Spatial clustering in energy system modelling: application to the case study of District heating

Given the strong decentralization of renewables and the raising interest in energy networks (e.g. heat, gas, hydrogen, etc.), spatial detail has become an important feature that energy system models have to take into account. However, a trade-off is needed between systems representation quality and models computational burden, also considering the temporal and technological resolution. In this context, clustering is a useful tool that can help reducing the complexity of energy system models maintaining at the same time a high quality of systems representation. The objective of this paper is to delve into the decision-making process behind the selection of parameters for clustering algorithms applied to energy system model data. This involves analysing the requirements of the algorithms and linking and adapting the clustering parameters to the specific physical quantities of these models. Furthermore, it aims to assist modelers in reading and interpreting the results obtained. As a case study, the Density-Based Spatial Clustering Algorithm with Noise (DBSCAN) algorithm is applied to six Italian Municipalities with the aim of aggregating the possible input data of an Energy System Optimization Model (ESOM). A parametric analysis is then performed to assess the behaviour of clustering under different conditions and to assess technical-economical characteristics of the resulting aggregations. Results show that relationships between input data and DBSCAN behaviour (such as the relationship between the average distance between elements and the value of the parameter ϵ that returns the maximum number of clusters) might subsist for spatial domains where the elements to aggregate are homogeneous. Since the optimal choice of the clustering parameters depends on the modelling requirements, the paper finally proposes four different cases a modeller could focus on, specifying the most suitable parameters for each case.