Periodformer: An Energy Consumption Prediction Model Based on Decomposition of Time Series
In recent years, deep learning technology has demonstrated remarkable potential across various prediction tasks. However, existing deep learning models still fall short in fully exploiting the periodicity, trends, and residual characteristics inherent in energy consumption data. To address these def...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | zho |
| Published: |
Editorial Office of Control and Information Technology
2025-06-01
|
| Series: | Kongzhi Yu Xinxi Jishu |
| Subjects: | |
| Online Access: | http://ctet.csrzic.com/thesisDetails#10.13889/j.issn.2096-5427.2025.03.007 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In recent years, deep learning technology has demonstrated remarkable potential across various prediction tasks. However, existing deep learning models still fall short in fully exploiting the periodicity, trends, and residual characteristics inherent in energy consumption data. To address these deficiencies, this paper proposes a novel prediction model called Periodformer. The model begins by decomposing time series into three components: trend, period, and residual. Each component is modeled separately, and the prediction results from these models are then integrated, leading to significantly improved prediction accuracy. Experimental results showed that Periodformer achieved reductions in both Mean Absolute Error (MAE) and Mean Squared Error (MSE) of 5.56% and 11.85%, respectively, compared to the existing Transformer model, while exhibiting strong robustness against data noise. |
|---|---|
| ISSN: | 2096-5427 |