Dimensionality Reduction of Environmental Data for Long-Term PV Performance Analysis Using Graph Based Methods

This study presents an innovative approach to solar photovoltaic (PV) performance analysis through the integration of temperature and solar irradiance data using a graph-based community detection method. While previous research often considered these factors independently, our method captures their complex interactions, offering a more comprehensive insight into the behaviour of solar PV systems. We introduce a novel data reduction technique that identifies 10 representative days from a year's worth of hourly data, significantly reducing computational requirements while preserving crucial microclimate patterns relevant to PV performance. Our methodology employs a graph-based community detection algorithm, representing days as nodes and similarities as edges, enabling the identification of nonspherical clusters that better represent complex weather patterns. We evaluated three normalization techniques— standard, min max, and robust—finding minmax normalization to yield the best-defined clusters for our dataset. The effectiveness of our approach is demonstrated using hourly average temperature and irradiance data. Results show that our method successfully captures annual weather patterns while substantially reducing data volume. This research will contribute to more efficient and accurate solar PV performance evaluations, balancing comprehensive data analysis with computational efficiency.