Aplicação de técnicas de inteligência artificial para identificação de faltas em módulos fotovoltaicos

Abstract

Photovoltaic solar energy has proven to be a viable alternative that contributes not only to sustainable development but also to ensuring energy supply around the world. The exponential growth of installed capacity in recent years has highlighted the need to ensure the safe operation and reliability of photovoltaic systems. In this context, the occurrence of faults in such systems is a crucial issue, as it can significantly impact the generated power, decrease the modules lifetime, and cause potential risks in the operation. Thus, this research applied artificial intelligence techniques to detect and diagnose faults in photovoltaic modules. The faults identified by the proposed methods are short-circuit modules, string disconnection, and partial shading. Algorithms that detect isolated faults were developed, namely: multilayer perceptron neural network, probabilistic neural network, and a neuro-fuzzy method, which combines the use of a neural network with fuzzy logic. All trained algorithms used data simulated through MATLAB/Simulink® software and tested with experimental data from three different photovoltaic systems. Two of the studied photovoltaic systems are power plants installed at the University of Huddersfield, with 2.2 kWp and 4.16 kWp of installed power. The third photovoltaic system has a maximum power of 5 kWp and is installed at the Federal Technological University of Paraná. Additionally, training situations in which the dataset was contaminated by random noise were also considered. The results indicated maximum accuracy of 99.1% for the lack of short-circuited modules, 100% for string disconnection, and 82.2% for the lack of partial shading. Furthermore, the analyzes allowed to reaffirm the robustness of the multi-layer perceptron network for fault detection in photovoltaic systems, even with the presence of noise in the training data.