5. 3 Operation, Performance and Maintenance of PV Systems
Summary / Abstract:
When estimating performance loss rates (PLR) of PV systems, proper filtering of operational and environmental data is essential to reduce bias and standard errors in the results. Uncertainties can be introduced by shading, soiling, mismatch losses, misalignment between sensors or panels, and more. These uncertainties are propagated into the performance metrics used for calculating the PLR, and ultimately the estimated PLR. Several filters can be applied to the data to remove data associated with high uncertainty and is often required to obtain PLR-results with sufficiently small confidence intervals. At the same time, small variations in how the filtering is done can have large impact on the calculated results, and care must be taken during the filtering process to not introduce bias. Finding appropriate values for the different filtering parameters can be challenging, however, especially when several filters are applied simultaneously. It is therefore a strong need for methods and procedures to set filter parameters in a way that is reproducible, efficient, and limits the influence of the analyst. In this work, we explore how Bayesian Optimization (BO) can be used to systematically obtain optimal values for the filter parameters with respect to the width of the confidence intervals. At the same time, we use the presented tools to gain insight on how filtering impacts bias and confidence intervals across the explored parameter space. The approach has been tested with the year-on-year (YOY) method for calculating PLRs, on a modelled PV system in Norway based on measured weather data.