Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Parallel-friendly Spatio-Temporal Graph Learning for Photovoltaic Degradation Analysis at Scale

Yangxin Fan, Raymond Wieser, Laura Bruckman, Roger French, Yinghui Wu

TL;DR

This work tackles fleet-scale estimation of long-term PV performance loss by decoupling a non-monotonic aging trend from seasonal fluctuations using ST-GTrend, a spatio-temporal graph autoencoder framework. The model employs a parallel array of GAEs to learn an aging term $h_a$ and multiple fluctuation terms $h_f$, optimized via a loss that enforces reconstruction accuracy while enforcing flatness and smoothness to achieve disentanglement. To scale to large PV networks, the authors introduce Para-GTrend, a three-level parallelization scheme that combines model, data, and pipeline parallelism, achieving near-linear speedups on HPC infrastructure. Experiments across PV, finance, and economy datasets show ST-GTrend outperforms state-of-the-art baselines in MAPE and ED for PLR estimation and demonstrate substantial scalability, with practical implications for improving LCOE calculations and fleet maintenance planning.

Abstract

We propose a novel Spatio-Temporal Graph Neural Network empowered trend analysis approach (ST-GTrend) to perform fleet-level performance degradation analysis for Photovoltaic (PV) power networks. PV power stations have become an integral 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 challenging problems in assessing the Levelized Cost of Energy (LCOE) of a PV system is to understand and estimate the long-term Performance Loss Rate (PLR) for large fleets of PV inverters. ST-GTrend integrates spatio-temporal coherence and graph attention to separate PLR as a long-term "aging" trend from multiple fluctuation terms in the PV input data. To cope with diverse degradation patterns in timeseries, ST-GTrend adopts a paralleled graph autoencoder array to extract aging and fluctuation terms simultaneously. ST-GTrend imposes flatness and smoothness regularization to ensure the disentanglement between aging and fluctuation. To scale the analysis to large PV systems, we also introduce Para-GTrend, a parallel algorithm to accelerate the training and inference of ST-GTrend. We have evaluated ST-GTrend on three large-scale PV datasets, spanning a time period of 10 years. Our results show that ST-GTrend reduces Mean Absolute Percent Error (MAPE) and Euclidean Distances by 34.74% and 33.66% compared to the SOTA methods. Our results demonstrate that Para-GTrend can speed up ST-GTrend by up to 7.92 times. We further verify the generality and effectiveness of ST-GTrend for trend analysis using financial and economic datasets.

Parallel-friendly Spatio-Temporal Graph Learning for Photovoltaic Degradation Analysis at Scale

TL;DR

This work tackles fleet-scale estimation of long-term PV performance loss by decoupling a non-monotonic aging trend from seasonal fluctuations using ST-GTrend, a spatio-temporal graph autoencoder framework. The model employs a parallel array of GAEs to learn an aging term and multiple fluctuation terms , optimized via a loss that enforces reconstruction accuracy while enforcing flatness and smoothness to achieve disentanglement. To scale to large PV networks, the authors introduce Para-GTrend, a three-level parallelization scheme that combines model, data, and pipeline parallelism, achieving near-linear speedups on HPC infrastructure. Experiments across PV, finance, and economy datasets show ST-GTrend outperforms state-of-the-art baselines in MAPE and ED for PLR estimation and demonstrate substantial scalability, with practical implications for improving LCOE calculations and fleet maintenance planning.

Abstract

We propose a novel Spatio-Temporal Graph Neural Network empowered trend analysis approach (ST-GTrend) to perform fleet-level performance degradation analysis for Photovoltaic (PV) power networks. PV power stations have become an integral 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 challenging problems in assessing the Levelized Cost of Energy (LCOE) of a PV system is to understand and estimate the long-term Performance Loss Rate (PLR) for large fleets of PV inverters. ST-GTrend integrates spatio-temporal coherence and graph attention to separate PLR as a long-term "aging" trend from multiple fluctuation terms in the PV input data. To cope with diverse degradation patterns in timeseries, ST-GTrend adopts a paralleled graph autoencoder array to extract aging and fluctuation terms simultaneously. ST-GTrend imposes flatness and smoothness regularization to ensure the disentanglement between aging and fluctuation. To scale the analysis to large PV systems, we also introduce Para-GTrend, a parallel algorithm to accelerate the training and inference of ST-GTrend. We have evaluated ST-GTrend on three large-scale PV datasets, spanning a time period of 10 years. Our results show that ST-GTrend reduces Mean Absolute Percent Error (MAPE) and Euclidean Distances by 34.74% and 33.66% compared to the SOTA methods. Our results demonstrate that Para-GTrend can speed up ST-GTrend by up to 7.92 times. We further verify the generality and effectiveness of ST-GTrend for trend analysis using financial and economic datasets.
Paper Structure (8 sections, 1 theorem, 10 equations, 9 figures, 2 tables, 1 algorithm)

This paper contains 8 sections, 1 theorem, 10 equations, 9 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminary
  3. ST-GTrend Framework
  4. Parallelization and Deployment
  5. Experiments
  6. Experimental Setup
  7. Experimental Results
  8. Conclusion

Key Result

theorem 1

The training algorithm Para-GTrend is scale-free with a total parallel cost in $O\left(\frac{T(G, M)}{\lvert P \rvert} + f(\theta)\right)$ time, where $T(G, M)$ is total cost without parallelism and $f(\theta)$ is independent of the size of $G_t$ and linear to the length of timeseries.

Figures (9)

  • Figure 1: Example of a PV system with 10-years power timeseries exhibiting non-monotonic degradation pattern.
  • Figure 2: Overview of ST-GTrend Framework (six nodes shown in $G_{t}$ for illustration).
  • Figure 3: : Para-GTrend
  • Figure 4: Proposed Para-GTrend ($k = 3$ for illustration: one aging channel and three fluctuation channels).
  • Figure 5: Illustration of ST-GTrend Workflow with Level I Parallelism Implemented for PV Systems from Our Industry Partners (deployed in CRADLE nihar2021toward).
  • ...and 4 more figures

Theorems & Definitions (2)

  • definition 1
  • theorem 1