A Comprehensive Regression Framework for Photovoltaic Systems: From Data Analysis to Real-Time Notifications

Abstract

This paper suggests a comprehensive and costeffective photovoltaic maintenance system intended to meet the growing demand for smart monitoring solutions in the market. The approach integrates various processes for data collection and analysis, regression modeling, real-time anomaly detection, and automated notification capabilities. The key objective is to enhance remote monitoring, data management, and fault detection by leveraging both historical meteorological and power databases. The workflow begins with the use of two wellestablished representatives of Machine Learning paradigms to produce regression models based on eight variables. Correlation analysis was then applied for feature selection to reduce computational load without compromising overall quality. This action leads to a refined and meaningful subset that was assessed against the first full model using well-known metrics such as the root mean squared error and the coefficient of determination. The results show that the optimized model maintains high accuracy while streamlining the protocol for rapid execution. Finally, to operationalize the method, a unified architecture for frequent data collection was implemented. The pipeline feeds a fault detection mechanism that tracks disparities between projected and real power and flags potential under performance in realtime, prompting appropriate physical inspections. This end-toend setup bridges renewable energy practical applications with the field of Machine Learning. It was specifically built for operators with limited technical expertise, introducing a versatile and affordable strategy to improve the reliability of PV systems.