환경데이터 기반 전압- 전류 곡선 시뮬레이션 및 태양광 스트링 인공지능 고장진단 방법

Abstract

This paper focuses on the development and analysis of a solar photovoltaic (PV) combiner box equipped with a module capable of obtaining Voltage–Current (I-V) characteristic curves to evaluate the performance degradation and faults in a solar power plant. A deep learning algorithm was proposed for the analysis of I-V curves. Direct Current circuit breakers were applied to obtain the I-V curves of each string circuit inside the combiner box. The fault dataset was labeled through classification based on the shape of I-V characteristic curves. The fault data were directly collected through field diagnosis at the power plant, amounting to a dataset of 3,200. A GoogLeNet transfer deep learning model with a convolution neural network (CNN) was used to develop a fault diagnosis algorithm, achieving training and validation accuracies of more than 95%. Empirical research was conducted at a 1.5 MW solar power plant site. Shading and bypass diode short-circuit faults were simulated in solar PV strings to evaluate the accuracy of the algorithm. The artificial intelligence algorithm demonstrated accuracy rates of 98.67% for shading simulation and 95.06% for bypass diode short-circuit fault simulation.