Machine Learning Applications for Renewable Energy Forecasting in Nigeria’s Electricity System

Accurate short-term forecasting of solar photovoltaic and wind energy generation is a critical operational requirement for electricity system operators as Nigeria's renewable energy share grows, yet the application of advanced machine learning methods to Nigerian renewable energy forecasting, accounting for local climate patterns including harmattan dust events, tropical convective cloud systems, and monsoon seasonality, represents a significant knowledge gap. This study develops and validates machine learning models for solar photovoltaic and wind generation forecasting tailored to Nigeria's electricity system operational requirements and local climatic conditions. Ground-based irradiance, wind speed, and atmospheric aerosol optical depth data from five monitoring stations in Lagos, Abuja, Kano, Enugu, and Port Harcourt are used to train and test forecasting models. Four machine learning architectures, including Long Short-Term Memory networks, Temporal Convolutional Networks, Gradient Boosting Machines, and Transformer-based models, are implemented and their forecasting performance compared across intra-day, day-ahead, and week-ahead horizons using skill score metrics relative to persistence and climatology baselines. Novel contributions include development of a harmattan aerosol attenuation module that improves solar forecasting accuracy by 22 percent during the November to February dust season, and a spatially interpolated wind ramp event detection algorithm for the Middle Belt transition zone. Results confirm that ensemble machine learning approaches outperform individual models by 15 to 28 percent on skill score metrics. The study provides operational forecasting tools ready for deployment by the Nigerian Electricity System Operator. Keywords: machine learning, renewable energy forecasting, Nigeria, solar photovoltaic, harmattan aerosol.