Search

As a renewable and clean energy source, wind energy has caught worldwide attention. To ensure the reliability and stability of wind energy production, wind speed forecasting is of great importance. Although existing forecasting models have shown promising results, they remain relatively small in scale and are incapable of learning comprehensive sequential representations from large-scale data to reason intricate spatio-temporal patterns obscured in entangled wind series. Recently, pre-trained large models have demonstrated impressive performance in natural language processing (NLP) and computer vision (CV). However, large models are rarely applied to wind speed forecasting due to the lack of large-scale wind data for training. To resolve this issue, we propose Spatio-Temporal Enhanced Pre-Trained Large Language Model, namely STELLM , innovatively leveraging the powerful reasoning ability of Large Language Models (LLMs) for accurate wind speed forecasting. Specifically, we first perform the series decomposition to separate wind series data into seasonal and trend components, capturing the distinct properties from disentangled patterns. Then, to empower LLMs with the spatio-temporal modeling capability, we introduce the spatial and temporal prompts, which fuse the geographic distribution information and short-term temporal patterns of wind turbines to enrich the input wind series. Moreover, we leverage an autoregressive fine-tuning strategy to align LLMs with wind series data and learn patch-level representations. Afterwards these representations are fed into a localized spatial module to capture the spatial dependencies between wind turbines. Extensive experiments on four public wind datasets demonstrate the performance superiority of STELLM. • We innovatively introduce Pre-Trained LLMs into the wind speed forecasting task. • We devise spatio-temporal prompts to boost LLM's performance. • We develop a localized spatial module to capture spatial dependencies. [ABSTRACT FROM AUTHOR]

Wind energy has attracted more and more attention due to its sustainability and pollution-free nature. As wind energy is highly dependent on wind speed, wind speed forecasting is of great importance for fully harvesting wind energy and ensuring the stability and reliability of wind energy systems. By accurately predicting the future wind speed, we can make efficient power schedules and consequently improve the utilization of wind power. Although existing forecasting methods have achieved satisfactory wind forecasting performance, they may still fail to discover the intricate spatio-temporal dependencies obscured in entangled patterns. To resolve this issue and further improve the prediction accuracy, we propose Mixformer , a novel mixture Transformer with hierarchical context for spatio-temporal wind speed forecasting. Mixformer first applies seasonal-trend decomposition, and then employs an MLP to predict the trend part and an attention model to predict the seasonal part. Besides, Mixformer innovatively proposes a spatio-temporal Gaussian mixture attention (ST-GMA) layer, which fuses the periodic temporal and long-term spatial context. To capture intricate long-term spatial characteristics, Mixformer extracts global information by the Dynamic Time Warping algorithm. Moreover, with the ST-GMA layer, Mixformer builds a hierarchical encoder–decoder architecture to fully utilize the context at different scales for seasonal forecasting and further enhance the spatio-temporal modeling. Empirical experiments on four real-world benchmark datasets show that Mixformer yields the lowest prediction errors with the MAE scores of 1.66, 1.68, 3.14 and 0.70 and the RMSE scores of 2.07, 2.01, 3.99 and 0.98, and improves the prediction accuracy by 8.43%, 5.36%, 4.14%, 6.57% in term of MAE scores and 9.18%, 6.92%, 4.01%, 5.75% in term of RMSE scores compared to state-of-the-art methods, which demonstrates the superiority of Mixformer. In a nutshell, Mixformer presents a promising method for spatio-temporal wind speed forecasting and may achieve desirable performance in real-world applications of wind power systems. [Display omitted] • Mixformer is proposed for spatio-temporal wind speed forecasting. • Mixformer leverages the series decomposition to extract disentangled patterns. • Mixformer devises a spatio-temporal Gaussian mixture attention module. • Experiments on 4 real-world datasets validate the superiority of Mixformer. [ABSTRACT FROM AUTHOR]