Short-term electric vehicle charging load forecasting based on TCN-LSTM network with comprehensive similar day identification

Tian, Jiarui, Liu, Hui, Gan, Wei, Zhou, Yue ORCID: https://orcid.org/0000-0002-6698-4714, Wang, Ni and Ma, Siyu 2025. Short-term electric vehicle charging load forecasting based on TCN-LSTM network with comprehensive similar day identification. Applied Energy 381 , 125174. 10.1016/j.apenergy.2024.125174 Item availability restricted.

PDF - Accepted Post-Print Version Restricted to Repository staff only until 24 December 2025 due to copyright restrictions. Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (1MB)

Abstract

With the rapid growth of the number of electric vehicles (EVs) and the considerable challenges this poses to the power distribution network, the necessity for accurate forecasting of EV charging loads has become increasingly critical. This paper proposes a novel short-term EV charging load forecasting approach that integrates temporal convolutional networks (TCN) and long short-term memory (LSTM) networks, along with comprehensive similar day identification. The combined use of TCN and LSTM allows for enhanced precise forecasting of EV charging loads, while a novel methodology is employed to identify comprehensive similar days, significantly enhancing the accuracy of EV charging load forecasts. This method incorporates both linear and nonlinear analyses through the Pearson coefficient and maximal information coefficient to identify meteorological factors that show strong correlations with the load. The forecasting accuracy is further improved by incorporating a broad spectrum of input features, including but not limited to meteorological factors, seasonal patterns, and day-type distinctions. A key component of this approach is the detailed analysis of EV charging load data from the days immediately preceding the forecast. By integrating insights from both historical data and the latest observations, the model is able to detect critical trends and anomalies. The effectiveness of the proposed method is validated using historical data from Palo Alto, USA. The effectiveness of the proposed method is validated using historical data from Palo Alto, USA. The TCN-LSTM prediction model reduces prediction error from approximately 8 % to 6 % compared to other models. Furthermore, incorporating comprehensive similar days improves performance, reducing the prediction error by an additional 2 % compared to using only meteorological similar days.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Engineering
Publisher:	Elsevier
ISSN:	0306-2619
Date of First Compliant Deposit:	31 January 2025
Date of Acceptance:	16 December 2024
Last Modified:	03 Feb 2025 12:30
URI:	https://orca.cardiff.ac.uk/id/eprint/175802

Actions (repository staff only)

Edit Item