A Hybrid Data-Driven Method for Online Power System Dynamic Security Assessment with Incomplete PMU Measurements
Qiaoqiao Li, Yan Hui Xu, Chao Ren, Junhua Zhao · 2019
With widely deployment of phasor measurement units (PMU), data-driven dynamic security assessment (DSA) has been a promising tool to help power system counteract large disturbance and avoid instability. In case of PMU lost events, the DSA performance may be dramatically degraded. This paper designs a new online DSA model based on hybrid ensemble learning to address the missing data problem. In this method, a set of missing data estimators is proposed to restore the system measurement, and potential inaccurately estimated features are detected in feature validation stage to further ensure the model performance. After excluding inaccurate estimations, the filled-up dataset is provided to support accurate DSA under PMU lost situation. A hybrid ensemble of extreme learning machine (ELM) and random vector functional link networks (RVFL) is designed as learning engine for estimators and DSA classifiers. Simulation results show this approach achieves low estimation error as well as high DSA accuracy and strong robustness.