Short-Term Sun Coverage Prediction Using Cloud Movement Pattern Feature in PV System

This study addresses the challenge of solar intermittency in Photovoltaic (PV) generation systems by proactively forecasting the cloud coverage effecting the irradiance from the sun. Using a hybrid color-intensity sun detection method, a specialized contour analysis technique is introduced to precisely identify the sun within sky images. Subsequently, a distinctive feature extraction process is proposed specifically for predicting cloud coverage over the sun. This process generates feature images representing cloud movement patterns and serving as inputs for the predictive model, a Convolutional Neural Network (CNN) based on the GoogLeNet architecture. The CNN model predicts ‘sun cover’ or ‘sun not cover’ events, facilitating forecast horizons of 1-minute and 2-minute ahead predictions. The proposed method can achieve notable accuracies of 93.15% for 1-minute and 90% for 2-minute ahead predictions. This study demonstrates robust predictive capabilities in dealing with solar intermittency effects in PV systems.