SOLAR PHOTOVOLTAIC (PV) SYSTEM FAILURE EVALUATION IN PARTIAL SHADING CONDITIONS USING MACHINE LEARNING METHODS (REGRESSION)

• Main Author: Sela Rahman Faza, Noer
• Format: Final Project
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/74425
• Institution: Institut Teknologi Bandung

Solar Power Generation System (PLTS) is one of the high-potential renewable energy utilization methods in Indonesia. The PLTS system is currently undergoing development in Indonesia. However, there are challenges in its development caused by the failure of the PLTS system. One of the common causes of PLTS system failure is partial shading, which results in a decrease in power output. Partial shading refers to the condition where PV modules are covered by shadows or dirt. The decrease in power output is influenced by the amount of shading that occurs on the PV modules. Therefore, an evaluation of the failure due to power output reduction is needed to prevent a decrease in the efficiency of the PLTS system, and an estimation system for the amount of shading on the Photovoltaic module is required. The average power loss from the normal condition to the shading condition of 11 modules is 97.49%. In the normal condition, the efficiency of the PV module reaches 13.27%, but in the shading condition of 11 PV modules, the efficiency of the PLTS system is 0.32%. The reduction that occurs is 12.95%. The estimation of the amount of shading on the Photovoltaic module is performed using machine learning methods, including Support Vector Regression, Random Forest, Decision Tree Regressor, and Extra Trees Regressor. These methods are compared based on the estimation accuracy determined through three performance indicators: RMSE (Root Mean Squared Error), MAE (Mean Absolute Error), and Rsquare, as well as training and testing time. The model with the Extra Trees Regressor method is the best model with an RMSE value of 0.307, MAE value of 0.113, and Rsquare value of 0.994, with a training time of 111.200 milliseconds and testing time of 1.130 milliseconds.