Cross-Dataset Data Augmentation Using UMAP for Deep Learning-Based Wind Speed Prediction

Computers, Год журнала: 2025, Номер 14(4), С. 123 - 123

Опубликована: Март 27, 2025

Wind energy has emerged as a cornerstone in global efforts to transition renewable energy, driven by its low environmental impact and significant generation potential. However, the inherent intermittency of wind, influenced complex dynamic atmospheric patterns, poses challenges for accurate wind speed prediction. Existing approaches, including statistical methods, machine learning, deep often struggle with limitations such non-linearity, non-stationarity, computational demands, requirement extensive, high-quality datasets. In response these challenges, we propose novel neighborhood preserving cross-dataset data augmentation framework high-horizon The proposed method addresses variability behaviors through three key components: (i) uniform manifold approximation projection (UMAP) is employed non-linear dimensionality reduction technique encode local relationships time-series while structures, (ii) localized (DA) approach introduced using UMAP-reduced spaces enhance diversity mitigate across datasets, (iii) recurrent neural networks (RNNs) are trained on augmented datasets model temporal dependencies patterns effectively. Our was evaluated from diverse geographical locations, Argonne Weather Observatory (USA), Chengdu Airport (China), Beijing Capital International (China). Comparative tests regression-based measures RNN, GRU, LSTM architectures showed that better at improving accuracy generalizability predictions, leading an average prediction error. Consequently, our study highlights potential integrating advanced reduction, augmentation, learning techniques address critical forecasting.

Язык: Английский

A physics-informed temporal convolutional network-temporal fusion transformer hybrid model for probabilistic wind speed predictions with quantile regression

Energy, Год журнала: 2025, Номер unknown, С. 136302 - 136302

Опубликована: Апрель 1, 2025

Язык: Английский