Short-term wind power forecasting through stacked and bi directional LSTM techniques

Area of Science:

• Computational intelligence
• Renewable energy systems
• Machine learning for time series forecasting

Background:

• Traditional recurrent neural networks (RNNs) struggle with long-term temporal dependencies, leading to vanishing gradient issues.
• This limitation hinders efficient resource utilization in wind prediction models.
• Computational intelligence (CI) offers potential solutions for enhancing prediction accuracy.

Purpose of the Study:

• To propose an advanced recurrent neural network (RNN) model using Long Short-Term Memory (LSTM) architecture.
• To address the vanishing gradient problem inherent in traditional RNNs for wind prediction.
• To improve the training performance and efficiency of wind prediction models.

Main Methods:

• Developed a recurrent neural network (RNN) model incorporating stack LSTM and bidirectional LSTM architectures.
• Evaluated model performance using standard metrics: Mean Absolute Error (MAE), Standard Deviation Error (SDE), and Root Mean Squared Error (RMSE).
• Compared the proposed model against state-of-the-art techniques using diverse wind farm datasets.

Main Results:

• The proposed LSTM-based RNN model demonstrated superior performance over existing techniques, achieving lower RMSE and MAE across all datasets.
• A notable reduction in SDE was observed for larger wind farm datasets.
• The model achieved comparable results with fewer parameters and minimal processing power requirements.

Conclusions:

• The advanced LSTM architecture effectively overcomes the vanishing gradient problem in RNNs for wind prediction.
• The proposed model offers a more efficient and effective approach to wind resource utilization.
• This method provides a robust and computationally efficient solution for enhancing wind prediction accuracy.