Solar Irradiance Prediction for Zaria Town Using Different Machine Learning Models

Abstract

The research is set to predict solar irradiation using various machine learning algorithms. This is done in order to construct and develop a high-efficiency prediction model that uses actual meteorological data to predict daily solar irradiance for the town of Zaria, Nigeria. To assist utilities working in various solar energy generation and monitoring stations in making effective solar energy generation management system decisions. Four machine learning models (artificial neural network (ANN), decision tree (DT), random forest (RF), and gradient boost tree (GBT).) were used to predict and compare actual and anticipated solar radiation values. The results reveal that meteorological characteristics (min-humidity, max-temperature, day, month, and wind direction) are critical in machine learning model training. The solar radiation prediction skills of multi-layer perceptron and decision tree models were low. In the prediction of daily solar irradiation, the ensemble learning models of random forest and gradient boost tree outperformed the other models. The random forest model is shown to be the most accurate in predicting solar irradiation.