.png){kind=icon}
.png){kind=icon}
Return
EDIformer: An iTransformer-based model for electricity demand interval forecasting
.png){kind=icon}
PRE
AI
L
Y
R
Y
DOI：10.1016/j.epsr.2026.113036.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Accurate forecasting of electricity demand is critical for power system scheduling and planning. To address the challenges in UK electricity demand forecasting, including utilization of time-related features, temporal information capturing at different scales, complex interactions among multiple variables, and quantification of prediction uncertainty, this paper proposes a novel electricity demand interval forecasting model, EDIformer, based on quantile regression. It integrates a time feature encoder-decoder strategy, a pyramid GRU embedding strategy, and a multivariable router attention strategy. The first strategy focuses on exploring the relationship between time-related features and electricity demand. The second strategy is used to extract and fuse information across different scales. The last strategy captures complex interactions among variables through a dynamic routing mechanism. Ablation experiments validate the effectiveness of these strategies and show average reductions in MSE, Quantile Loss (QL), and Interval Score (IS) by 17.1 %, 10.5 %, and 10.3 %, respectively, at prediction lengths of 96, 240, and 336. Comparative experiments indicate EDIformer outperforms state-of-the-art models at different prediction lengths, and MSE and QL are 41.3 % and 39.1 % lower on average, respectively. Visual analysis shows that EDIformer has strong robustness in interval forecasting and provides a reliable decision basis for power system planning and scheduling.
Keywords:
Electricity demand interval forecasting
Time-related features
Multiscale embedding
Multivariable router attention
Quantification of uncertainty
Journal
IF:
4.2
Papers:
1.1W
Citations:
2.2W
Organization
Cited Papers