DETECTION OF SOLAR POWER PLANT SYSTEM ANOMALIES USING SUPPORT VECTOR MACHINE BASED ON I-V CURVE CHARACTERISTICS

• Main Author: Satya Widhitama, Abdeebarr
• Format: Final Project
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/77375
• Institution: Institut Teknologi Bandung

Indonesia as one of the countries with abundant natural resources has its own advantages in utilizing the many sources of renewable energy. Indonesia has various types of EBT (Energi Baru Terbarukan) sources such as hydro, biomass, and solar energy. Seeing Indonesia as a tropical country, the potential for electrical energy generated through solar power plants (PLTS) can be considered quite large. Besides that, solar power plants itself is just a technology that can fail which can result in a decrease in the efficiency and potential of the system. One of the failures that can cause a decrease in the efficiency of solar power plants is the presence of shading. The shading in this solar power plants system can be detected using machine learning. In this study, the failure detection system that will be used is I-V curve detection using a support-vector machine (SVM) based algorithm. This algorithm will compare training data with test data to be able to determine whether a failure has been detected in the existing solar power plants system. This research is expected to reduce the risk of damage to solar power plants caused by failure of the shading type. In this study, there were 4 shading conditions in the solar power plants system, namely shading 1 PV module, shading 3 PV modules, shading 6 PV modules, and shading 9 PV modules. From the four shading conditions, an average efficiency decrease of 5.02% was obtained and an average decrease in fill factor was 27.15% of the efficiency and fill factor of the PV string. Modeling using SVM can very well detect the type of shading that occurs in the PLTS system. This is evidenced by the accuracy value which is the value of the entire modeling reaching 0.98 out of 1.