Stochastic Grey Wolf Optimization for Hyperparameter Tuning of LSTM and RNN Models in Energy Forecasting.

Area of Science:

• Renewable Energy Systems
• Artificial Intelligence in Energy
• Time Series Forecasting

Background:

• Accurate photovoltaic (PV) power forecasting is crucial for integrating solar energy into electric grids.
• Recurrent Neural Networks (RNNs) and Long Short-Term Memory (LSTM) networks are effective for time series forecasting but sensitive to hyper-parameter selection.

Purpose of the Study:

• To develop a Stochastic Grey Wolf Optimizer (SGWO) based system for optimizing RNN and LSTM hyper-parameters for PV power prediction.
• To enhance the exploration capabilities of the Grey Wolf Optimizer (GWO) by introducing stochastic elements to prevent premature convergence.

Main Methods:

• Implementation of a novel Stochastic Grey Wolf Optimizer (SGWO) by integrating randomness into the standard Grey Wolf Optimizer (GWO).
• Application of the SGWO algorithm for hyper-parameter optimization of RNN and LSTM models.
• Evaluation using a real-world PV dataset, comparing SGWO-LSTM against manual tuning, random search, and standard GWO.

Main Results:

• The SGWO-LSTM configuration demonstrated the highest out-of-sample prediction accuracy.
• Performance metrics included Mean Absolute Error (MAE) of 0.018, Root Mean Squared Error (RMSE) of 0.041, and a Coefficient of Determination (R²) of 0.978.
• The stochastic nature of SGWO improved exploration and prevented premature convergence compared to standard GWO.

Conclusions:

• The proposed SGWO-based hyper-parameter optimization system significantly enhances the accuracy of PV power forecasting using LSTM models.
• SGWO offers a robust approach to optimizing complex machine learning models for renewable energy applications.
• This method provides a pathway for more stable and efficient integration of solar power into existing power grids.