Team @EPFL: Prof. Jean-Louis Scartezzini, Dr Roberto Castello, Alina Walch
Team @UNIL: Prof. Mikhail Kanevski, Fabian Guignard
External project supervisor: Dr Nahid Mohajeri (postdoc mobility fellow at University of Oxford)
Funding: Swiss National Science Foundation – Nation Research Program 75 “Big Data”
Project duration: 2017-2020
The main objective of this project is to develop a methodology for forecasting the spatio-temporal potential of a combination of renewable energy resources for built areas. Using the available environmental and urban data together with the most advanced data learning techniques, the project aims at:
- Estimating the hybrid renewable energy potential in the built environment, in order to mitigate the effects of variability in individual energy resources and improve the reliability of power generation
- Developing machine learning algorithms for spatio-temporal environmental data processing and analysis as well as for urban and buildings dataset classification
- Applying the developed algorithms to the built environment for predicting energy generation and potential energy savings of hybrid renewable resources
- Providing forecasting models according to the projected climate scenarios for 2035 and 2050
- Estimating uncertainty and validate models using measurement data from weather stations and energy providers
- Proposing a Building Renewable Energy Database (BRED), geo-visualisation tools and renewable energy mapping to support evidence-based decision-making processes.
Using Big Data and the most advanced learning technologies, our project will make possible to assess for the first time how much of the energy demand from buildings can be reduced throughout Switzerland. The results will guide and influence urban energy policies in the country and the developed generic data-driven methods can also be used for other countries. Finally, the Building Renewable Energy Database and the mapping tools will allow building owners, communities and investors to visualise energy savings and supply for individual buildings as well as for groups of buildings anywhere in Switzerland.
The activity of EPFL team is focused on the assessment of renewable energy generation potential for the built environment, with particular focus of rooftop solar photovoltaic (PV) and solar thermal (STC) energy generation. This work includes the estimation of solar horizontal irradiance at high spatial and temporal resolution, as well as the quantification of urban factors impacting PV and STC potential such as shading and available rooftop area. For the analysis, data driven and Machine Learning approaches (in particular Random Forests and Extreme Learning Machines) are combined with GIS tools to extract information from various sources including LiDAR-based surface models, 3D building data, satellite imagery and meteorological weather stations.
Near-future work includes the combination of the estimated PV potential at hourly resolution with additional renewable resources such as wind, to assess the potential of a hybrid renewable energy potential across Switzerland. Parallel activities include the forecast of future development of PV potential based on projected population statistics and climate scenarios, as well as a comprehensive assessment of the currently installed PV and STC capacity in Switzerland using the most advanced Convolution Neural Networks and semantic image segmentation techniques on aerial photo maps of Switzerland.
The activity of UNIL team is focused on the development and application of analytical tools for the spatio-temporal analysis and visualization of environmental multivariate data using cutting-edge approaches in (geo)statistics and data mining. In particular, we are extensively exploring high frequency wind speed data at local and global scales using both linear and non-linear tools such as spatio-temporal variography, Fisher-Shannon complexity, wavelet variance, random fractals, temporal point processes, Alan factor, and complex network science. At present, we are also working on fundamental methodological developments for the quantification of uncertainties using Extreme Learning Machines.
Nearest future work will involve geostatistical spatio-temporal predictions and simulations, as well as adaptation and application of advanced machine learning algorithms for deep exploration and forecasting of environmental data relevant to the renewable energy potential assessment.