FAILURE DETECTION OF PLTS SYSTEM IN OPERATION MODE AND PERFORMANCE DEGRADATION USING XGBOOST METHOD

• Main Author: Fuad Hassan Siregar, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/83735
• Institution: Institut Teknologi Bandung

Solar technology is one of the renewable energy technologies with significant sustainable potential because solar radiation is available worldwide and does not produce greenhouse gas emissions. The solar energy potential in Indonesia is substantial, around 4.8 KWh/m² or equivalent to 112,000 GWp, but only about 10 MWp has been utilized. Photovoltaic (PV) is a technology that converts solar energy into electricity and is the main component of Solar Power Plants (PLTS). Despite significant advancements in PV technology, challenges such as performance degradation and the decrease in Performance Ratio (PR) persist. Physical modeling and machine learning can aid in detecting faults in PLTS. This final project employs the XGBoost algorithm to detect, classify, and perform PR regression on a hybrid PLTS system at the Energy Management Laboratory (Lab.ME) ITB. Operational data were collected over one year with minute-level resolution. The research results indicate that essential features for detecting operational modes were determined using Pearson correlation methods and theoretical considerations, resulting in 15 important features. The PR calculation showed an error rate of only 1.49%. Classification of operational modes with XGBoost demonstrated high accuracy, reaching 99.49%, and 99.87% with hyperparameter tuning. PR regression using XGBoost showed good performance with RMSE and MAE values close to 0, though Linear Regression was less capable of representing annual PR changes. Weekly frequency provided the most stable representation of annual PR changes. This approach is expected to produce a more accurate and efficient fault detection system, supporting the broader development of renewable energy in Indonesia. Keywords: renewable energy, photovoltaic, XGBoost, Performance Ratio, fault detection, machine learning.