Data-Driven Machine Learning and Visualization for Solar PV System Fault Detection and Power Forecasting

With the continuous increase in the relevance of solar energy in the world renewable energy portfolio, the concern of ensuring that solar photovoltaic (PV) systems remain dependable and effective has acquired a paramount importance. The paper implements the use of sophisticated data analytics and visualization tools to examine the trends in the generation of renewable energy and find out the origin of the faults in the system. Two sets of primary data, namely, weather sensor data (ambient temperature, module temperature, irradiation) and inverter generation data (DC power, AC power, daily yield, total yield), are the basis of the analysis. After thorough preprocessing, a new labeled fault column was introduced to help to conduct supervised classification and support fault detection. All kinds of visual tools were introduced to explore the relationships between features and the results of their performance, including correlation heatmaps, violin plots, confusion matrices, sunburst charts, etc. The final classification model achieved a high accuracy of 99.26%, and the values of precision, recall, and F1, scores, the final classification model was well balanced between normal and faulty classes. These results indicate that information-motivated solutions can be the path toward a more intelligent solar PV management, increased energy generation control, and better system diagnosis.