Teaching and popularization

About 20h of teaching load to a group of 2nd year students of engineering degree at ENSAE Paris ("groupe de statistiques appliquées"). The project title was "Deep learning for detecting individual solar systems from aerial images". The students tasks were to carry out a short literature review, define the fundamental notions of machine learning and deep learning (empirical risk minimization, perceptron, neural network, convolutional layers, CNN) and to implement a custom-made and a ready-to-use neural network for binary classification on BDAPPV. The students leveraged PyTorch to train and evaluate the classification model. The project repository can be accessed here and the students' implementation here.

About 20h of teaching load to a group of 1st year students of the engineering degree at Mines Paris ("projet d'informatique"). The students task was to implement a website to visualize the outputs of DeepPVMapper (in a similar fashion as this website). The main tasks were to implement an architecture with a client and a host based on Flask, to implement spatial indexing and preprocessing of the raw data to minimize the loading times and to propose some complementary statistics (e.g. the number of systems or the cumulated installed capacity per city). The project repository can be accessed here. This project was co-supervised with R. Jolivet.

2021-2022

Internships

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 PyPVRoof. PyPVRoof is a Python package to extract essential PV installation characteristics. These characteristics are tilt angle, azimuth, surface, localization, and installed capacity. 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 DOI , and the package code is accessible at this URL.

The main tasks of the internship included an extensive literature review of the existing works in the field of extracting characteristics of PV systems, a benchmark of these methods and an implementation of new methods (e.g. Theil-Sen regression). All methods were extensively compared both in terms of accuracy and in terms of computational cost on BDAPPV. Yann put a particular emphasis on the environmental impact of his coding practices and the methods that he implemented. The idea to discuss the environmental impact of DeepPVMapper (appendix A of the thesis manuscript) was brought by him. This internship was co-supervised with L. Dubus.

Photovoltaic (PV) energy plays a major role in the energy transition in France and in Europe. Between 2022 and 2023, the PV installed capacity grew by 16.6% in France 26.1% in the European Union. PV energy is characterized by a large diversity of systems and its growth has been possible notably thanks to rooftop PV systems. Measures such as feed-in tariffs incentivized households to adopt PV systems, and currently, the legislative framework has shifted towards encouraging self-consumption [3]. In the former, the user was paid a fixed amount for its energy produced. In the latter, the user is no longer paid for the energy produced by its system but instead directly consumes its energy.

In the context of strong growth, self-consumption will impact on the electric system and will affect the depth of PV adoption. The motivations behind self-consumption are not only purely economic or technical, as ethical factors also come into play. On the other hand, access to self-consumption is uneven, and its requirements (e.g., home ownership) can restrict the diffusion of rooftop PV systems among some underprivileged communities.

Understanding the drivers of rooftop PV adoption is of interest for to wide range of stakeholders, ranging from policymakers to utilities such as transmission system operators. On the one hand, it can help assess whether the current legislative framework is adapted to meet the adoption goals and determine if additional mechanisms should be implemented or potential limitations should be lifted to foster PV adoption. On the other hand, it can help utilities such as TSOs to calibrate their projections regarding the adoption of rooftop PV, thus preventing potential under or over-investments in the system. One of the key limitations for studying the drivers of residential PV adoption is the lack of disaggregated data regarding the PV systems. Over the last years, this limitation has been gradually lifted in several European countries and regions such as France, the Netherlands, Northern Italy, or the German state of North Rhine-Westphalia. Indeed, these regions’ rooftop PV systems have been mapped thanks to overhead imagery and artificial intelligence algorithms.

The goal of the internship is to study the drivers behind residential solar adoption in Europe, with an emphasis on France. Using the rooftop PV registry acquired by RTE with DeepPVMapper, the intern will build on existing works to construct a methodology to identify the socio-economic and demographic drivers behind PV adoption.

Talks and popularization

Press articles