Multi-scale wind speed prediction model based on improved escape algorithm for optimizing time-varying filtering...

Area of Science:

• Renewable Energy Systems
• Computational Intelligence
• Signal Processing

Background:

• Wind energy's inherent stochasticity and volatility pose significant challenges to grid stability and reliable power dispatch.
• Existing decomposition-based forecasting methods struggle with mode mixing and non-stationary signal separation, limiting accuracy.

Purpose of the Study:

• To develop an advanced forecasting method for wind energy that enhances grid stability and dispatch reliability.
• To improve the accuracy and robustness of wind power forecasting, particularly for multi-step predictions.

Main Methods:

• Proposed an Improved Escape Algorithm (IESC) with chaotic mapping to optimize Time-Varying Filter Empirical Mode Decomposition (TVF-EMD) hyperparameters, mitigating mode mixing.
• Implemented a frequency-adaptive hybrid model utilizing XLSTM for high-frequency, LSTM for medium-frequency, and ELM for low-frequency wind signal components.
• Evaluated the IESC-optimized TVF-EMD and the hybrid forecasting model on a large-scale wind energy dataset.

Main Results:

• The IESC algorithm demonstrated superior performance over standard ESC, GWO, and DE optimization methods.
• The hybrid model achieved a 29.8% lower 1-step Mean Absolute Error (MAE) and a 65.6% higher 15-step R² compared to XLSTM alone.
• Error growth over 15 steps was limited to 12%, with R² degradation 35% slower, indicating significant improvements in long-term accuracy and stability.

Conclusions:

• The proposed IESC-optimized TVF-EMD and hybrid forecasting model significantly enhance wind power forecasting precision.
• The method effectively bridges multi-step forecasting accuracy with real-time dispatch requirements, ensuring better grid-demand matching.
• This approach contributes to improved operational stability in power grids integrating variable wind energy sources.