Hybrid attention-based temporal convolutional bidirectional LSTM approach for wind speed interval prediction.

Area of Science:

• Renewable Energy Systems
• Data Science and Machine Learning
• Atmospheric Science

Background:

• Accurate wind speed prediction is vital for efficient wind power generation management.
• The inherent stochasticity of wind speed complicates precise interval prediction.
• Existing forecasting models often struggle with performance degradation over longer prediction horizons.

Purpose of the Study:

• To propose a robust hybrid approach for wind speed interval prediction (WSIP).
• To enhance the accuracy and reliability of wind speed forecasts for wind power applications.
• To evaluate the proposed model's performance across different prediction lead times.

Main Methods:

• A hybrid model integrating Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN), Temporal Convolutional Network with Attention Mechanism (ATCN), and Bidirectional Long Short-Term Memory (Bi-LSTM) network.
• ICEEMDAN for signal decomposition and noise reduction.
• ATCN for feature extraction and uncertainty reduction.
• Bi-LSTM for high-quality interval forecasting.

Main Results:

• The proposed ICEEMDAN-ATCN-Bi-LSTM framework demonstrated superior performance in wind speed interval prediction.
• Significant percentage improvements were observed: 36% for 5-min ahead, 47% for 10-min ahead, and 17% for 30-min ahead WSIP.
• The model maintained high accuracy across varying prediction intervals, unlike conventional methods.

Conclusions:

• The hybrid ICEEMDAN-ATCN-Bi-LSTM approach offers a powerful solution for complex wind speed interval prediction.
• This method effectively addresses the challenges posed by wind speed variability, improving wind power management.
• The study validates the model's efficacy using real-world wind farm data, highlighting its practical applicability.