Can We Reliably Improve the Robustness to Image Acquisition of Remote Sensing of PV Systems?

Photovoltaic (PV) energy is crucial for the decarbonization of energy systems. Due to the lack of centralized data, remote sensing of rooftop PV installations is the best option to monitor the evolution of the rooftop PV installed fleet at a regional scale. However, current techniques lack reliability and are notably sensitive to shifts in the acquisition conditions. To overcome this, we leverage the wavelet scale attribution method (WCAM), which decomposes a model's prediction in the space-scale domain. The WCAM enables us to assess on which scales the representation of a PV model rests and provides insights to derive methods that improve the robustness to acquisition conditions, thus increasing trust in deep learning systems to encourage their use for the safe integration of clean energy in electric systems.

PyPVRoof: a Python package for extracting the characteristics of rooftop PV installations using remote sensing data

Photovoltaic (PV) energy grows at an unprecedented pace, which makes it difficult to maintain up-to-date and accurate PV registries, which are critical for many applications such as PV power generation estimation. This lack of qualitative data is especially true in the case of rooftop PV installations. As a result, extensive efforts are put into the constitution of PV inventories. However, although valuable, these registries cannot be directly used for monitoring the deployment of PV or estimating the PV power generation, as these tasks usually require PV systems {\it characteristics}. To seamlessly extract these characteristics from the global inventories, we introduce {\tt PyPVRoof}. {\tt PyPVRoof} is a Python package to extract essential PV installation characteristics. These characteristics are tilt angle, azimuth, surface, localization, and installed capacity. {\tt PyPVRoof} is designed to cover all use cases regarding data availability and user needs and is based on a benchmark of the best existing methods. Data for replicating our accuracy benchmarks are available on our Zenodo repository \cite{tremenbert2023pypvroof}, and the package code is accessible at this URL: \url{https://github.com/gabrielkasmi/pypvroof}.

A crowdsourced dataset of aerial images with annotated solar photovoltaic arrays and installation metadata

Photovoltaic (PV) energy generation plays a crucial role in the energy transition. Small-scale PV installations are deployed at an unprecedented pace, and their integration into the grid can be challenging since public authorities often lack quality data about them. Overhead imagery is increasingly used to improve the knowledge of residential PV installations with machine learning models capable of automatically mapping these installations. However, these models cannot be easily transferred from one region or data source to another due to differences in image acquisition. To address this issue known as domain shift and foster the development of PV array mapping pipelines, we propose a dataset containing aerial images, annotations, and segmentation masks. We provide installation metadata for more than 28,000 installations. We provide ground truth segmentation masks for 13,000 installations, including 7,000 with annotations for two different image providers. Finally, we provide installation metadata that matches the annotation for more than 8,000 installations. Dataset applications include end-to-end PV registry construction, robust PV installations mapping, and analysis of crowdsourced datasets.

DeepSolar tracker: towards unsupervised assessment with open-source data of the accuracy of deep learning-based distributed PV mapping

Photovoltaic (PV) energy is key to mitigating the current energy crisis. However, distributed PV generation, which amounts to half of the PV energy generation, makes it increasingly difficult for transmission system operators (TSOs) to balance the load and supply and avoid grid congestions. Indeed, in the absence of measurements, estimating the distributed PV generation is tough. In recent years, many remote sensing-based approaches have been proposed to map distributed PV installations. However, to be applicable in industrial settings, one needs to assess the accuracy of the mapping over the whole deployment area. We build on existing work to propose an automated PV registry pipeline. This pipeline automatically generates a dataset recording all distributed PV installations' location, area, installed capacity, and tilt angle. It only requires aerial orthoimagery and topological data, both of which are freely accessible online. In order to assess the accuracy of the registry, we propose an unsupervised method based on the {\it Registre national d'installation} (RNI), that centralizes all individual PV systems aggregated at communal level, enabling practitioners to assess the accuracy of the registry and eventually remove outliers. We deploy our model on 9 French {\it d\'epartements} covering more than 50 000 square kilometers, providing the largest mapping of distributed PV panels with this level of detail to date. We then demonstrate how practitioners can use our unsupervised accuracy assessment method to assess the accuracy of the outputs. In particular, we show how it can easily identify outliers in the detections. Overall, our approach paves the way for a safer integration of deep learning-based pipelines for remote PV mapping. Code is available at {\tt https://github.com/gabrielkasmi/dsfrance}.

Machine learning models show similar performance to Renewables.ninja for generation of long-term wind power time series even without location information

Driven by climatic processes, wind power generation is inherently variable. Long-term simulated wind power time series are therefore an essential component for understanding the temporal availability of wind power and its integration into future renewable energy systems. In the recent past, mainly power curve based models such as Renewables.ninja (RN) have been used for deriving synthetic time series for wind power generation despite their need for accurate location information as well as for bias correction, and their insufficient replication of extreme events and short-term power ramps. We assess how time series generated by machine learning models (MLM) compare to RN in terms of their ability to replicate the characteristics of observed nationally aggregated wind power generation for Germany. Hence, we apply neural networks to one MERRA2 reanalysis wind speed input dataset with no location information and one with basic location information. The resulting time series and the RN time series are compared with actual generation. Both MLM time series feature equal or even better time series quality than RN depending on the characteristics considered. We conclude that MLM models can, even when reducing information on turbine locations and turbine types, produce time series of at least equal quality to RN.