Variable-Weighted Ensemble Forecasting of Short-Term Power Load Based on Factor Space Theory

Yundong Gu (), Dongfen Ma (), Jiawei Cui (), Zhenhua Li and Yaqi Chen
Additional contact information
Yundong Gu: North China Electric Power University
Dongfen Ma: Xinjiang University of Science and Technology
Jiawei Cui: North China Electric Power University
Zhenhua Li: North China Electric Power University
Yaqi Chen: North China Electric Power University

Annals of Data Science, 2022, vol. 9, issue 3, No 5, 485-501

Abstract: Abstract The power load forecasting plays an important role in the economical and safe operation of the modern power system. However, the characteristics of power load such as non-stationarity, nonlinearity, and multiple quasi-periodicities make power load forecasting a challenging task. The present work focuses on developing a multi-model ensemble forecasting strategy by using prediction phase space construction, similar scenario improved support vector machine, and variable-weighted ensemble method, based on the Factor Space Theory. Firstly, the concept of “Prediction Scenario” is proposed to describe the “Internal historical facts in time series form” and the “External space–time environment composed of external influence factors” of power load forecasting. Next, the candidate input features for power load forecasting are selected based on the correlation analysis between the power load to be predicted, its historical load, and external influence factors. Then, based on the Factor Space Theory, the feature description of the “Prediction Scenarios” is studied and a series of prediction phase space are constructed by randomly selecting some strongly correlated features. An improved support vector machine is proposed based on similar historical scenario screening to set up the unit prediction sub models in each corresponding prediction phase space. The performance of these models is tested by simulation experiments and the variable weight of each model is designed based on the results. Finally, the power loads are forecasted by variable weighted ensemble of multiple models in different prediction phase spaces. The results of mid-Atlantic region load forecasting analysis suggest that the proposed method has better performance in almost cases, comparing with Support Vector Machine, Recurrent Neural Network, Self-partitioning Local Neuro Fuzzy method, Random Forest, Ensemble Neuro-fuzzy method and other state of art forecasting methods.

Keywords: Short-term load forecasting; Prediction scenario; Factor space theory; Prediction phase space; Similar historical scenario; Support vector machine; Multi-model variable-weighted ensemble; 90C05 (search for similar items in EconPapers)
Date: 2022
References: View references in EconPapers View complete reference list from CitEc
Citations:

Downloads: (external link)
http://link.springer.com/10.1007/s40745-022-00398-5 Abstract (text/html)
Access to the full text of the articles in this series is restricted.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:spr:aodasc:v:9:y:2022:i:3:d:10.1007_s40745-022-00398-5

Ordering information: This journal article can be ordered from
https://www.springer ... gement/journal/40745

DOI: 10.1007/s40745-022-00398-5

Access Statistics for this article

Annals of Data Science is currently edited by Yong Shi

More articles in Annals of Data Science from Springer
Bibliographic data for series maintained by Sonal Shukla () and Springer Nature Abstracting and Indexing ().