Estimation of Hourly Solar Radiations on Horizontal Surface from Daily Average Solar Radiations Using Artificial Neural Network

Abstract

Hourly solar radiation information is necessary to increase the effectiveness of photovoltaic (PV) energy systems. However, compared with daily average solar radiation, the measurement of hourly solar radiation is much less available. In this study, a multilayer perceptron artificial neural network based on daily average solar radiation is proposed to estimate the amount of hourly solar radiation. The proposed method also relies on the location of the PV system, the hour when the estimate is needed, and the month when the estimate is needed. Two separate networks were built to estimate hourly direct solar radiation and hourly diffuse solar radiation, the summation of which is the estimated value of global hourly solar radiation. The networks were trained using three years of data from 2016 to 2018 from five locations around Japan. The data were taken from the website of the Japan Meteorological Agency (JMA). The number of hidden neurons for each network was determined by comparing the regression value obtained during the training process. The proposed method was validated by comparing the estimated value with the actual measured solar radiation value for each month in 2019 for the same five locations used for training the networks. The reliability of the proposed method was confirmed with the minimum R2 of 0.951 for the estimation of hourly direct solar radiation and 0.983 for hourly diffuse solar radiation. Further improvement in the accuracy of estimating the daily peak radiation value is considered to improve the total accuracy of the estimation model in the future.