Photovoltaic Site Architecture Estimation Using Performance Data

Most photovoltaic power generation sites schedule maintenance as a result of physical inspections and observations. For example, a site may use aerial infrared imaging to determine the fault status of individual combiner boxes and/or strings. However, the costs to perform aerial scans result in infrequent, typically annual, application. As a result, DC faults can often go unnoticed for months at a time. While this repetitive, expensive task is attractive for automation, the limited granularity of modern sensor suites makes it difficult. The authors' work has enabled continuous, real-time PV anomaly detection using existing, installed sensor suites. This relies on a detailed knowledge of the site layout to correctly predict expected performance. Previously this information was manually codified using available site drawings for each site. However, the manual review and codification of metadata is time-consuming, increasing the investment required for an M&D center to implement the code. This difficulty is exacerbated by the competitiveness of the PV market, which has led to leaner O&M investments. This work presents a new method to estimate the site architecture using performance data and a fraction of the metadata. The setup speed is accelerated by more than a factor of 15, while achieving similar anomaly detection quality to the previous work using manually codified site layouts.