Innovative second-generation wavelets construction with recurrent neural networks for solar radiation forecasting

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Engineering
• Meteorological Forecasting

Background:

• Accurate solar radiation prediction is crucial for optimizing photovoltaic (PV) module power generation.
• Existing methods often struggle to capture complex correlations within meteorological data.

Purpose of the Study:

• To develop and evaluate a novel wavelet recurrent neural network (WRNN) for precise 2-day solar radiation forecasting.
• To exploit the relationship between solar radiation and timescale-dependent meteorological variables.

Main Methods:

• Utilized observed meteorological data (wind speed, humidity, temperature) from the University of Catania.
• Employed WRNNs that process timescale-related wavelet coefficients from meteorological time series.
• Implemented WRNNs to perform predictions directly in the wavelet domain, including inverse transform for output.

Main Results:

• Achieved a very low root-mean-square error in solar radiation prediction.
• Demonstrated superior performance compared to existing hybrid neural network approaches.
• Validated the effectiveness of the wavelet domain prediction strategy.

Conclusions:

• The proposed WRNN offers a highly accurate and novel approach to solar radiation prediction.
• This method effectively integrates meteorological data analysis with advanced neural network techniques.
• The WRNN shows significant potential for enhancing the reliability of PV power systems.