Comparative analysis of daily global solar radiation prediction using deep learning models inputted with stochastic...

Area of Science:

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

Background:

• Photovoltaic power generation is directly influenced by daily global solar radiation (DGSR).
• DGSR data precision is often limited due to high instrument costs and data intermittency from equipment failures.
• Accurate DGSR prediction is essential for reliable photovoltaic power production.

Purpose of the Study:

• To investigate and compare the performance of various artificial neural network (ANN) models for DGSR prediction.
• To evaluate the impact of different meteorological variables on DGSR prediction accuracy.
• To identify the most effective ANN architecture and input variable combination for precise DGSR forecasting.

Main Methods:

• Development and comparison of four ANN models: Radial Basis Function Neural Network (RBFNN), Long Short-Term Memory Neural Network (LSTMNN), Modular Neural Network (MNN), and Transformer Model (TM).
• Utilizing five meteorological stochastic variables as inputs: wind speed, relative humidity, minimum temperature, maximum temperature, and average temperature.
• Evaluating model performance based on prediction accuracy metrics, specifically Mean Absolute Relative Error (MARE).

Main Results:

• The Transformer Model (TM) achieved a Mean Absolute Relative Error of 1.98% when using average, maximum, and minimum temperatures as input variables.
• All investigated ANN models demonstrated superior predictive accuracy compared to traditional forecasting methods.
• Different combinations of meteorological variables yielded varying levels of accuracy across the ANN models.

Conclusions:

• ANN models, particularly the Transformer Model, offer a significant advancement in DGSR prediction accuracy.
• The selection of input meteorological variables critically impacts the performance of DGSR forecasting models.
• This research highlights the potential of advanced AI techniques for optimizing photovoltaic energy production through improved solar radiation forecasting.