Regional Wind Power Ramp Forecasting through Multinomial Logistic Regression

Xiaomei Chen, Jie Zhao, Miao He

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Wind power ramps are the abrupt yet significant change in wind power productions. The information on the ordinal levels of impending wind power ramp could help power system operator to arm operation or ramping reserves in a timely manner. This paper presents novel approaches for regional wind power ramp level forecasting using real-time meso-scale wind speed measurements. Motivated by the correlation of the meso-scale wind speed measurements with the regional wind power data, the proposed approach utilizes multinomial logistic regression for wind power ramp forecasting. An approach that combines the probabilistic output of individual regressive models in a weighted manner is proposed, with the weights calculated by minimizing the Brier skill score of the combined model. The proposed methods are tested by using real-world data, and is compared with benchmark methods. The results reveal the effectiveness of the proposed approaches.

Original languageEnglish
Title of host publicationProceedings of the 2020 IEEE Green Technologies Conference, GreenTech 2020
EditorsPierre F. Tiako, Robert Scolli, Tom Jobe
PublisherIEEE Computer Society
Pages36-41
Number of pages6
ISBN (Electronic)9781728150178
DOIs
StatePublished - Apr 1 2020
Event2020 IEEE Green Technologies Conference, GreenTech 2020 - Virtual, Oklahoma City, United States
Duration: Apr 1 2020Apr 3 2020

Publication series

NameIEEE Green Technologies Conference
Volume2020-April
ISSN (Electronic)2166-5478

Conference

Conference2020 IEEE Green Technologies Conference, GreenTech 2020
CountryUnited States
CityVirtual, Oklahoma City
Period04/1/2004/3/20

Keywords

  • Multinomial logistic regression
  • sparse primary component analysis
  • wind power ramp forecasting

Fingerprint Dive into the research topics of 'Regional Wind Power Ramp Forecasting through Multinomial Logistic Regression'. Together they form a unique fingerprint.

Cite this