Application of deep learning for solar irradiance and solar photovoltaic multi-parameter forecast
Olusola Bamisile, Ariyo Oluwasanmi, Sandra Obiora, Emmanuel Osei-Mensah, Gaylord Ogbonna Asoronye, Qi Huang
Abstract
In this study, the use of artificial neural network (ANN) models for solar irradiance and solar PV parameters forecasting in Nigeria is presented. Although solar irradiance prediction exists in literatures, the use of ANN for solar PV parameter prediction has not been considered in previous studies. Six different locations are selected in Nigeria and data from these locations have been used to train/test the ANN models developed. This study aims to model ANN algorithms that can forecast solar irradiance and solar PV parameters based on an hourly time-step more accurately. A deep learning regression model built on Levenberg-Marquardt back propagation algorithm is used to train and test the model for all the locations considered. Four different ANN models were developed for each location in Keras python using all the input parameters. The evaluation metrics used in this study are; R, R-squared, RMSE, and MAE. The models developed are capable of predicting solar irradiance and solar PV parameters. The R values for the ANN models range from 0.9046–0.9777 for solar irradiance and 0.7768–0.8739 for solar PV multi-parameters prediction.