DETECTION OF SOLAR POWER PLANT SYSTEM FAULTS USING DEEP NEURAL NETWORK BASED ON I-V AND P-V CURVE CHARACTERISTICS

• Main Author: Jauhar Sulaiman, Fakhri
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/79869
• Institution: Institut Teknologi Bandung

The demand for decarbonization and the transition to new renewable energy sources are increasing the future demand for Photovoltaic (PV) to generate solar energy. With the rising trend in PV demand, discussions about failures in the Photovoltaic Solar Power System (PLTS) have become necessary. Undetected failures for an extended period have the potential to significantly reduce electricity production. Therefore, it is crucial to develop methods capable of detecting and diagnosing failures that occur in the PLTS. Each type of failure in the PLTS system results in differences in the shapes of the I-V and P-V curves. These differences enable the reading and detection of a specific type of failure by observing and analyzing the I-V and P-V curve shapes over a specific time span. A failure detection system for the PLTS will be developed using the Deep Neural Network (DNN) method with the main variables being the I-V and P-V curves. The focus of the detected failures will be on partial shading due to its frequent occurrence and ease of engineering and variation. Partial shading is applied by covering PV modules in the PLTS system at the ITB CAS Building with a carpet. Various amounts of shading are applied, ranging from 1 to 10 covered PV modules. I-V and P-V curve data are modeled using the DNN algorithm and then compared with modeling results using the Support Vector Machine (SVM) algorithm. The DNN training accuracy is 83.67%, while the SVM training accuracy is 63.89%. For validation accuracy, the DNN machine learning has a validation accuracy of 81.65%, while SVM has a validation accuracy of 64.69%.