|
| Short-term load forecasting for dedicated transformer users based on diffusion data augmentation and Transformer |
| DOI：10.19783/j.cnki.pspc.260191 |
| Key Words:short-term load forecasting dedicated transformer users diffusion model data augmentation Transformer framework |
| Author Name | Affiliation | | LI Jiawei | 1. Zhuzhou Power Supply Company, State Grid Hunan Electric Power Co., Ltd., Zhuzhou 412000, China
2. State Key Laboratory of Disaster Prevention & Reduction for Power Grid, Changsha University of Science & Technology, Changsha 410114, China | | YANG Aiwen | 1. Zhuzhou Power Supply Company, State Grid Hunan Electric Power Co., Ltd., Zhuzhou 412000, China
2. State Key Laboratory of Disaster Prevention & Reduction for Power Grid, Changsha University of Science & Technology, Changsha 410114, China | | ZHOU Boyu | 1. Zhuzhou Power Supply Company, State Grid Hunan Electric Power Co., Ltd., Zhuzhou 412000, China
2. State Key Laboratory of Disaster Prevention & Reduction for Power Grid, Changsha University of Science & Technology, Changsha 410114, China | | ZOU Yina | 1. Zhuzhou Power Supply Company, State Grid Hunan Electric Power Co., Ltd., Zhuzhou 412000, China
2. State Key Laboratory of Disaster Prevention & Reduction for Power Grid, Changsha University of Science & Technology, Changsha 410114, China | | YIN Youpeng | 1. Zhuzhou Power Supply Company, State Grid Hunan Electric Power Co., Ltd., Zhuzhou 412000, China
2. State Key Laboratory of Disaster Prevention & Reduction for Power Grid, Changsha University of Science & Technology, Changsha 410114, China |
|
| Hits: 7 |
| Download times: 1 |
| Abstract:Dedicated transformer users are characterized by strong abrupt load fluctuations and high noise. In practical metering operations, field personnel face significant challenges, including uncertainty in load conditions, difficulty in conducting on-site energy meter calibration during high-load periods, and low operation and maintenance efficiency. To address these issues, this paper proposes a short-term load forecasting method for dedicated transformer users based on diffusion data augmentation and an improved Transformer model. First, the Pearson correlation coefficient combined with the RReliefF algorithm is used for feature selection to eliminate irrelevant and redundant variables. Second, a denoising diffusion probabilistic model (DDPM) is constructed based on an improved temporal 1D-UNet to learn the latent spatial distribution of load data and generate high-quality augmented samples. Finally, an improved Transformer forecasting model integrating the long short-term memory (LSTM) coding layer and a gated linear unit (GLU) is developed to simultaneously capture the short-term temporal dependencies and long-period comprehensive load characteristics. Simulation results based on the measured dataset of dedicated transformer users in Hunan Province and a public industrial load dataset from South Korea show that the proposed method outperforms traditional forecasting models across multiple evaluation metrics. Ablation experiments and data distribution fitting analysis fully verify the effectiveness of the diffusion data augmentation module. The proposed method can accurately predict the load operation state of dedicated transformer users and provide reliable decision-making support for scheduling on-site metering operations. |
| View Full Text View/Add Comment Download reader |
|
|
|