This paper proposes a novel method for the machine learning-based online prediction of generator dynamic behavior in large interconnected power systems. Unlike the existing literature in this domain, which assumes faults occur immediately after a steady-state situation, the proposed method takes the possibility of multiple disturbances into account. It is founded on a simulation-based classification approach to indirectly take advantage of phasor measurement unit (PMU) data, which leads to improvements in robustness against load model uncertainties. Relying on offline scenarios, the method developed conducts multiple time-domain simulations (TDSs) in parallel for a set of feasible two-machine dynamic equivalent models (DEMs) for each case. Thereafter, common descriptive statistics are computed for the rotor angles obtained to form the feature space. The values taken via a feature selection process are then applied as inputs to ensemble decision trees, which train models capable of predicting both stability status and generator grouping ahead of time. In online situations, PMU data are used to create DEMs and the predictors are collected by performing parallel TDSs for DEMs. The functionality of the proposed hybrid machine learning and TDS-based approach is verified on several IEEE test systems, followed by a discussion of results.

中文翻译：

---

基于仿真的多大扰动下发电机动态行为在线预测分类方法

本文提出了一种基于机器学习的大型互连电力系统中发电机动态行为在线预测的新方法。与该领域现有文献假设故障在稳态情况后立即发生不同，所提出的方法考虑了多重干扰的可能性。它建立在基于仿真的分类方法之上，可以间接利用相量测量单元 (PMU) 数据，从而提高对负载模型不确定性的鲁棒性。依赖于离线场景，所开发的方法对每个案例的一组可行的两机动态等效模型 (DEM) 并行进行多个时域仿真 (TDS)。此后，为形成特征空间而获得的转子角度计算常见的描述性统计。然后将通过特征选择过程获取的值作为输入应用于集成决策树，从而训练能够提前预测稳定性状态和生成器分组的模型。在在线情况下，PMU 数据用于创建 DEM，并通过对 DEM 执行并行 TDS 来收集预测变量。所提出的混合机器学习和基于 TDS 的方法的功能在几个 IEEE 测试系统上得到验证，然后讨论结果。PMU 数据用于创建 DEM，并通过对 DEM 执行并行 TDS 来收集预测变量。所提出的混合机器学习和基于 TDS 的方法的功能在几个 IEEE 测试系统上得到验证，然后讨论结果。PMU 数据用于创建 DEM，并通过对 DEM 执行并行 TDS 来收集预测变量。所提出的混合机器学习和基于 TDS 的方法的功能在几个 IEEE 测试系统上得到验证，然后讨论结果。