Machine Learning allows Synthesis and Functional Interpolation of Computational and Field-Data for Worldwide Utility-Scale PV Systems

There has been an accelerated pace of installation variety of utility-scale solar farms across the world. Field data are streaming in from existing plants. Also, highly sophisticated physics-based numerical models and software are being developed to estimate the energy yield of a solar PV system over its lifetime. By necessity, the results are specific to a finite number of arbitrarily chosen geographical locations and require substantial computational resources for a worldwide energy estimation, making it difficult to extrapolate the results to new locations. In this paper, we demonstrate a novel machine learning-based tool that reduces the computational time by 4-orders of magnitude with an accuracy of R
²
~ 0.99. The tool utilizes the simulation-derived utility-scale solar PV energy yield to functionally interpolate the location-specific inputs and energy yields. This compact, efficient, and versatile representation will transform how large-scale modeling is used to predict the energy yield of various farm configurations for any location around the world.