Multivariant Short-Term Residential Load Forecasting with Dual Path Trigonometric Encoding LSTM and Attention Mechanism

Manthila Wijesooriya Mudiyanselage*, Haimeng Wu, Abbas Mehrabidavoodabadi

*Corresponding author for this work

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

16 Downloads (Pure)

Abstract

Accurate short-term residential load forecasting is essential for the efficient operation of power systems. However, existing forecasting methods face several challenges, including the inability to effectively handle the rapid accumulation of residential load data, the complexity of capturing patterns and dependencies, and the limited consideration of relative factors influencing load behaviours. This paper proposes a novel approach for short-term residential load forecasting that integrates Long Short-Term Memory (LSTM) and attention mechanism (AM) with trigonometric encoding to capture periodic and seasonal variations. Additionally, Season Trending Loess (STL) decomposition method is applied to identify seasonal trends in the data. Key input features include historical load data, price data, and solar and wind generation data. The dual-path mechanism optimizes system performance using mean squared error (MSE) for the proposed ensembled method. This approach provides highly accurate short-term load forecasts, enhancing power system management. Finally, the proposed model achieves significant improvements, with gains of 28.71%, 15.57%, and 17.99% in MSE, RMSE, and MAE, respectively, compared to existing methods.
Original languageEnglish
Title of host publication2024 4th International Symposium on Electrical, Electronics and Information Engineering (ISEEIE)
Place of PublicationPiscataway, US
PublisherIEEE
Pages450-456
Number of pages7
ISBN (Electronic)9798350355772
ISBN (Print)9798350355789
DOIs
Publication statusPublished - 28 Aug 2024
Event4th International Symposium on Electrical, Electronics and Information Engineering - University of Leicester, Leicester, United Kingdom
Duration: 28 Aug 202430 Aug 2024
https://conferences.ieee.org/conferences_events/conferences/conferencedetails/62461

Conference

Conference4th International Symposium on Electrical, Electronics and Information Engineering
Abbreviated titleISEEIE 2024
Country/TerritoryUnited Kingdom
CityLeicester
Period28/08/2430/08/24
Internet address

Keywords

  • short-term residential load forecasting
  • trigonometric encoding
  • STL Decomposition
  • LSTM
  • AM

Cite this