Photovoltaic Power Output Estimation with Digital Visible Cameras

Monitoring and estimating the production of a photovoltaic (PV) system is crucial to optimise energy efficiency, detect performance problems and improve its financial viability. Monitoring allows the operator to detect faults early, improve maintenance planning and maximise energy production, thus improving return on investment and sustainability. On the other hand, estimating PV field production helps to plan energy consumption and integrate it into the grid. This study aims to explore a deep learning approach to estimate power generation in real time using images captured every ten minutes by three fixed cameras positioned at different perspectives (left, lateral and right) of the PV field at CIESOL research center. A hybrid neural network combining convolutional neural networks (CNN) and long-term memory networks (LSTM) is used to analyse the images and establish correlations with the actual power production. The results show that an accuracy of up to 8% (NRMSE) can be achieved in production estimation, as well as adequately capturing power variations and environmental influences. These findings underscore the potential of image-based monitoring as a non-intrusive and efficient method for improving PV system management and maintenance.