Fast-powerformer achieves accurate and memory-efficient mid-term wind power forecasting.

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Power Grid Management

Background:

• Mid-term wind power forecasting (WPF) is vital for grid stability and economics.
• Existing Transformer models face a trade-off between predictive accuracy and computational efficiency.
• Current methods struggle with redundant computations, weak inter-variable coupling, or loss of local temporal dynamics.

Purpose of the Study:

• To develop an efficient and accurate mid-term wind power forecasting model.
• To address the limitations of existing Transformer-based architectures in WPF.
• To balance high predictive accuracy with reduced computational resource consumption.

Main Methods:

• Proposed Fast-Powerformer model based on the Reformer architecture.
• Implemented an Input Transposition Mechanism for optimized multivariate coupling and reduced complexity.
• Introduced a lightweight temporal embedding module to capture local sequential features.
• Integrated a Frequency Enhanced Channel Attention Mechanism (FECAM) for spectral pattern analysis.

Main Results:

• Fast-Powerformer demonstrated superior overall performance compared to existing methods.
• The model achieved a significant balance between high accuracy and reduced resource consumption.
• Experimental validation on real-world wind farm datasets confirmed the model's effectiveness.

Conclusions:

• Fast-Powerformer offers a practical solution for mid-term wind power forecasting.
• The model shows significant potential for deployment in resource-constrained environments.
• The proposed strategies effectively overcome the accuracy-efficiency trade-off in WPF.