Hourly PV potential profiles for individual roof surfaces: A large-scale approach

This repository represents the data and codes that were used to create a dataset of hourly PV potentials for 9.6M roof surfaces in Switzerland. The computed potentials account for:

• The spatio-temporal variation of the solar radiation
• The effects of surrounding trees and buildings on roof shading and the sky view factor
• The impact of roof geometry and roof superstructures on the available area for installing PV panels
• The temperature-dependence of the PV module efficiency
• The spatio-temporal patterns of uncertainty of the estimated technical rooftop PV potential

The method includes a combination of GIS processing tools (grassGIS and geopandas), Machine Learning algorithms (Random Forest and Extreme Learning Machine Ensemble) and physical models (*Perez* model for tilted radiation, *PVWatts* model for module and inverter efficiency).

As a proof of concept, the dataset included here contains only 1000 example roofs. This means that some of the code may appear as "overkill", because it was designed to operate on datasets containing millions of roofs. As some of the sources are under a research license, the raw data is not included in this repository. However, the processed data is provided. Please note that in some cases small processing steps may be necessary to move from one stage to the next.

If you are interested in the full hourly PV potential dataset, please do not hesitate to get in touch (email in "References").

Processing steps

Data

In the example data, timestamps are given as the 15th of each month for the year 2001. This is merely a convention to indicate the mean hourly profiles for each month, in a non-leap year. The data represents the 12-year average hourly profiles of solar radiation for each month, based on the data from 2004-2015.

Prerequisites

The code is based almost entirely on python scripts (using python 3.6.5), with exception of the shading and SVF analysis, which makes use of the GRASS GIS bash API (v7.4). The following python libraries are heavily used in the present scripts:

• pandas
• xarray
• pvlib
• scipy
• matplotlib

In the *libs*-folder, several scripts performing background functions are provided. It is recommended to add this folder to the python path in order to run the codes provided. The script *util.py* in *libs* provides several utility functions and is a general prerequisite to run the provided scripts.

References

This work was performed by Alina Walch (alina.walch@epfl.ch) as part of the National Research Program 75 (Big Data) for the project "HyEnergy" (https://www.epfl.ch/labs/leso/research/domains/urbanenergysystems/hyenergy/)

The rooftop data was provided by the Swiss Federal Office for Energy and the Swiss Federal Office for Topography.