Battery electric vehicle charging load forecasting using LSTM on STL trend, seasonality, and residual decomposition

Syahrizal, Salleh and Roslinazairimah, Zakaria and Siti Roslindar, Yaziz (2024) Battery electric vehicle charging load forecasting using LSTM on STL trend, seasonality, and residual decomposition. In: Recent Advances on Soft Computing and Data Mining. Lecture Notes in Networks and Systems, 1078 . Springer Cham, Switzerland, pp. 327-336. ISBN 978-3-031-66965-1

Preview	Pdf Battery electric vehicle charging load forecasting using LSTM on STL trend_ABST.pdf Download (133kB) | Preview
	Pdf Battery electric vehicle charging load forecasting using LSTM on STL trend.pdf Restricted to Repository staff only Download (2MB) | Request a copy
Preview	Pdf Recent Advances on Soft Computing and Data Mining.pdf Download (281kB) | Preview

Abstract

To overcome the challenge of limited high-resolution Battery Electric Vehicle (BEV) charging data, a unique feature engineering technique was implemented. The start-stop electricity charging data from the My Electric Avenue project underwent transformation into a count of concurrent active charging events at 1-min intervals. In an effort to enhance prediction accuracy, the transformed BEV charging data was subjected to decomposition into its trend, seasonality, and residual components using the Seasonal-Trend decomposition using Loess (STL) procedure. Acknowledging the nonlinear, dynamic, and noisy characteristics inherent in BEV charging behavior, the Long Short-Term Memory (LSTM) network was chosen to model electricity demand arising from multiple concurrent BEV charging events. In the model optimization process, hyperparameter tuning focused on adjusting the number of epochs at intervals of 1, 10, 20, 30, 40, and 50. The prediction values for the STL-LSTM decomposed trend, seasonality, and residual components achieved their lowest Mean Absolute Percentage Error (MAPE) values against decomposed testing data at 0.03%, 4.37%, and 27.11%, respectively. The corresponding Root Mean Squared Error (RMSE) values were 0.01, 0.08, and 0.49. Upon reconstruction and comparison against observed testing data, the resulting lowest MAPE occurred at epochs 10, 30, and 40 for trend, seasonality, and residuals, respectively, with values of 1.21% and RMSE 0.51. This is marginally lower than the forecast using the LSTM model on observed data, which recorded a MAPE of 1.38% at RMSE 0.51. The findings underscore the suitability of the STL-LSTM model, tailored for 1-min resolution electricity demand, for electric utility companies aiming to forecast very short-term loads.

Item Type:	Book Chapter
Additional Information:	Indexed by Scopus
Uncontrolled Keywords:	Battery Electric Vehicle; STL Decomposition; Long Short-Term Memory; Feature Engineering; Charging Behavior
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science T Technology > TK Electrical engineering. Electronics Nuclear engineering
Faculty/Division:	Institute of Postgraduate Studies Center for Mathematical Science
Depositing User:	Mrs Norsaini Abdul Samat
Date Deposited:	05 Sep 2024 00:49
Last Modified:	05 Sep 2024 00:49
URI:	http://umpir.ump.edu.my/id/eprint/42504
Download Statistic:	View Download Statistics

Actions (login required)

View Item