DOD-Boost: a temporal and distribution-optimized deep boosting framework for solar radiation modeling

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Statistical Modeling

Background:

• Accurate solar radiation forecasting is crucial for designing efficient clean energy systems.
• Traditional methods often struggle with the inherent randomness and temporal variability of solar data.
• Deep learning and statistical distribution fitting offer promising avenues for improved solar radiation modeling.

Purpose of the Study:

• To develop and evaluate hybrid solar radiation temporal modeling approaches for clean energy system design.
• To integrate statistical distribution fitting with deep learning techniques for enhanced prediction accuracy.
• To establish a transferable framework for photovoltaic-based energy planning, particularly in developing countries.

Main Methods:

• Analysis of solar radiation data using probability distributions and parameter optimization via Maximum Likelihood Estimation (MLE), Whale Optimization Algorithm (WOA), and Particle Swarm Optimization (PSO).
• Development of hybrid temporal modeling using Cumulative Distribution Function (CDF) with Long Short-Term Memory networks (LSTMs), Gated Recurrent Units (GRUs), and Extreme Gradient Boosting (XGBoost).
• Evaluation of models using Jensen-Shannon Divergence (JSD) to assess distributional accuracy.

Main Results:

• The proposed DOD-Boost framework, integrating data preprocessing, optimization, and temporal modeling, achieved highly accurate solar radiation predictions.
• The Weibull (WOA) - LSTM - XGBoost hybrid model exhibited the best performance, yielding the lowest JSD value of 0.0084.
• Jensen-Shannon Divergence (JSD) proved effective in evaluating the similarity between predicted and actual solar radiation data distributions.

Conclusions:

• Hybrid models combining statistical distributions and deep learning significantly enhance solar radiation temporal modeling accuracy.
• The developed DOD-Boost framework offers a robust and transferable solution for photovoltaic energy planning.
• The findings support the deployment of advanced renewable energy systems, especially in regions with limited data infrastructure.