Robust Clustering Methods for 5GDHC Networks in Densely Populated Urban Areas

Abstract

Fifth generation district heating and cooling (5GDHC) networks enable the integration of low temperature sources and the utilization of synergies between heating and cooling demands. To implement them in urban areas a modular approach is suitable which implies the partition of the respective area into smaller cluster networks. This study addresses the effects of participation uncertainty in potential networks and the lack of comparison of clustering methods in the context of 5GDHC networks. Within this work the performance of different clustering methods is measured by optimization metrics that capture the requirements for a suitable partition, such as compact and balanced clusters with sufficient energy supply. This enables the comparison of the selected methods K-Means, K-Medians, DBSCAN and Single Linkage Clustering (SLC). To investigate the robustness with respect to participation uncertainty, a scenario-based approach that combines the minimax regret method with a cluster viability analysis is developed. The developed methodology is applied to the city centre of Amsterdam. The findings indicate that the participation rate influences the viability of the cluster networks. Nevertheless, K-Means, K-Medians and SLC are robust in their performance measured by the averages of the optimization metrics. Additionally, K-Means leads to the lowest variance in performance and is the best performing and most robust method in terms of average cluster compactness and demand fulfilment. Moreover, promising starting clusters for the implementation in the city centre of Amsterdam are identified. The study gives insights into the behaviour of the tested clustering methods and how they may be an advantageous alternative to state-of-the-art planning tools. Finally, the proposed approach to analyse the robustness of clustering methods contributes to the planning of district heating and cooling networks.