Parallelizable Spatio-Temporal Graph Learning for Photovoltaic Degradation Analysis

Photovoltaic (PV) systems have become an important component to the global sustainable energy production landscape. Accurately estimating the performance of PV systems is critical to their feasibility as a power generation technology and as a financial asset. One of the most fundamental problems in assessing the Unit Energy Cost of a PV system application is to understand and estimate the long-term Performance Loss Rate (PLR) for large-scale PV fleets. We propose a novel parallel-friendly Spatio-Temporal Dynamic Graph Neural Network (Para-st-DynGNN) to perform large-scale fleet-wide PLR estimation. Para-st-DynGNN integrates spatio-temporal coherence and graph attention to extract PLR as a long-term "aging" trend by filtering out multiple fluctuation terms in the PV input data. Para-st-DynGNN leverages a three-level think-like parallelisms including the think-like branch model parallelism, the think-like snapshot data parallelism, and the think-like layer pipeline parallelism. Our results show that Para-st-DynGNN reduces Mean Absolute Percent Error (MAPE) by 34.74% compared to the state-of-the-art methods while achieving upto 7.56 times as we increase the number of processors from 1 to 8.