Energy Efficiency and Renewables meet Big Data – CISBAT2019
Author
Beatrice Huber
NRP 75 “Big Data”
CISBAT2019, an international conference on climate resilient cities, energy efficiency and renewables in the digital era, aims to further sustainability in the built environment through interdisciplinary dialogue and presentations of innovative research and development. Innovative research also means Big Data and therefore two sessions were dedicated to data-driven methods where work done within NRP 75 was presented.
CISBAT2019 was held from 4 to 6 September in Lausanne at the EPFL organised by the EPFL Solar Energy and Building Physics Laboratory. The conference has a long tradition. The first national conference was held in 1979, since 1991 the conference is international.
After Welcome Addresses by Jean-Louis Scartezzini, chair of the conference and PI of an NRP 75 project, and Rolf Schmitz, Head of Energy Research at SFOE, Isabelle Bey, MeteoSwiss, presented in her keynote lecture the big difference of “Climate Change: 1.5°C versus 2°C”. The second keynote of the day was from Christian Cajochen, University of Basel, on the “non-visual impact of light on human sleep and circadian physiology”.
Work presented from NRP 75 project “HyEnergy”
Roberto Castello, one of the NRP 75 researchers, spoke about “Deep Learning in the built environment: automatic detection of rooftop solar panels using Convolutional Neural Networks”. Mapping the location and size of solar installations in urban areas can be a valuable input for policymakers and for investing in distributed energy infrastructures. Machine Learning techniques, combined with satellite and aerial imagery, allow to overcome the limitations of surveys and sparse databases in providing this mapping at large scale.
In his talk Roberto Castello presented how they applied a supervised method based on convolutional neural networks to delineate rooftop solar panels and to detect their sizes by means of pixel-wise image segmentation. As input to the algorithm, they relied on high resolution aerial photos provided by the Swiss Federal Office of Topography. Different data augmentation was explored and network parameters were varied in order to maximize model performance. Preliminary results show that they were able to automatically detect in test images the area of a set of solar panels at pixel level with an accuracy of about 0.94 and an Intersection over Union index of up to 0.64. This work lays the foundation for predicting the existing solar panels deployment at the Swiss national scale. The correlation with local environmental and socio-economic variables would allow to extract predictive models to foster future adoption of solar technology in urban areas. The developed framework could be adapted in the context of the HyEnergy project and being trained to directly segment rooftop PV available area to provide a highly realistic estimate of the solar PV technical potential at the Swiss national scale.
What is the solar rooftop photovoltaic potential in Switzerland? A critical comparison of the methods
Alina Walch, the other researcher presenting NRP 75 work, introduced her “critical comparison of methods to estimate solar rooftop photovoltaic potential in Switzerland”. She compared six studies on rooftop photovoltaic potential in Switzerland carried out in recent years, in order to understand the opportunities and challenges of large shares of solar photovoltaics (PV) in Switzerland’s electricity mix. A qualitative comparison outlined different methodologies that have been applied in the studies, while a quantitative comparison showed how these methods can impact the potential estimate. They observed a strong trend towards increasing spatial and temporal resolutions, using larger datasets in better quality for the analysis. Data-driven estimations integrate Machine Learning with physical models and Geographic Information Systems to obtain an accurate large-scale potential estimate. Their analysis shows that the largest differences are caused by the source of the solar radiation input data, the computation of shading effects on rooftops and the estimation of available roof area for PV panel installation. The latter is the most uncertain parameter in the presented studies and offers opportunities for future work.
The presented results are relevant to researchers, policy makers and federal offices alike, as they highlight the importance of a critical interpretation of the numbers presented in different studies. The analysis also shows the impact that big data has on the estimation of renewable energy potential at large scales, and how Machine Learning methodologies can be effectively coupled with physical models to enhance their performance and accuracy.
Further talks in the sessions on data-driven methods were about a “combined geospatial and techno-economic analysis of deep building envelope retrofit”, “retrofitting a building stock: modelling and optimization for decision aiding at territory scale”, “cost-optimal retrofit analysis for residential buildings”, “statistical modelling of the energy reference area based on the Swiss building stock”, “using deep neural networks for predictive modelling of informal settlements in the context of flood risk”, a “morphological based PV generation and energy consumption predictive model for Singapore neighbourhood”, “machine learning and geographic information systems for large-scale wind energy potential estimation in rural areas”, “SCAFE: automated simultaneous clustering and non-linear feature extraction of building energy profiles”, and “data-driven short-term load forecasting for heating and cooling demand in office buildings”.
AI and buildings
The second day started with two keynote lectures, one from Arno Schlueter, ETH Zürich, on “Climate Changes: towards low-carbon, liveable cities”, and the other one from Charlotte Matthews, Sidewalk Labs, New York, on “District energy and the divergence from research into reality”.
After that, participants had the opportunity to join a workshop on the topic of “AI: Building Science 2.0” moderated by Roberto Castello. The fast pace of the digital transformation happening in our society generates a deluge of opportunities to increase productivity, efficiency and sustainability in many sectors. Building science helps architects during the design process and supports decisions for the improvement of building energy performance. The large amount of accessible digital data is a key asset to ensure that current and future urban developments are and will be in line with the objectives of countries’ energy strategy.
After an introduction recalling the basics of machine learning and the impact of artificial intelligence (AI) in different domains, from science to art and games, the speakers of the workshop followed one another to present their work. The examples proposed highlighted the role of Big Data and AI in optimizing the energy consumption of buildings and detecting structural weaknesses. This generated a lively debate about how open data can work as catalysts for AI in building science, on the importance of AI in data analytics and data structuring and, more in general, on the role of human supervision in AI-controlled processes.
New opportunities with open science
The third and final day was reserved for visits as well as a hands-on workshop on Open Science. This concept is rapidly gaining the adherence of both funding organisations like the SNFS and universities. It not only allows equal access to precious scientific information for all, but also offers new opportunities of collaboration and sharing knowledge and a much better use of already available data. The workshop was supported by the NRP 75 and SCCER projects and conducted by Roberto Castello and Dasaraden Mauree.
During the workshop, participants have been guided through a typical end-to-end workflow, from data collection to the publication of results. Starting from freely available datasets, they learned how to retrieve them from the Zenodo portal. After a short introduction on the Git language, the participants were able to retrieve a publicly available code from GitHub repository, to customize it for their own analysis and to push back their modifications back into the repository. Finally, it has been discussed how to ensure the analysis workflow reproducibility by means of Renku, a platform provided by the Swiss Data Science Center. The workshop aimed at letting participants experience a typical real data analysis, fully compliant with the FAIR principle.
A big thank you goes to the co-authors of this post: Alexandre Luyet, Responsible for Suisse romande at SATW, and the NRP 75 researchers Roberto Castello and Alina Walch.