3.1.3.3. Iterative process for calculating the additional capacity need

B.3. The ‘mid-term flow-based’ modelling framework used in this study

As described in the previous section, the flow-based capacity calculation is a complex process involving many stakeholders and many parameters. To build market models where market exchanges adhere to the rules depicted in a flow-based coupled market, multiple approaches are possible. For short term forecasts and analyses, a framework relying on the flow-based domains conceived in the SPAIC process was developed [JAO-1]. This framework however leans heavily on historical data. As historical domains are strongly related to the historical grid & generation situation this approach is not suited for studies on a longer time horizon where significant evolutions on the grid and generation mix occur.

Elia has developed a mid-term flow-based framework which does not rely on historical domains, but instead aims to mimick the operational flow-based capacity calculation workflow, for which the required inputs are forecasted for the targeted time horizons.

B.3.1. CALCULATION OF PTDFS

The first step of the mid-term flow-based framework is the definition of a set of PTDFs2. To obtain those, a European grid model is built, which is for this study based on the TYNDP 2018 reference grid, upon which grid modifications for Belgium are applied at the different target time horizons. This grid model is then used to calculate the PTDFs.

A PTDF matrix consists of lines/rows representing the different CNEC’s that are taken into account, and columns representing the variables in the flow-based domain. y Each CNEC refers to the combination of a Critical Network Element and a Contingency. In the grid model that was used for this study, many hundreds of CNECs were considered; y The variables can represent the net positions of the grid nodes under consideration, the HVDC3 flows, PST positions, etc; depending on the degrees of freedom that are given to the market coupling algorithm.

Aside from a PTDF matrix, the flow-based mid-term framework also requires the capacity of each Critical Network Element. These correspond to the steady-state seasonal ratings of the network elements. B.3.2. CALCULATING THE INITIAL LOADING OF

EACH CNEC

For this study, to be in line with current market operations, only CWE is modelled as a flow-based region. The variables are the net positions of the countries (BE, DE (and LU), NL, FR, AT) toward CWE. Flows outside of CWE are subject to NTC constraints, and the interaction between the flowbased region and flows on external borders to CWE are modelled using standard hybrid coupling. Only cross-border (XB) CNECs are considered for the 5 CWE countries. ALEGrO is modelled using ‘evolved flow-based’, introducing a 6th variable in the PTDF matrix.

As described in Section 2.7.2, once fully set up, the midterm flow-based framework first performs a market dispatch simulation to determine the initial loading of each CNEC. In this simulation, 2/3rd of the PST tap ranges are allowed to be used to optimize initial flows in order to maximize welfare of the system. The flows from this simulation determine the “Reference Flows (Fref)” (see Section, ). These flows are then scaled back to zero-balance flows “F0” per Bidding Zone through the use of GSKs. This procedure mimicks the CACM CC process and allows for a good estimation of the pre-loading on CNECs.

B.3.3. CALCULATING THE FB CAPACITY DOMAIN

European legislation requires minimum margins to be made available to the market. For this reason, every time a CNEC’s margin after preloading is less than the required minimum margin given to the market, the minimum margin is guaranteed (see also Section 2.7.2).

B.3.4. CLUSTERING OF DOMAINS

In this study, a series of climatic years is used to model variability in climatic variables such as renewable generation, electricity demand… The use of hourly domains for the market simulations is however not deemed computationally efficient. The calculated hourly domains were therefore clustered into groups, identifying one representative domain per group. Furthermore, the relationship between each group and the climate conditions was analysed in order to map them onto the model. This approach is line with what is done in the strategic reserve volume determination assessments.

2. A PTDF coefficient for a CNEC & zone represents the change in flow on the

CNEC related to the change in net position of the zone (see Section ). 3. An HVDC link is a controllable device by nature. Power electronics allow for completely control the flow on the link, therefore not making it subject to Kirchhoff laws. For this study, and after analyzing multiple combinations of pre-clustering data split (seasonal split, day type split, …) no clear trends were identified therefore the decision was made not to apply any pre-clustering data split, but to cluster the entire set of 8760 domains as a whole. Indeed, no clear advantage in the distinction of domains using pre-cluster splits was found, as higher winter initial flows are offset by higher steady-state seasonal ratings of the network elements. Each flow-based domain is a 6 dimensional shape, one dimension for each of the 6 variables. The clustering of the 8760 domains is based on their geometrical shape. For this it is important to define a good distance metric between domains. Next, one needs to define the number of clusters to retain. For this study, all domains were clustered into three groups. An advantage of choosing a low number of clusters is that many domains are present in each of the clusters, therefore reinforcing the stability of the chosen medoid. After defining the number of groups, a representative domain per group is chosen. This is done by means of a k-medoid algorithm. Here the medoids are elements which are part of the initial domains, and therefore have physical meaning.

The quality of the clustering can be visually observed by plotting all domains for each cluster, as well as the centroid, as is shown in For each of the clusters a correlation analysis with climatic variables (in this case CWE load & wind) was performed. Based on this analysis French load & German wind infeed were identified as the main axes to which the clusters can be correlated. for 2025.

THE GROUPING OF DOMAINS PER CENTROID SHOWS THE QUALITY OF THE CLUSTERING ALGORITHM (EXAMPLE SHOWN FOR 2025) [FIGURE 6-7]

FR balance to CWE [GW] 12 10 8 6 4 2 0 -2 -4 -6 -8 -10 -12 Cluster 1 Cluster 2 Cluster 3 12 10 8 6 4 2 0 -2 -4 -6 -8 -10 -12 12 10 8 6 4 2 0 -2 -4 -6 -8 -10 -12

-10 -8 -6 -4 -2 0 2 4 6 8 10 -10 -8 -6 -4 -2 0 2 4 6 8 10 -10 -8 -6 -4 -2 0 2 4 6 8 10 BE balance to CWE [GW] Belgian simultaneous import capacity (including Nemo Link): 7500 MW All domains in cluster Cluster representative domain used in the study

For each of the clusters a correlation analysis with climatic variables (in this case CWE load & wind) was performed. Based on this analysis French load & German wind infeed were identified as the main axes to which the clusters can be correlated.

EXAMPLE OF THE CORRELATION BETWEEN GERMAN WIND AND CLUSTER FOR THE CLUSTERS USED IN THIS STUDY ON THE 2025 HORIZON [FIGURE 6-8]

German wind [GW] 50 45 40 35 30 25 20 15 10 5 0

Cluster 1 Cluster 2 Cluster 3 Figure 6-8 shows the occurrence of each cluster for different German wind conditions. A few observations can be made from the figure: : y For a range of wind infeed in Germany between 10

GW and 25 GW (y-axis), a significant overlap is observed between the three different clusters. This means that within this range, each of the three clusters can be associated to the given level of wind infeed in Germany, each still with a different probability of occurence. Therefore a fully deterministic linking of a each cluster to a certain threshold of German wind infeed is not possible. The same is generally true for the French load correlation. y The second observation is however that the size of the boxplots above is not equal indicating e.g. that cluster 1 will have a relative higher probability of apparence in the high range of wind infeed (~15 GW-23 GW), whereas cluster 3 will have a relative higher probability of appearence in the range of low wind infeed, below 5 GW.

Therefore, both climatic variables wind infeed in Germany and French load were split into three groups (threshold defined by the 33% and 66% percentiles) and for each of the possible nine combinations a probability of finding cluster 1, 2 or 3 is calculated. These probabilities for each of the nine combinations are shown in Figure 2-55.