Challenge provided by Ghent Municipality

Energy communities inclusive of residents vulnerable to energy poverty

Encouraging residents to join energy communities can have environmental and financial benefits, including reduced energy consumption and carbon emissions. Previous analysis estimated potential cost reductions of 10-26% and emission reductions of 5-13% in Belgium.

Ghent is the second largest city in Flanders, Belgium, with a population of approximately 260’000. The City of Ghent has huge ambitions in the fields of climate mitigation and adaptation, including making clean and renewable energy accessible to all its citizens.

Energy communities are an effective way to allocate and distribute energy equitably. Recently the EU launched the Energy Communities Repository, defining different types of energy communities with the core tenant of their primary purpose being to generate social and environmental benefits rather than financial profits.

The city of Ghent has collected an extraordinary dataset of its solar potential using laser measurements during several flights in 2013, creating a point cloud from which their data team calculated a solar potential map spanning the whole city.

In combination with data on energy consumption, existing local energy production, and demographic data, Ghent is looking to use this dataset to create energy communities in Ghent with a special focus on including residents vulnerable to energy poverty.

Figure 1 - A small section of the solar potential map of Ghent is provided on their website.


Leverage the existing data to propose how energy communities could be formed in the city of Ghent with a special focus on including residents vulnerable to energy poverty.

United Nations SDG 

GOAL 7: Affordable and clean energy

  • Target 7.1: Ensure universal access to affordable, reliable and modern energy services
  • Target 7.2: Increase substantially the share of renewable energy in the global energy mix


The following datasets were provided to the participants:

  • Geographical data on sun irradiation on a 3D model of Ghent which can be explored on the city of Ghent’s website.
  • The energy company data on energy consumption in Ghent.The energy company data on production, storage devices, and EV charging points.
  • Ghent average income per sector.
  • Current rent prices in different locations in Ghent.


This year’s final challenge came with a comprehensive dataset about the sun irradiation and solar energy potential on all roofs in the city of Ghent, derived from a 3D model based on a point cloud gathered from 16 points per m2 point cloud.

One team decided to augment that dataset with data on hours of sunshine in the city of Ghent, as well as installation costs of solar panels and energy tariffs in Belgium.

Another team discovered the recently published Belgian Index of Multiple Deprivation (BIMD) by Otavova et al (2023) and incorporated it into their model alongside open street data and demographic information provided by the city of Ghent.

Methods and Techniques

The city of Ghent posed their challenge as an open problem, looking to find ideas and ways In order to define their data product, one team defined their own key performance indicators (KPI), such as annualized profitability for each household in the energy community, start-up capital needed and geographical proximity of members and combined them into one outcome measurement.

Several teams (1,2) achieved clustering using K-means, which is well suited to a problem wanting to minimize the distance between cluster members. One team derived a custom iterative tiered optimization algorithm the team developed. The models were compared to a random assignment baseline.

Another team in additional to K-means tried spatial optimization algorithms with queen contiguous weights and max-p regionalization, Spectral clustering HDBSCAN as well as reinforcement learning arriving at a combined method with rule application before clustering using K-means for their MVP (minimal viable product) in order to optimize for less compute time needed.

Main Insights from Data

One team showed that their predicted energy communities would on average save each member of an energy community 1000 euro per year. They also pointed out that higher returns often are connected to higher upfront costs, meaning that to serve households at risk of energy poverty a tradeoff has to be considered.

Figure 2 - Plots of clustering results on a subset of the data. The left plot shows the results of K-means clustering scoring at 2621.7 in the team combined KPI metric. The plot on the right side shows the proposed energy communities cluster created by their iterative algorithm, scoring 3398.5 in their combined KPI metric, both models meaningfully beat the random assignment baseline.

Another team plotted the area with the highest solar potential, showing it to be highest around the city center, where individuals with higher risk to energy poverty were grouped as well (Figure 3).

Figure 3 - Roofs with the highest solar potential as black dots, ranked income domains as colored areas (red being the lowest income on average)


One product proposed was a web app open to the public that can give recommendations to the inhabitants of Ghent with whom and how to start an energy community in their neighborhood. The app highlights both potential financial gain, as well as a reduction in carbon emission per household (Figure 4).

Figure 4 - The web app of one team shows the public the benefits of an energy community

Another team developed their own open source Python package to be used to optimize the formation of energy communities. In addition they created a dashboard for non-technical users to explore energy communities suggested based on user-defined inputs.

Figure 5 - A second team approach to a dashboard showing different energy communities based on user-defined values.

A third team applied a K-means algorithm to the whole dataset solving for the problem of optimal energy communities if all individuals living in Ghent should be assigned to one (Figure 6).

Figure 6 - Plot of K-means clustering including all households with the ideal K of 8 determined with the Elbow method.

Social Impact

When analysing the potential impact of their solutions several teams noted that the challenge proposed is as much a technical as a social one. Helping residents see and understand the environmental and financial benefits of being part of an energy community is key and secondary benefits like social cohesion and empowerment of citizens

With data gathering already in place concrete impact metrics can be reduction in energy consumed as well as reduction of carbon emissions. Previous analysis by Felice et al (2022) on the potential for energy communities in Belgium estimated a potential cost-reduction of 10-26% and emission reductions of 5-13%, with the most important challenge being to reduce the barrier to entry for citizens, which was one of the core aspects addressed by proposed solutions.

Open-source code

Other challenges