Wadhwani et al. (2026) Attention-Enhanced Recurrent Neural Networks for Wind Speed Downscaling from Global Climate Models: Case Study of Pune City

Identification

• Journal: Lecture notes in networks and systems
• Year: 2026
• Date: 2026-01-01
• Authors: Vishwas Wadhwani, Rajesh Wadhvani, Pragati Agrawal
• DOI: 10.1007/978-3-032-10664-3_18

Research Groups

• Center of Artificial Intelligence, Maulana Azad National Institute of Technology, Bhopal, India
• Department of Computer Science and Engineering, Maulana Azad National Institute of Technology, Bhopal, India

Short Summary

This research introduces attention-enhanced recurrent neural networks (RNNs) to downscale wind speed from Global Climate Models (GCMs) for local-scale renewable energy assessments in Pune city, demonstrating significantly improved prediction accuracy and interpretability compared to traditional RNN models.

Objective

• To enhance the modeling of temporal dependencies for wind speed downscaling from Global Climate Models (GCMs) to provide detailed local-scale data for renewable energy assessments.

Study Configuration

• Spatial Scale: From broad spatial resolution (Global Climate Models) to detailed local-scale (Pune city, India).
• Temporal Scale: Focuses on modeling temporal dependencies for wind speed prediction.

Methodology and Data

• Models used: Attention mechanisms integrated into Recurrent Neural Networks (RNNs), specifically Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models.
• Data sources: Outputs from Global Climate Models (GCMs).

Main Results

• The proposed attention-augmented LSTM and GRU models significantly improve wind speed prediction accuracy.
• The models provide enhanced interpretability through visual attention weights, aiding in the identification of crucial historical data.
• These attention-enhanced models outperform traditional RNN models for precise wind resource forecasting.

Contributions

• Introduction of an innovative approach integrating attention mechanisms into RNNs for wind speed downscaling, offering improved accuracy and interpretability.
• Demonstration of the suitability of attention mechanisms for precise wind resource forecasting at local scales, addressing the limitations of broad GCM spatial resolutions.

Funding

• Not explicitly stated in the provided text, but computational resources and facilities were supplied by Maulana Azad National Institute of Technology, Bhopal.

Citation

@article{Wadhwani2026AttentionEnhanced,
  author = {Wadhwani, Vishwas and Wadhvani, Rajesh and Agrawal, Pragati},
  title = {Attention-Enhanced Recurrent Neural Networks for Wind Speed Downscaling from Global Climate Models: Case Study of Pune City},
  journal = {Lecture notes in networks and systems},
  year = {2026},
  doi = {10.1007/978-3-032-10664-3_18},
  url = {https://doi.org/10.1007/978-3-032-10664-3_18}
}