CLASSIFICATION OF PV FAILURES USING SVC AND XGBOOST METHODS BASED ON SIMULATION AND EXPERIMENTS OF I-V CURVE CHARACTERISTICS

• Main Author: Cahyadi, Aulia
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/83647
• Institution: Institut Teknologi Bandung

In order to achieve the target of a 23% renewable energy mix by the end of 2025, Indonesia has set a target of implementing rooftop solar power plants (PLTS) to reach 3.6 GW in that year. However, until 2023, the implementation figure is still far from the target. In addition to the bureaucratic aspect, the attractiveness of implementing rooftop PLTS depends on the cooling system in producing energy and the service life of the components. To improve these two aspects, the operation of the PLTS system must be in optimal condition and far from failure conditions. This study focuses on maximizing performance by designing a monitoring system that can classify the condition of the PLTS system. This condition classification aims to detect the condition of the system, whether it is normal or experiencing failure so that mitigation action can be more effective. The failures reviewed include line to line, or the creation of a low-resistance electrical path in the middle of the panel series, with a mismatch of 3,6, and 9 modules. The mismatch here is the number of modules that are skipped between 2 line to line location points. The second failure is partial shading at 3,6 and 9 modules. Finally, the third failure reviewed for classification is an open circuit in one of the PLTS series. The study was conducted on a PLTS array object consisting of 2 strings, or 2 series of modules. This condition review was carried out through experiments and simulations of the PLTS model using Matlab Simulink software. The results of the failure observations showed that there was almost the same decrease in energy production even though the failure conditions were different. Therefore, the classification of system conditions based on other parameters was used. The characteristics of the system's IV curve are used as the basis for classification using machine learning methods based on SVC (support vector classifier) and XGBoost (extreme gradient boosting). A performance comparison between these 2 models was used and resulted in SVC as a model that classifies failure conditions better with an accuracy value of up to 0.97 out of 1. Keywords: Solar power plant, failure classification, line to line, open circuit, partial shade, modeling, IV curve.