Data-augmented sequential deep learning for wind power forecasting

Accurate wind power forecasting plays a critical role in the operation of parks and dispatch energy into grid. With excellent automatic pattern recognition nonlinear mapping ability for big data, deep learning is increasingly employed forecasting. However, salient realities are that in-situ measured data relatively expensive inaccessible correlation between steps omitted most multistep forecasts. This paper first time augmentation applied to by systematically summarizing proposing both physics-oriented data-oriented time-series approaches considerably enlarge primary datasets, develops encoder-decoder long short-term memory networks enable sequential input output The proposed techniques algorithm deployed on five turbines with diverse topographies an Arctic park, outcomes evaluated against benchmark models different augmentations. main findings reveal one side, average improvement RMSE model over benchmarks 33.89%, 10.60%, 7.12%, 4.27% before augmentations, increases 40.63%, 17.67%, 11.74%, 7.06%, respectively, after other side unveils effect augmentations prediction intricately varying, but without all boost outperformance from 7.87% 13.36% RMSE, 5.24% 8.97% MAE, similarly 12% QR90. Finally, general, slightly better than physics-driven ones.