition, the data-driven method is further transformed on the basis of a cost-sensitive approach where the error tolerance for different actual CCT values should be differentiated during the training process instead of being treated equally in the common data-driven method. To mitigate the negative effect caused by such transformations, an ensemble learning structure is also constructed. In this paper, an integrated extended equal-area criterion (IEEAC) and an extreme learning machine (ELM) are applied as model-driven and data-driven methods, respectively. A genetic algorithm (GA) is used in the ensemble learning structure construction. Validations show that the proposed integrated method with the transformed data-driven method can improve the CCT prediction accuracy and avoid the polarization of the error distribution.

中文翻译：

---

一种基于模型驱动和集成成本敏感数据驱动方案的临界清除时间预测综合方法

摘要 临界清除时间（CCT）是大扰动转子角稳定裕度评估的重要指标之一。在实践中，模型驱动的方法通常是基于简化模型来实现的，以减轻计算负担，但牺牲了准确性。为了解决这个问题，本文采用数据驱动的方法对模型驱动的方法进行快速纠错，创造了一种集成方法。使用这种集成方法可以获得可靠的精度和可接受的计算速度。同时，模型驱动方法的参与有助于增强集成方法对训练样本不足、测量误差和电力系统规模的鲁棒性。此外，数据驱动方法在成本敏感方法的基础上进一步转换，在训练过程中应区分不同实际 CCT 值的容错，而不是在常见的数据驱动方法中平等对待。为了减轻这种转换造成的负面影响，还构建了一个集成学习结构。在本文中，集成扩展等面积准则（IEEAC）和极限学习机（ELM）分别被用作模型驱动和数据驱动的方法。在集成学习结构构建中使用遗传算法（GA）。验证表明，所提出的集成方法与转换数据驱动方法可以提高 CCT 预测精度并避免误差分布的极化。