Advanced Signal Decomposition Analysis and Anomaly Detection in Photovoltaic Systems

Abstract

With the rapid expansion of large-scale photovoltaic (PV) plants, it is paramount for solar stakeholders to understand the reliability and efficiency of their plants to inform maintenance decisions, increase production, and understand the design factors that impact performance. Diagnosing underperformance in PV plants is challenging due to the relatively few monitoring points with respect to the large geographic footprint of the plant. This study introduces a cutting-edge method that transforms the analysis and management of key factors influencing PV plant performance, including performance loss rate, recoverable soiling, and major system changes. Identifying these factors is critical for deriving actionable insights. Leveraging advanced analytical techniques, such as wavelet transformation, robust regression, and extreme point analysis, this approach provides a nuanced understanding of these factors. This method has been tested across two synthetic datasets and one real dataset, consistently surpassing existing benchmarks by achieving a lower median mean absolute error and reduced error variability across all comparable components.