4 Neighborhood zoning and typification
For the CEL to be implemented, it is important to first set a delineation of the community. For SIG, what is at stake is to propose a coherent and generic definition of a neighborhood that would facilitate the deployment of CEL. It is indeed important to know which building and units are to be part of a CEL that is about to be settled. Therefore having generic perimeter for CEL for the whole canton is very valuable for an industrial entity like SIG. Besides, there are also other benefits of neighborhood zoning and typification for SIG such as proposing coherent energy concept for a neighborhoods, comparing the CO2 intensity of each neighborhood in order to identify which neighbors to prioritize for renovation. Typification allows to extract general trends on energy and resources consumption, energy harvesting over the territory of Geneva. However, the implementation of energetic concept needs to take into account the local characteristics of each neighborhood.
4.1 Neighborhood zoning
The aim is to propose a suitable neighborhood zoning of the entire Geneva’s canton to have a proper framework for the implementation of CEL in the canton. There are different ways of tackling this problem.
First, the question of the neighborhood zoning could be adressed through a technical prism which consists of considering only the technical possibilities of neighborhoods : how a neighborhood could be supplied in electricity and thermal energy and form a neighborhood on the basis of these characteristics. Second, one could consider this question through the lens of the public administration of the territory. Indeed, establishing CEL implies a technical knowledge but also a legal and organizational framework on which every stakeholders involved in the deployment of CEL need to agree.
Since this problem mixes a technical and a social dimension and involves different stakeholders it is important to state that the problem is case-dependent on the territory and the administration that rules it. Here, pros and cons of different approaches for tackling this issue are presented for the specific case of the Canton of Geneva.
4.1.1 Technical dimension
Here, the focus is put on technical characteristics of each possible neighborhood in order to propose a complete zoning (in the sense that it covers the entire territory) of the canton. The energy demand of buildings being split between the thermal needs and the electricity needs, two zonings are hereafter proposed that are based on the supply of thermal and electrical energy.
4.1.1.1 Thermal neighborhoods
As most of the buildings energy needs are thermal needs, it makes sense to establish groupings of buildings in terms of their thermal demand : a common strategy for buildings inside a thermal neighborhood to fulfill their thermal needs efficiently can thus be determined easily. CEL could be implemented at the level of these neighborhoods with the aim of creating synergies between electrical and thermal resources.
Pros:
- This is especially interesting when the neighborhood can be connected to a thermal network like RTS (Réseau Thermique Structurant, meaning structuring thermal network) in Geneva or if there is a renewable thermal source that could be exploited locally.
- The unit that operates thermal networks at SIG works already with a zoning based on the demand profile of buildings inside a neighborhood
Cons:
- This is not really relevant when considering the electrification of the thermal demand to come.
- The granularity of the zoning established by the unit responsible of thermal networks at SIG is quite small, making planning for energetic supply at this neighborhood level and thus CEL less relevant. Moreover, this zoning does not cover the whole territory of Geneva’s canton.
4.1.1.2 Electrical neighborhoods
A more relevant aspect to consider is the electrical demand and means of supply of buildings for the conception of a zoning as CEL are especially about exchanging electricity locally. Establishing grouping based on means of supply of each building, therefore by grouping buildings that are in a same Low Voltage (LV) network, makes sense as in such a disposition, only the LV network would be used for exchanging electricity within the CEL.
Pros:
- It would simplify the accounting of the energetic flux between buildings and the flux between the CEL and the rest of the grid.
- Buildings inside that kind of neighborhood are directly connected to one another. Thus constraints on grid connection are easily manageable.
Cons:
- The LV network is meshed is such a way that buildings are linked to multiple (MV/LV) transformers. This makes the delimitation of CEL more complex. From the zoning in electrical neighborhood established by the DSO of SIG, some buildings are even sometimes split in two where one part is in one electrical neighborhood and the other part in another.
- Moreover, since the network is meshed, it is also dynamic : Buildings can be supplied by a certain MV/LV transformer at a point \(t\) in time, but supplied by another transformer later on at time \(t'\). Thus electric neighborhoods are not the same at time \(t\) and at time \(t'\).
- Does not reflect the administrative architecture of the organization of the territory
4.1.2 Public administration
In this section the emphasis is put on social dimensions that infer a zoning of the territory. It is believed that being attentive to the administrative organization of the territory is particularly important in order to implement rapidly and efficiently new consumption strategies like CEL would leverage. To this extend, a first step was to analyze if urban layers used in the conception and construction phase of a new neighborhood (here neighborhood is to be understood as an ensemble of new building to construct) : the layers PLQ (Plan Localisé de Quartier) and CET (Concept Énergétique Territorial). A second step was to look at any administrative layers of the information system of Geneva’s territory (SITG).
4.1.2.1 Urban data
When a new neighborhood, in the sense of an ensemble of new buildings, is built a local development plans, PLQ in French and a energetic concept for the territory, CET in French have to be emitted.
The PLQs are special or detailed land allocation plans that are legally binding on third parties, specifying the conditions for the realization of new constructions. They cover aspects such as the volume (number of floors, footprint) and allocation of each proposed building, access points, parking, land use, requested easements and transfers, etc. They are primarily composed of a plan and a set of regulations.
The CETs aim to propose strategies for valuing local resources and supply strategies based on an assessment of resources, needs, stakeholders, and infrastructure. The objective is to meet the short and long-term needs of the relevant area in line with the non-nuclear energy policy objectives as a long-term vision.
It was thus natural to assess if these urban data could help designing a zoning of the territory.
Pros:
- These urban plans tackle technical aspects that are important for designing an optimal energy system for the neighborhood. It could be a great help for the implementation of CEL if the delimitation of the community was fixed by these urban plans
Cons
- None of these two plans covers the entire territory if the canton
An overview of these two layers are given in Annex [METTRE EN ANNEX]
4.1.2.2 Administrative data
In order to get data about the administrative architecture of the territory, one search on the SITG catalog of geographical data of the Canton. The search have been done with keywords in order to filter out only layers relative to political and administrative limits : In the catalog query research interface, under the section Type of data (type de donnée) the keyword polygone was selected to get perimeters type of data; under the section ISO theme (thème iso) the keyword limites politiques et administratives was selected.
Within the proposed list, layers have been even more filtered out using two criterion :
- That the proposed layer covers the entire territory
- That it proposed small enough neighborhoods in order to have a realizable centralized strategy for neighborhood energy production facilities but big enough to make it relevant.
finally it appears that only one layer fulfilled all these criterion : the layer of statistical sub-sector of GIREC (Groupe Interdépartemental de Représentation Cartographique). According to SITG, statistical sub-sectors cover a territorial entity at an intermediate scale between the building and the commune. They allow for distinguishing, within urban areas, different neighborhoods, or within rural areas, villages and hamlets from agricultural or forested areas.
Pros:
- The use of this zoning is widespread both within the cantonal administration, municipal administrations, and specialized planning offices, as well as at the University and other educational institutions. Moreover, federal administration services or those of other cantons use it for national or regional level comparisons. (“Le découpage Du Canton de Genève En Sous-Secteurs Statistiques. Révision 2005,” n.d.)
- The DSO of SIG uses also this layer for the medium voltage (MV) network : the transformer that supplies each sector is known which is a advantage for implementing CEL. Besides it is an asset that the DSO uses already this layer, since it would ease its collaborative work with the other stakeholders.
- The unit that operates thermal networks at SIG use a layer which is a based on this layer. Similarly this is also an asset since it would ease the collaborative work between every stakeholder.
Cons:
- This zoning is not directly correlated to the consumption of electricity nor heat and does not take into account energy harvesting potentials (such as PV potental for instance).
4.2 Typification
To go even further, the typification of the neighborhood provide valuable additional information on the energetic composition of the canton. Establishing a typification based on a layer identified as relevant can thus be a great interest for SIG in order to propose a coherent energetic concept and business model for a neighborhood. Even if implementing a specific energy concept to a neighborhood is case-dependent, typification gives valuable macroscopic information.
The typification has been done using the clustering algorithm developed by Loustau (Loustau et al. 2023). The clustering features considered are the neighborhood energetic characteristics: the energy demands on one side (electrical, heating and domestic hot water demand), and the endogenous resources on the other side (solar, distance to a thermal network RTS, and biomass). Two clustering algorithms are investigated in Loustau’s work, a centroid-based (Kmedoids) and a density-based (Gaussian Mixture). However here only the Gaussian Mixture Model (GMM) approach has been conducted as it appeared to be the best clustering algorithm in Loustau’s work.
Prior clustering, data are normalized and the ideal number of clusters, as well as the shape of the covariance matrix, have to be estimated. This is done by runing multiple times the algorithm and by assessing the performance of each run with scores depending on the number of parameters included in the model. In this study, the number of iterations made to find the optimal number of clusters is 500.
The GMM summarizes a multivariate probability density function with a mixture of Gaussian probability distributions. For \(M\) Gaussian components, the probability densities are given in equation (4.1) :
\[\begin{equation} p(\textbf{x}|\lambda)=\sum_{i=1}^{M}\omega_i \cdot g(\textbf{x}|\mu_i, \Sigma_i) \tag{4.1} \end{equation}\]
Where :
- \(\textbf{x}\) is a D-dimensional data-vector representing the selected features of a neighborhood;
- \((\omega_i)_{i=1}^M\) is the set of mixture weights, satisfying \(\Sigma_{i=1}^M \omega_i=1\)
- \(g(\textbf{x}|\mu_i, \Sigma_i)\) is the probability density of the ith Gaussian component. Each component is a Gaussian function given by equation (4.2) :
\[\begin{equation} g(\textbf{x}|\mu_i, \Sigma_i) = \dfrac{1}{(2\pi)^{D/2}|\Sigma_i|^{1/2}}\exp{(-\frac{1}{2}(\textbf{x}-\mu_i)^t\Sigma_i^{-1}(\textbf{x}-\mu_i))} \tag{4.2} \end{equation}\]
with \(\mu_i\) the D-dimensional mean vector and \(\Sigma_i\) the DxD-dimensional covariance matrix. The covariance matrix can be full rank or diagonal with common a element to all the components or not. \(\lambda\) represents the parameters of the GMM : \(\lambda = ((\omega_i)_{i=1}^M, (\mu_i)_{i=1}^M, (\Sigma_i)_{i=1}^M)\).
Because the GaussianMixture relies on probabilities, the best clustering is the one that maximises the likelihood. The likelihood function evaluates the joint probability of observed data as a function of the chosen statistical model. Given a set of \(N\) training vectors \(\textbf{X}\), the GMM likelihood function \(L(\lambda)\) can be written as equation (4.3):
\[\begin{equation} L(\lambda) = p(X|\lambda) = \prod_{j=1}^N p(x_j|\lambda) \tag{4.3} \end{equation}\]
However, adding components helps increase the likelihood while it may lead to overfitting the data. Criteria introduce penalty terms on the number of parameters to solve the issue :
- Akaike Information Criterion (AIC) : defined by equation (4.4), it should be as small as possible :
\[\begin{equation} AIC = 2\cdot\ln{(k)} - 2\cdot\ln{(\hat{L})} \tag{4.4} \end{equation}\]
where \(k\) is the number of parameters in the model and \(\hat{L}\) is the maximized likelihood of the GMM.
- Bayesian Information Criterion (BIC) : defined by equation (4.5), it should be as well as small as possible. It has consistency (meaning it would asymptotically select the candidate model having the correct structure), as its penalty term contains \(N\).
\[\begin{equation} BIC = k\cdot\ln{(N)} - 2\cdot\ln{(\hat{L})} \tag{4.5} \end{equation}\]
For each of the 500 iterations, both scores are computed for \(M=2,...30\), thus 29 times. For each iteration the optimal number of Gaussian components is thus selected by minimizing the AIC and BIC score. If one of the scores doesn’t have a minimum on the range \(M=2,...30\), the optimal number of components is obtain at the elbow of the score curve. Therefore, after the 500 iteration the overall optimal number of components is the optimal number of component that occurred the most across the 500 iterations.