Data‐driven transient frequency stability assessment: A deep learning method with combined estimation‐correction framework

Transient frequency stability assessment (TFSA) is important for operators to understand dynamic power system frequency characteristics. However, because of the problem of combinatorial explosion, exacerbated by uncertain renewable generation and various contingencies, applying traditional time‐doma...

Full description

Saved in:
Bibliographic Details
Published inEnergy conversion and economics Vol. 1; no. 3; pp. 198 - 209
Main Authors Wen, Yunfeng, Zhao, Rongzhen, Huang, Mingzeng, Guo, Chuangxin
Format Journal Article
LanguageEnglish
Published Singapore John Wiley & Sons, Inc 01.12.2020
Wiley
Subjects
Accuracy
Algorithms
Artificial neural networks
Combinatorial analysis
Computer simulation
Deep learning
Domains
Error analysis
Error correction
Error reduction
Feature extraction
Frequency stability
Machine learning
Neural networks
Parameter estimation
Parameter modification
Support vector machines
Online AccessGet full text

Cover

More Information
Summary:Transient frequency stability assessment (TFSA) is important for operators to understand dynamic power system frequency characteristics. However, because of the problem of combinatorial explosion, exacerbated by uncertain renewable generation and various contingencies, applying traditional time‐domain simulation methods is very time‐consuming for TFSA of a low‐inertia power system. To address this, this study evaluates a fast and online TFSA using a data‐driven tool, based on deep learning. A novel combined estimation–correction learning framework is proposed. In the estimation step, a deep neural network is used to obtain the network parameters based on actual input–output feature data, which effectively realize automatic dimension reduction and feature extraction. In the subsequent correction step, the multidimensional frequency metrics produced by the deep neural network‐based estimator are input to a stacked extreme learning machine‐based corrector to further reduce the prediction error. A case study on a modified IEEE RTS‐79 system demonstrates that the proposed approach has extremely high computation speed, compared to the time‐domain simulation method, and achieves higher prediction accuracy and superior generalisation performance in comparison with other state‐of‐the‐art algorithms.
ISSN:2634-1581
2634-1581
DOI:10.1049/enc2.12015