A wind power noise prediction model training and wind power prediction method
By overlaying wind power and meteorological conditions time-frequency data into a multi-channel graph and adding noise, and utilizing a converter structure with a block embedding layer and an attention layer, the problem of insufficient frequency domain feature fusion in the diffusion model is solved, and more accurate wind power prediction is achieved.
Patent Information
- Application Number
- CN202610753313.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-05-28
- Publication Date
- 2026-07-17
AI Technical Summary
Existing wind power prediction methods based on diffusion models fail to effectively integrate the frequency domain characteristics of the sequence, making it difficult to fully capture the inherent periodic fluctuations and continuous evolution of wind power, thus affecting prediction accuracy.
By overlaying power time-frequency data and condition variable time-frequency data into a multi-channel time-frequency graph, and adding noise only to the power channel, the time-frequency unit is divided into power and condition variable units using a block embedding layer and an attention layer. Combined with the attention mechanism of the converter structure, a global dependency relationship is established, which solves the problems of frequency domain feature fusion and long-term memory weakening.
It improves the accuracy of wind power prediction, reduces long-term memory weakening and multi-step error accumulation, and enhances the robustness and prediction efficiency of the model.
Smart Images
Figure CN122413079A_ABST