ShuffleTransformerMulti-headAttentionNet network for user load forecasting

Linfei Yin and Linyi Ju

Energy, 2025, vol. 322, issue C

Abstract: With the development of intelligent power load forecasting, optimizing and improving algorithms to increase the accuracy and speed of load forecasting is a key part of responding to policy development opportunities. However, since the load data is affected by many factors, the output is characterized by volatility, nonlinearity and non-smoothness, and the volume is large, and it is difficult to mine the hidden information, the stable operation and economic benefits of the power system will become a challenge. To improve data prediction accuracy and maintain power system stable operation, this study proposes a ShuffleTransformerMulti-headAttentionNet network model. The structural feature of the network model is that two CNN networks are input to a Transformer model in a parallel way. The ShuffleTransformerMulti-headAttentionNet network model is compared with 20 network models such as Shufflenet and GooleNet. To better improve the applicability of the model, we discuss small-scale, high temporal resolution and large-scale, low temporal resolution historical power load data. The feasibility and reliability of the ShuffleTransformerMulti-headAttentionNet algorithm is verified by simulating the input data of Case 1:292 × 96 and Case 2:31720 × 24, respectively. The experimental results show that the precision of the ShuffleTransformerMulti-headAttentionNet network is the highest in the comparison experiment, with root mean squared error, mean squared error, mean absolute error and symmetric mean absolute percentage error as the reference standards. In case 1, RMSE, MSE, MAE, MAPE, and SMAPE of ShuffleTransformerMulti-headAttentionNet network are optimized 20.27 %–100 %, 36.43 %–100 %, 11.51 %–100 %, 0.76 %–100 %, 5.97 %–96.00 % compared with other algorithms, respectively. In case 2, RMSE, MSE, MAE, MAPE, and SMAPE of ShuffleTransformerMulti-headAttentionNet network are optimized by 21.65 %–100 %, 38.62 %–100 %, 36.20 %–100 %, 12.72 %–100 %, 26.00 %–96.05 % compared with other algorithms, respectively.

Keywords: Deep learning; Transformer; Parallel network; Load forecasting; Shufflenet (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S036054422501179X
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:energy:v:322:y:2025:i:c:s036054422501179x

DOI: 10.1016/j.energy.2025.135537

Access Statistics for this article

Energy is currently edited by Henrik Lund and Mark J. Kaiser

More articles in Energy from Elsevier
Bibliographic data for series maintained by Catherine Liu ().