A novel PV power prediction method with TCN-Wpsformer model considering data repair and FCM cluster
Abstract Short-term day-ahead photovoltaic power prediction is of great significance for power system dispatch plan formulation. In this work, to improve the accuracy of photovoltaic power prediction, a TCN-Wpsformer (temporal convolutional network-window probability sparse Transformer) day-ahead ph...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-04-01
|
| Series: | Scientific Reports |
| Subjects: | |
| Online Access: | https://doi.org/10.1038/s41598-025-95843-9 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Short-term day-ahead photovoltaic power prediction is of great significance for power system dispatch plan formulation. In this work, to improve the accuracy of photovoltaic power prediction, a TCN-Wpsformer (temporal convolutional network-window probability sparse Transformer) day-ahead photovoltaic power prediction model based on combining data restoration and FCM (fuzzy C means) cluster is proposed. The time code of the dataset obtained after data restoration and FCM clustering was spliced with the location code. A temporal convolutional neural network is introduced to extract temporal segment features and incorporate a self-attention mechanism. The short-term photovoltaic power prediction is outputted by the window probability sparse Transformer model in multiple steps. Compared with the original Transformer model, the window probability sparse Transformer model uses the window probability sparse self-attention mechanism. It captures the long-term dependencies while filtering out the time segment features with relatively high importance for computation, which improves the prediction accuracy and reduces the computational cost. The computing time is reduced to 68.83% and R squared is improved by 5.3% compared to Transformer. The comparison is made through 11 models, and the R squared of this model is above 99% while different data volume and different power station data. It proves that the model stability and cross scene generalisation ability is well. Meanwhile, it can also provide more accurate confidence intervals on the basis of point prediction, which has certain application value. |
|---|---|
| ISSN: | 2045-2322 |