Power system state estimation using conditional generative adversarial network

Yi He, Songjian Chai, Zhao Xu, Chun Sing Lai, Xu Xu

Research output: Journal article publicationJournal articleAcademic researchpeer-review

17 Citations (Scopus)

Abstract

Accurate power system state estimation (SE) is essential for power system control, optimisation, and security analyses. In this work, a model-free and fully data-driven approach was proposed for power system static SE based on a conditional generative adversarial network (GAN). Comparing with the conventional SE approach, i.e. weighted least square (WLS) based methods, any appropriate knowledge of the system model is not required in the proposed method. Without knowing the specific model, GAN can learn the inherent physics of underlying state variables purely relying on historic samples. Once the model has been trained, it can estimate the corresponding system state accurately given the system raw measurements, which are sometimes characterised by incompletions and corruptions in addition to noises. Case studies on the IEEE 118-bus system and a 2746-bus Polish system validate the effectiveness of the proposed approach, and the mean absolute error is <1.2 × 10−3 and 5.3 × 10−3 rad for voltage magnitude and phase angle, respectively, which indicates a high potential for practical applications.

Original languageEnglish
Pages (from-to)5816-5822
Number of pages7
JournalIET Generation, Transmission and Distribution
Volume14
Issue number24
DOIs
Publication statusPublished - 18 Dec 2020

Keywords

  • State estimation
  • Adversarial networks
  • Data-driven approach
  • IEEE 118-bus system
  • Mean absolute error
  • Power system controls
  • Power system state estimation
  • Voltage magnitude
  • Weighted least squares
  • Least squares approximations

ASJC Scopus subject areas

  • Control and Systems Engineering
  • Energy Engineering and Power Technology
  • Electrical and Electronic Engineering

Fingerprint

Dive into the research topics of 'Power system state estimation using conditional generative adversarial network'. Together they form a unique fingerprint.

Cite this