Articles
Estimating Window-to-Wall Ratio for Urban Energy Modelling of Amsterdam centrum by Deep Learning
In this week’s blog post we are giving some insight into how AI can help in estimating the Window-to-Wall Ratio for the Urban Energy Modelling of the city. Availability of high-quality data for determining the feasibility of specific energy conservation measures on an urban scale is crucial but often insufficient. The Window-to-wall ratio (WWR), defined as the percentage area of a building's facade that is glazed, is one of the parameters affecting energy consumption of buildings but is currently lacking. Based on a deep learning approach using semantic segmentation to detect windows on panoramic images, the present project intends to enrich the current building database and to link WWR data to the design of energy-efficient retrofit strategies for Amsterdam city centre.
Sustainable heating
Amsterdam aims to be natural gas-free by 2040 and to reduce CO2 emissions from the built environment by 25%. The city centre has the highest heat demand density in Amsterdam: buildings are now mainly served with natural gas boilers and deep energy retrofits are difficult due to the monumental status of the buildings, designated as Unesco World Heritage sites. The city centre of Amsterdam also faces multiple challenges such as the vulnerability of heritage buildings to climate change, low-insulated building envelopes and poor thermal indoor comfort.
The objective of the research programme “High-hanging fruit: the sustainable heating of Amsterdam’s historic buildings”, initiated by AMS Institute and TU Delft (Chair Climate Design and Sustainability), is to define a generic approach for building energy retrofitting, thus identifying energy conservation measures (post-insulation, equipment upgrade) allowing for maximum impact while preserving the city’s historic and aesthetic values. The combinations of solutions should allow buildings to be heated at lower temperature (eg. with a supply temperature less than 55°C), improve indoor comfort and minimize environmental impact.
Building archetyping
The first part of the project consisted in characterizing the building stock of Amsterdam centrum into building archetypes based on function, construction year, typology and WWR. For each archetype, the goal is to define best-performing energy retrofitting strategies using energy simulation engines and parametric modelling (eg. Grasshopper and Ladybug Tools). Parametric tools are able to concentrate in a single workflow the energy simulation and the optimization of the building performance via a set of selected input parameters. From a very broad range of interventions, it is then possible to narrow down the number of options, and identify the best-balanced scenarios through multi-objective optimisation.
The combination of window insulation quality and WWR plays a role in space heating demand reduction and occupant comfort, thus is important for decision-making in selecting energy-savings measures. Figure 2 illustrates the influence of different WWR and the choice of different types of glazing on the potential heating savings.
By linking input data such as the WWR, it is possible to more accurately evaluate the impact of different retrofitting solutions, including potential energy savings.
Gemeente Amsterdam provided a dataset that covers the front facades of 48% of the buildings in the city centre of Amsterdam (in total about 7,700 buildings). By extracting façade textures from panoramic images (Figure 3), it is possible to identify, rectify and extract the texture region of a building and link these to building footprint data. More details on the methodology are available in the post “A deep learning approach to enhance 3D city models” by Chris Eijgenstein.
An example of classification of buildings built before 1945 based on typology and WWR is shown below (Figure 4).
The final results of the spatial distribution of different WWR in the city center of Amsterdam is displayed on Figure 5. Most buildings have a front facade with a WWR ranging between 33 to 57%.
To improve the coverage of the dataset (Figure 6), several options could be further investigated:
- using drones to collect images from facades which are non-visible from the street or distorted (eg. in narrow streets panoramic images of the top of buildings may be stretched);
- running the system on panoramic images on datasets from other years;
- using AI to predict facade elements hidden behind an obstacle (eg. trees, trucks, street signs,..).
Future Prospects
Based on the classification of the buildings, it will be possible to assess the “retrofittability” of each archetype and to link it to a suitable retrofit package. The next article will present how the parametric modelling tools Ladybug tools for Grasshopper (tutorials are available here https://docs.ladybug.tools/ladybug-tools-academy/) can be used to compute all possible retrofit scenarios for each archetype and evaluate the associated impact in terms of energy performance and thermal comfort.
Related Resources
AI not only stands for Artificial Intelligence but also for “Amsterdam Intelligence”. With this website we wish to share our knowledge, process and results. Check our blog posts and projects to keep up to date. Feel free to get in touch.
