Enhancing Photovoltaic Maintenance with Real-Time Data Integration: A Sensorless And Scalable Approach Validated By Intelligent Correlation Analysis

This study introduces an innovative and cost-effective approach to gathering, storing and managing photovoltaic and meteorological data, avoiding the use of traditional sensors, wires and cables. Conventionally, systems for data collection often require costly tools as well as sophisticated cabling networks; they revolve around the same methodologies with difficult setups and complicated maintenance needs, which can limit their application, long-term reliability and scalability in countries with budget constraints and challenging environmental conditions such as Morocco. The presented solution leverages built-in functionalities of photovoltaic inverters to collect crucial data such as real-time generated power, current, voltage and consumed energy. Regarding weather parameters, a set of online open-source techniques has been explored to compile all necessary data required by the researchers, especially those working merely with simulated metrics, to successfully carry out their projects. Machine learning models have been developed to forecast PV power and verify the accuracy of the data, showing a high degree of correlation with inverter-based information. Particularly, decision tree regression and support vector regression were employed. The results on the test sets were as follows: the decision tree regression model achieved a mean squared error of 0.2199 and an $\mathbf{R}^{2}$ score of 0.7769, while the support vector regression model presented a mean squared error of 0.0936 and a higher $\mathbf{R}^{2}$ score of 0.9051. In the end, an online accessible dashboard that showcases important graphs and includes a notification center to alert users as the need arises has been designed, representing a significant step forward in the quest for photovoltaic system maintenance excellence.