A Comparative Study of Machine Learning Models for Solar Power Generation Forecasting Using Weather Parameters: A Case of Benin City

Abstract

Solar power generation as a renewable energy source is one of the
most used and highest in demand in recent times, solar systems totally
depend on weather changes. Because of weather fluctuations, one
cannot manually determine the amount of energy produce by a solar
system due to the amount of data and computational power needed.
Machine learning having this computational advantage and being
able to perform almost accurate predictions was implemented in this
study. A comparison of three different machine learning models
(decision tree regression, support vector regression and random
forest) used to predict solar power generation were carried out using
dataset generated over a given period of time across different
locations in Benin City. An IoT -based data logger was developed to
generate the dataset using temperature, humidity and light-insensitive
sensors interfaced with an Atmega 238p Microcontroller. The data
acquired by sensors are sent to the cloud through a GSM module
connected to the Microcontroller for storage. This dataset constitutes
of weather parameters (temperature, humidity and light flux) and
electrical parameters (voltage and current). The preferred model was
selected using performance metrics and minimal error value. The
results of the comparison shown that Random Forest regression model
had a model score of 0.9506, decision tree regression and support
vector regression with model score of 1.0594 and 3.7632 respectively.
Random forest regression model gives the most accurate predictions
of solar power over a given period. The model score was calculated
by addition of the values gotten from root mean square error, mean
absolute error, mean square error and R-squared value