Modelagem e simulação de arranjos de painéis fotovoltaicos para predição de condições operacionais usando curvas I-V e redes neurais convolucionais

Abstract

The growth of investments in renewable energies, especially solar energy, increases the search for more reliable and profitable generation systems. Monitoring the operation of these systems in order to analyze power losses, as well as to identify anomalies and their possible causes, is one of the alternatives, and automated methods, including those supported by machine learning algorithms, are increasingly relevant in this scenario. This research resulted in a convolutional neural network (CNN) model, capable of identifying whether the operating condition of a photovoltaic system (PV system) is normal or under some anomaly. The model was trained using digital images with the graphs of the voltage behavior as a function of the electric current generated (I-V curve), produced through mathematical modeling and simulation of the operation of a PV system’s equivalent circuit. operation of a PV system’s equivalent circuit. Normal operational condition, mismatch (tolerance in electrical parameters depending on manufacturing methods and materials, degradation due to age or external events or agents, etc.), short circuit, open circuit, and partial shading were evaluated alone or combined. In order to ensure wide use and offer adequate precision in the identification of the operating condition, the work was developed in the most varied scenarios, with several arrangements of dozens of photovoltaic panels operating under various temperatures and radiations, resulting in the generation of hundreds of thousands of images of the respective characteristic curves I-V. Significant results were obtained demonstrating that the proposed modeling method and the CNN model offer generalization capacity for different types of photovoltaic panels, operating in a wide range of temperatures and radiations.