<![CDATA[Solar photovoltaic (PV) power plants are increasingly deployed in tropical regions such as Indonesia, yet their performance is often degraded by undetected faults including partial shading, dust accumulation, and module mismatch. This study presents a computational intelligence framework for real-time monitoring and fault diagnosis of grid-connected PV systems from a computer science perspective. The framework consists of three main components: (1) a data acquisition module that simulates 12 months of PV system operation (25 kWp capacity) using meteorological data from Medan, Indonesia, generating 8,760 hourly samples of voltage, current, power, irradiance, and temperature; (2) a machine learning-based fault classifier using Random Forest (RF) and Support Vector Machine (SVM) algorithms to distinguish between four fault types (normal operation, partial shading, dust accumulation, and module mismatch) and one healthy state; and (3) a web-based dashboard built with PHP and MySQL for real-time visualization and alerting. Experimental results show that the Random Forest classifier achieves 97.3% accuracy, 95.8% precision, and 96.2% recall, outperforming SVM (91.6% accuracy). The algorithm detects faults within 1.8 seconds of occurrence, enabling rapid operator response. The proposed system is implemented as an open-source prototype and can be deployed on low-cost hardware (Raspberry Pi 4) with an average response time of 1.8 seconds. The framework is validated using tropical climate data from Medan, Indonesia, addressing a gap in existing PV fault diagnosis research. This research contributes a practical, software-based fault diagnosis tool for PV system operators in tropical environments]]>
Copyrights © 2026
