Research on Wind Power Short-Term Forecasting Method Based on Temporal Convolutional Neural Network and Variational...

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Time Series Forecasting

Background:

• Wind energy is a vast resource but faces challenges due to its inherent randomness and volatility.
• Accurate wind power prediction is crucial for efficient grid integration and maximizing wind energy utilization.

Purpose of the Study:

• To develop an accurate short-term wind power prediction model.
• To improve wind energy forecasting by incorporating environmental factors and historical power data.

Main Methods:

• Variational Mode Decomposition (VMD) was employed to denoise and decompose non-smooth environmental data.
• Pearson correlation coefficient and Maximal Information Coefficient (MIC) were used to select relevant modal components.
• A Time Convolutional Neural Network (TCN) model was trained using selected components and historical power data.

Main Results:

• The proposed VMD-TCN model achieved high prediction accuracy on a public dataset.
• The model demonstrated a Mean Absolute Percentage Error (MAPE) of 2.79% and an R-squared (R²) of 0.9985.
• The VMD-TCN model showed superior accuracy and robustness compared to the Long Short-Term Neural Network (LSTM) model.

Conclusions:

• The VMD-TCN model effectively addresses the challenges of wind power prediction posed by data volatility.
• The methodology provides a robust framework for accurate short-term wind power forecasting.
• This approach enhances the reliability of wind energy integration into power grids.