Predictive Model for Solar Insolation Using the Deep Learning Technique

Abstract

In this study, prediction performances of a regression model and deep learning-based predictive models were comparatively analyzed for the prediction of hourly insolation in regions located at the temperate climate and microthermal climate with high precipitation. Unlike linear regression models, artificial neural networks (ANN) and long short-term memory- (LSTM-) based models achieved reliable predictive performances with CV(RMSE) of 14.0% and 15.8%, respectively. This study proposed the direction of future research by improving the performance of predicting insolation at 1 hour after the current time-step, which has time-dependent characteristics, by utilizing insolation at 24 hours before the current time-step and insolation at the current time-step in addition to the forecasted weather data. In the proposed models, a large error occurred at sunrise and sunset times, suggesting the possibility of improving predictive performance by utilizing variables related to sunrise and sunset in the future. Along with Cheongju, the proposed model could properly predict the hourly insolation in other regions around the world. The results of predicting other regions derived slightly higher prediction errors than Cheongju. However, it is expected that it will be possible to predict the hourly insolation in other regions with better prediction performance if variables related to geographical location are additionally considered in the future.

Zero-Shot Photovoltaic Power Forecasting Scheme Based on a Deep Learning Model and Correlation Coefficient

“…Correlation coefficients can be used to determine the relationship between an input and output variable. The most commonly used correlation coefficient is PCC[29]. It measures the strength and direction of the linear association between two variables, with values ranging from -1 to 1.…”