Remote-Sensing-Based Estimation of Rooftop Photovoltaic Power Production Using Physical Conversion Models and Weather Data

Gabriel Kasmi, Augustin Touron, Philippe Blanc, Yves‐Marie Saint‐Drenan, Maxime Fortin, Laurent DubusEnergies 2024 · DOI

TL;DR

This paper tackles the challenge of accurately estimating rooftop solar panel (PV) power output, a critical step for integrating this growing energy source into the grid. By combining remote sensing data for PV system characteristics with weather data and a simple conversion model, the researchers achieved an average estimation error of about 10%, significantly improving "observability" for grid operators.

What Problem Does This Paper Address?

The rapid growth of solar PV installations, especially rooftop systems, presents a challenge for grid operators. These smaller, distributed systems are often not directly monitored, leading to a lack of "observability" – the ability to accurately know their real-time power production. This gap hinders effective grid management, potentially causing imbalances, congestion, and increased reserve requirements. While larger PV plants are generally well-monitored, rooftop PV systems, which constitute a significant portion of installed capacity in countries like France, remain largely unobserved. This paper aims to bridge this gap by developing a method to estimate rooftop PV power production using readily available data.

Methods

The core of the paper's methodology involves using a physical conversion model, specifically the PV Watts model, to estimate PV power production. The inputs to this model are:

  • PV System Characteristics: These are derived from remote sensing techniques, specifically using a deep learning algorithm called DeepPVMapper. This algorithm extracts information like the system's location, tilt and azimuth angles, and installed capacity from aerial imagery and existing PV registries.
  • Weather Data: This includes solar irradiance (global, direct, and diffuse) and 2-meter air temperature data, sourced from the Copernicus Atmosphere Monitoring Service (CAMS) and ERA5 reanalysis data, respectively.

The researchers then compare the power production estimated using these derived characteristics (from DeepPVMapper) against ground-truth measurements. They also establish an "Oracle" estimation, which uses the true, known characteristics from their ground-truth dataset to represent the best possible estimation achievable with the chosen conversion model and weather data.

Datasets

  • Ground-Truth PV Power Measurements: The study utilized a dataset of ground-truth PV power production measurements from 1793 individual rooftop PV systems across mainland France, provided by the non-profit association Asso BDPV. This dataset was filtered and curated to 906 systems after quality checks.
  • PV Registry Data: A PV registry generated by the DeepPVMapper algorithm was used, containing characteristics for over 100,000 installations. For this study, they focused on installations present in both the BDPV dataset and the DeepPVMapper registry, resulting in 276 unique installations for analysis.
  • Weather Data:
    • Solar Irradiance: Copernicus Atmospheric Monitoring Service (CAMS) data, providing 30-minute averaged values of global, direct, and diffuse solar irradiation.
  • Temperature: ERA5 reanalysis data, providing hourly 2-meter air temperature, interpolated to 30-minute intervals.

Key Insights and Ideas

  • Limited Information is Sufficient: The study demonstrates that even with minimal information about rooftop PV systems (location, installed capacity, tilt, and azimuth), it's possible to achieve reasonably accurate power production estimations. This is crucial because detailed information for these small systems is often unavailable.
  • Remote Sensing as a Key Enabler: The integration of remote sensing techniques (specifically DeepPVMapper) to extract PV system characteristics is a critical component. It bridges the gap between satellite/aerial imagery and actual PV system performance.
  • Off-the-Shelf Models Work: The use of a simple, established conversion model like PV Watts, rather than complex, custom-built models, highlights the practicality and scalability of the approach. This suggests that existing tools can be effectively leveraged.
  • Temporal Error Patterns: The estimation error is not constant throughout the day. It tends to be lower at night and peaks during the day, with higher variability observed during winter and at dawn/dusk, potentially due to factors like shading not being explicitly modeled.
  • Geographical Variability: While the overall error is manageable, there are geographical clusters of higher error, particularly in France's southeast and northwest. This is hypothesized to be linked to terrain complexity and potentially more significant shading effects, or variations in the accuracy of the solar irradiance data.

Key Results

  • Average Estimation Error: The study reports an average estimation error, measured by the percentage Root Mean Squared Error (pRMSE), of approximately 10% relative to the system's installed capacity when using characteristics derived from DeepPVMapper.
  • Oracle Comparison: The DeepPVMapper-based estimations showed a pRMSE of 10.10% (mean) compared to the Oracle's 8.36%, indicating a small but measurable difference when using estimated versus true system parameters.
  • Scalability: The analysis suggests that the approach is scalable. As the number of installations considered increases, the aggregated error behaves similarly to the Oracle, indicating no significant systematic biases that would prevent scaling to thousands of systems.
  • Bias Analysis: While the average error can be high, the median error is much lower, suggesting that the method generally has few biases. Interestingly, the DeepPVMapper parameterization showed slightly better bias metrics than the Oracle in some cases.

Limitations and Future Work

  • Limited Aggregated Analysis: The study's analysis of aggregated systems was limited to a few dozen systems, making direct comparison with existing TSO methods challenging. More data is needed for robust comparisons at scale.
  • Detection Errors: The framework relies on the accuracy of the remote-sensing-based registry. Discrepancies between estimated and actual installed capacity due to detection errors are not explicitly addressed.
  • Shading and Complexities: The current model does not account for shading, which could explain some of the observed temporal error patterns (e.g., higher error at dawn/dusk and in winter).
  • Future Work: The authors suggest further work on comparing their approach with TSO practices at the aggregated level, exploring more advanced conversion models (e.g., incorporating shading), and gathering data from more systems to enhance the significance of their findings.

Summary updated

Abstract

The global photovoltaic (PV) installed capacity, vital for the electric sector’s decarbonation, reached 1552.3 GWp in 2023. In France, the capacity stood at 19.9 GWp in April 2024. The growth of the PV installed capacity over a year was nearly 32% worldwide and 15.7% in France. However, integrating PV electricity into grids is hindered by poor knowledge of rooftop PV systems, constituting 20% of France’s installed capacity, and the lack of measurements of the production stemming from these systems. This problem of lack of measurements of the rooftop PV power production is referred to as the lack of observability. Using ground-truth measurements of individual PV systems, available at an unprecedented temporal and spatial scale, we show that by estimating the PV power production of an individual rooftop system by combining solar irradiance and temperature data, the characteristics of the PV system inferred from remote sensing methods and an irradiation-to-electric power conversion model provides accurate estimations of the PV power production. We report an average estimation error (measured with the pRMSE) of 10% relative to the system size. Our study shows that we can improve rooftop PV observability, and thus its integration into the electric grid, using little information on these systems, a simple model of the PV system, and weather data. More broadly, this study shows that limited information is sufficient to derive a reasonably good estimation of the PV power production of small-scale systems.
BibTeX
@inproceedings{kasmi2024remote,
  author = {Gabriel Kasmi and Augustin Touron and Philippe Blanc and Yves‐Marie Saint‐Drenan and Maxime Fortin and Laurent Dubus},
  title = {{Remote-Sensing-Based Estimation of Rooftop Photovoltaic Power Production Using Physical Conversion Models and Weather Data}},
  booktitle = {Energies},
  year = {2024},
  doi = {10.3390/en17174353},
}
Figure