Optimizing solar irradiance forecasting: ANN models enhanced with ADAM and Cuckoo search algorithm.

Area of Science:

• Energy Systems Engineering
• Artificial Intelligence
• Meteorological Forecasting

Background:

• Increasing demand for energy and environmental concerns necessitate the integration of renewable energy sources (RES) into the electrical grid.
• The intermittent nature of RES requires reliable forecasting for effective grid management.
• Artificial neural network (ANN) accuracy in solar irradiance prediction is limited by random weight initialization.

Purpose of the Study:

• To enhance artificial neural network (ANN) based solar irradiance prediction accuracy.
• To propose and evaluate hybrid forecasting models optimizing ANN weights using optimization algorithms.
• To compare the performance of ADAM-optimized ANN and a novel CSA-ADAM optimized ANN model.

Main Methods:

• Developed two hybrid solar irradiance forecasting models: ADAM-optimized ANN and CSA-ADAM optimized ANN.
• Employed the cuckoo search algorithm (CSA) and adaptive moment estimation (ADAM) to optimize ANN edge weights.
• Validated models using actual weather data and standard error metrics (MSE, MAPE, MAE, RMSE).

Main Results:

• The CSA-ADAM optimized ANN model achieved superior accuracy with MSE=0.25, MAPE=0.17%, MAE=0.43, and RMSE=0.50.
• The ADAM-optimized ANN model yielded MSE=0.52, MAPE=0.18%, MAE=0.64, and RMSE=0.72.
• Prediction times were comparable, with CSA-ADAM optimized ANN at 0.1110 ± 0.0058 seconds and ADAM-optimized ANN at 0.1093 ± 0.0085 seconds.

Conclusions:

• Optimizing ANN weights with CSA significantly improves solar irradiance prediction accuracy.
• The CSA-ADAM hybrid model offers a more accurate and reliable approach to solar forecasting.
• Hybrid optimization techniques are effective in overcoming limitations of standard ANN models for RES forecasting.