Optimizing solar power forecasting with metaheuristic algorithms and deep learning models for photovoltaic grid...

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Power Grids
• Deep Learning for Forecasting

Background:

• Global acceleration of photovoltaic system integration into power grids necessitates precise solar power forecasting.
• Optimizing energy dispatch, grid reliability, and large-scale renewable energy production depend on accurate forecasting.
• Existing forecasting methods may lack the robustness required for complex grid integration.

Purpose of the Study:

• To propose and evaluate a novel hybrid deep learning model for enhanced solar power forecasting.
• To leverage the strengths of Gated Recurrent Units (GRU) and Long Short-Term Memory (LSTM) networks through sequential combination.
• To optimize model hyperparameters using the nature-inspired Fire Hawk optimization (FHO) algorithm for superior performance.

Main Methods:

• Development of a FHO-GRU-LSTM model combining GRU and LSTM networks with FHO-optimized hyperparameters.
• Training the model using time-based temporal indexing and a recursive forecasting strategy.
• Evaluation on two distinct photovoltaic technologies (Poly-crystalline Array 1 and Mono-crystalline Array 2) within the PEARL system.

Main Results:

• The FHO-GRU-LSTM model demonstrated superior accuracy and robustness in solar power forecasting.
• Achieved high R2 scores of 0.9964 (Array 1) and 0.9966 (Array 2).
• Significant reductions in Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) by 12.67-24.52% across both arrays.

Conclusions:

• Advanced hyperparameter tuning, exemplified by FHO, is critical for enhancing deep learning model generalization.
• The FHO-GRU-LSTM model offers a robust solution for improving grid stability and sustainable renewable energy integration.
• The study underscores the potential of hybrid deep learning approaches in addressing challenges in modern power grid management.