Data-Driven Approach for Generator Rejection Prediction to Prevent Transient Instability in Power System Using Wide-Area Measurements

Abstract

This paper presents a novel data-driven approach to predict generator rejection/tripping for preventing transient instability in power systems. Since calculating the total amount of generator rejection and assigning the optimal amount of tripping to each generating facility is a time-consuming process, the optimal generator tripping calculation might be impractical for a real-life interconnected power system. In addition, communication delays deteriorate the efficiency of any wide-area remedial control action (RCA) in response to fault events which quickly evolve into transient instability. The presented framework predicts the optimal generator rejection for critical generators based on voltage data of generator terminals before and after the occurrence of the contingency. To simplify the problem and enhance the prediction accuracy, the framework is designed for each transmission line independently. The proposed framework is comprised of two stages: offline optimization which involves calculating proper RCAs using a full dynamic model of the power system for training the machine learning engine, and online prediction. In the offline stage, bulk scenarios are generated for individual transmission lines, the unstable cases are determined, then the critical generator patterns and generator rejection patterns are extracted for each unstable scenario. In the online stage, the proposed framework predicts the stability status, critical generators, and the optimal amount of generator tripping for each critical generator in real-time. The performance of the proposed framework is tested on the IEEE 9-bus system and the Nordic test system. The obtained results show the effectiveness of the proposed framework in responding to critical fault events in real-time.

Steady-State Data-Driven Dynamic Stability Assessment in the Energy Management System

“…Second, the existing studies related to the data-driven model often ignore the problem of how to access and acquire the quality data because they assume that all data are readily accessible. Thus, spectrum analysis and data preprocessing vary signi cantly [24]- [25]. Since the data being used vary, numerous preprocessing steps exist to deal with an exceptional case.…”