Hiformer: Hybrid Frequency Feature Enhancement Inverted Transformer for Long-Term Wind Power Prediction

TL;DR

Hiformer is a novel transformer-based model that enhances long-term wind power prediction accuracy and efficiency by integrating signal decomposition and weather feature extraction techniques.

Contribution

It introduces a hybrid architecture combining signal decomposition with weather features and an encoder-only design for improved long-term wind power forecasting.

Findings

Improves prediction accuracy by up to 52.5%.

Reduces computational time by up to 68.5%.

Effectively models meteorological and wind power correlations.

Abstract

The increasing severity of climate change necessitates an urgent transition to renewable energy sources, making the large-scale adoption of wind energy crucial for mitigating environmental impact. However, the inherent uncertainty of wind power poses challenges for grid stability, underscoring the need for accurate wind energy prediction models to enable effective power system planning and operation. While many existing studies on wind power prediction focus on short-term forecasting, they often overlook the importance of long-term predictions. Long-term wind power forecasting is essential for effective power grid dispatch and market transactions, as it requires careful consideration of weather features such as wind speed and direction, which directly influence power output. Consequently, methods designed for short-term predictions may lead to inaccurate results and high computational costs in long-term settings. To adress these limitations, we propose a novel approach called Hybrid Frequency Feature Enhancement Inverted Transformer (Hiformer). Hiformer introduces a unique structure that integrates signal decomposition technology with weather feature extraction technique to enhance the modeling of correlations between meteorological conditions and wind power generation. Additionally, Hiformer employs an encoder-only architecture, which reduces the computational complexity associated with long-term wind power forecasting. Compared to the state-of-the-art methods, Hiformer: (i) can improve the prediction accuracy by up to 52.5\%; and (ii) can reduce computational time by up to 68.5\%.