Jie Wu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Jie Wu has a total of 5 co-authors. Here are the top co-authors by number of shared publications:
- Fuhong Qiu
School of Automation and Information Engineering, Sichuan University of Science and Engineering, Yibin 644000, China
1 shared publication - Linghao Zhang
Research Institute of Electric Power Science, State Grid Corporation of Sichuan Province, Chengdu 610041, China
1 shared publication - Mingju Chen
School of Automation and Information Engineering, Sichuan University of Science and Engineering, Yibin 644005, China
1 shared publication - Senyuan Li
School of Automation and Information Engineering, Sichuan University of Science and Engineering, Yibin 644005, China
1 shared publication - Zhengwei Chang
Research Institute of Electric Power Science, State Grid Corporation of Sichuan Province, Chengdu 610041, China
1 shared publication
Latest Publications
- Open Access
- 25 June 2025
- Published by MDPI AG in Electronics
- Vol. 14 (13) , 2566
A Dual-Branch Transformer Network with Multi-Scale Attention Mechanism for Microgrid Wind Turbine Power Forecasting
Wind power generation provides clean and renewable electricity for microgrids, but its intermittency and uncertainty pose challenges to the operation and power quality of microgrids. Accurate forecasting is conducive to maintaining the stability of microgrids and improving the efficiency of energy management. Therefore, this study proposes a dual-branch frequency transformer (DBFformer), which leverages multi-scale spectral transformation and the multi-head attention mechanism to improve the prediction accuracy of microgrid wind turbines. In the encoder, two parallel branches are designed to extract the global features and local dynamic features of meteorological data based on Fourier transform and wavelet transform, respectively. In the decoder, an exponential smoothing attention (ESA) mechanism and a frequency attention (FA) mechanism are introduced to extract multi-scale temporal features. ESA enhances the model’s ability to capture long-term growth trends, whereas FA focuses on periodic pattern recognition. Additionally, to further optimize the model’s performance, a periodic weight coefficient (PWC) mechanism is employed to dynamically adjust the fusion coefficients to further improve the fusion performance and prediction accuracy. The factors influencing wind turbine power are analyzed; then, the most relevant factors are selected for the experiment. According to the experimental results, the proposed DBFformer accurately predicts the output power of wind turbines and exhibits superior performance. It achieves lower mean squared error (MSE) and mean absolute error (MAE) values than other state-of-the-art models. Specifically, its MSE values are 0.195, 0.216, 0.457, and 0.583, and the corresponding MAE values are 0.318, 0.335, 0.474, and 0.503 for different rated wind turbines. Furthermore, comprehensive ablation experiments validate that the dual-branch structure, frequency transformations, dual-attention mechanisms, and PWC module have a positive impact on the proposed model. Therefore, this research offers a novel and effective approach for wind power forecasting and supports the broader goal of integrating clean energy into microgrids.
- 25 June 2025
- Published by MDPI AG in Electronics
- Vol. 14 (13) , 2566
Open AccessA Dual-Branch Transformer Network with Multi-Scale Attention Mechanism for Microgrid Wind Turbine Power Forecasting
Wind power generation provides clean and renewable electricity for microgrids, but its intermittency and uncertainty pose challenges to the operation and power quality of microgrids. Accurate forecasting is conducive to maintaining the stability of microgrids and improving the efficiency of energy management. Therefore, this study proposes a dual-branch frequency transformer (DBFformer), which leverages multi-scale spectral transformation and the multi-head attention mechanism to improve the prediction accuracy of microgrid wind turbines. In the encoder, two parallel branches are designed to extract the global features and local dynamic features of meteorological data based on Fourier transform and wavelet transform, respectively. In the decoder, an exponential smoothing attention (ESA) mechanism and a frequency attention (FA) mechanism are introduced to extract multi-scale temporal features. ESA enhances the model’s ability to capture long-term growth trends, whereas FA focuses on periodic pattern recognition. Additionally, to further optimize the model’s performance, a periodic weight coefficient (PWC) mechanism is employed to dynamically adjust the fusion coefficients to further improve the fusion performance and prediction accuracy. The factors influencing wind turbine power are analyzed; then, the most relevant factors are selected for the experiment. According to the experimental results, the proposed DBFformer accurately predicts the output power of wind turbines and exhibits superior performance. It achieves lower mean squared error (MSE) and mean absolute error (MAE) values than other state-of-the-art models. Specifically, its MSE values are 0.195, 0.216, 0.457, and 0.583, and the corresponding MAE values are 0.318, 0.335, 0.474, and 0.503 for different rated wind turbines. Furthermore, comprehensive ablation experiments validate that the dual-branch structure, frequency transformations, dual-attention mechanisms, and PWC module have a positive impact on the proposed model. Therefore, this research offers a novel and effective approach for wind power forecasting and supports the broader goal of integrating clean energy into microgrids.