Chasing the Sun: A Journey in monitoring solar panels: Data science approaches for analysis of PV performance

Abstract

Solar photovoltaic (PV) energy is a decentralized renewable energy source that can be installed in diverse locations, from open fields to residential rooftops, for electricity generation. Rooftop-mounted residential PV systems are widely used, contributing to the European Union's plan for zero-energy buildings and empowering households to become small-scale energy producers. PV systems offer low maintenance requirements compared to other energy sources. However, malfunctions can occur, leading to significant energy losses. Consequently, monitoring PV systems is crucial to ensure proper functioning and grid stability. In the Netherlands, most PV systems are small-scale installations on residential and commercial rooftops, with exponential growth anticipated since 2010. Monitoring these systems presents challenges, including the lack of tilted solar irradiance measurements, shading from surrounding objects, and the time-intensive nature of supervised monitoring. To address these challenges, researchers have developed algorithms. One study conducted a public campaign in the Netherlands, where participants measured PV system energy generation during specific weeks. Analyzing weekly yield and performance ratios revealed overall satisfactory performance. However, deeper analysis uncovered issues such as undersized inverters and shading impacts. Citizen science approaches proved valuable, emphasizing the need for unsupervised monitoring of rooftop-mounted PV systems. Another research paper introduced the "Real PR" monitoring algorithm, which detects and isolates malfunctioning PV systems. By comparing energy production to reference data, the algorithm identifies deviations from linearity, indicating anomalies or malfunctions. This enables the detection of underperforming systems. Additionally, a data analysis method addresses challenges in partially shaded PV systems. By comparing long-term and high-resolution yield data between shaded and unshaded PV systems, the algorithm automatically detects energy losses caused by expected shading from surrounding obstacles. This helps distinguish shading-related losses from other malfunctions. Furthermore, a novel algorithmic tool tackles two major issues in residential PV systems: the absence of reference data sources and noisy data sets. The tool combines data from nearby unshaded PV systems to create a reference data set, enabling the comparison and detection of energy losses in the target PV system caused by shading from surrounding objects. To summarize, this dissertation investigates various aspects of PV system performance on residential rooftops, including owner awareness, PV technology comparisons, unsupervised malfunction detection, unsupervised shadow detection, reference set selection, and noise reduction in monitoring data. These findings enhance our understanding and monitoring of PV system performance, supporting the effective implementation of solar photovoltaic energy as a decentralized renewable energy source.