•  
  •  
 

Keywords

electric vehicle, charging pile, charging load, LSTM, load forecasting

Abstract

To eliminate the impact of spatial distribution uncertainty on the accuracy of ultra-short-term forecasting of electric vehicle charging load, a method based on the utilization rate of charging piles for electric vehicle charging load ultra-short-term forecasting is proposed. Firstly, the charging load power of each charging pile within the region is extracted from massive charging transaction data, and then quantified values of the utilization rate of charging piles are obtained through encoding. Then, the utilization rate of charging piles and charging load power data are merged to obtain training samples and test sets for long short-term memory (LSTM) neural networks, forming a deep learning model for ultra-short-term forecasting of electric vehicle charging load, with a time resolution of up to 0.5 h. Finally, the effectiveness and accuracy of the proposed method are validated in scenarios with different scales of charging load. The results indicate that compared to the unoptimized LSTM neural network load forecasting method, the proposed method achieves an increase in the average absolute percentage error of approximately 5%. This can provide significant support for the optimization operation of distribution grids under future vehicle-grid interaction.

DOI

10.19781/j.issn.1673-9140.2024.01.011

First Page

115

Last Page

123,133

Share

COinS