TITLE:
Optimized Neural Network-Based Solar Energy Forecasting Using Genetic Algorithm
AUTHORS:
Intan Azmira Wan Abdul Razak, Muhammad Zulhafiz Mohd Nizam, Mohamad Fani Sulaima, Farah Anishah Zaini
KEYWORDS:
Solar Energy Forecasting, Neural Networks, Genetic Algorithm, Hyperparameter Optimization, Photovoltaic Power Generation, Grid Stability
JOURNAL NAME:
Journal of Power and Energy Engineering,
Vol.14 No.11,
November
20,
2026
ABSTRACT: Accurate solar energy forecasting is vital for grid stability, optimal power dispatch, and effective integration of renewable resources, particularly in tropical climates characterized by extreme weather intermittency. While Artificial Neural Networks (ANNs) excel at modeling complex non-linear relationships, standard backpropagation algorithms are prone to getting trapped in local minima and are highly sensitive to initial hyperparameter selection. This study develops a hybrid forecasting framework combining a Feedforward Neural Network (NN) with a Genetic Algorithm (GA) to optimize hyperparameter configurations, including hidden neurons, learning rate, and momentum constant. Using empirical operational and meteorological data from a utility-scale solar photovoltaic plant located in a tropical climate zone, models were built using multi-day lagged solar irradiance and ambient temperature variables. On the evaluated September 2022 test period, the proposed NN-GA hybrid model achieved lower errors than the conventional neural network configurations. For solar irradiance-based models, GA optimization increased the coefficient of determination (
R
2
) from 0.589 to 0.798 while reducing the Mean Absolute Percentage Error (MAPE) from 18.70% to 13.53% and Mean Absolute Error (MAE) from 33.89 kW to 27.11 kW. For ambient temperature-based models, the hybrid architecture sustained an
R
2
of 0.803, reduced MAE from 32.97 kW to 27.18 kW, dropped MAPE to 13.05%, and produced no negative power predictions in the evaluated test samples. These findings indicate that evolutionary hyperparameter optimization improves the selected network configuration for utility-scale solar forecasting in tropical environments.