Machine Learning-Based Solar Photovoltaic Power Forecasting for Nigerian Regions

Abstract

This study explores machine learning-based forecasting of solar photovoltaic (PV) power generation across distinct climatic regions in Nigeria. Machine learning techniques, particularly support vector machines (SVM) and artificial neural networks (ANN), were employed to predict solar PV output, utilizing a comprehensive data set spanning 12 years of climatic parameters, including solar irradiation, cloud cover, temperature, and humidity. Model training, validation, and testing were conducted in MATLAB using the ANN approach, with results indicating a notable improvement in prediction accuracy with the addition of hidden layers. The model achieved optimal performance with 1000 hidden layers, achieving a low mean squared error (MSE) and high correlation coefficient (R) values across all regions. Forecasted power generation values revealed region-specific insights, with the Northern region exhibiting the highest solar potential, attributable to its hot, dry climate and minimal cloud cover. Conversely, regions with high humidity and frequent cloud cover, such as the Southern region, showed reduced PV output. These findings highlight the critical role of machine learning in enhancing solar PV forecasting accuracy across diverse environments. The study's insights provide a foundation for policymakers and stakeholders to make informed decisions, promote sustainable energy initiatives, and optimize solar energy resource management in Nigeria.