MESCO: a Clustering framework for the design Optimization of future Multi-Energy Systems

Clustering techniques are the standard to identify representative days of annual trends of energy demand, prices and climatic conditions in Multi-Energy Systems (MES) design. However, the literature lacks guidelines for clustering techniques leading to the ‘best’ design solution of a MES, usually neglecting a complete testing phase on multi-year datasets. This paper presents MESCO, a MES Clustering-Optimization framework to (1) generate representative days using different clustering algorithms (k-means, substitution, k-medoids) and extreme days criteria (null, replacing, adding, iterative); (2) validate the clustering-based design solutions on a ‘past’ dataset (2010-2018) and assess their robustness against two ‘future’ scenarios (Covid-19 pandemic, 2019-2020; Russia-Ukraine war, 2021-2022). K m e a n s − i t e r a t i v e clustering-based solutions with 7–9 representative days lead to the lowest relative error in total cost compared to perfect knowledge design solutions based on full time series, with errors of + 4 % and + 25 % for 2019-2020 and 2021-2022 scenarios, respectively. While results in other cases may differ, the application of the proposed general framework remains effective in evaluating the accuracy of different clustering algorithms and extreme day criteria in MES design. • MESCO: a novel clustering-optimization framework for MES design under uncertainty. • Clustering-based design solutions tested vs those based on multi-year time series. • Guidelines for selecting optimal clustering algorithms and extreme day criteria. • K-means algorithm has higher accuracy and stability than k-medoids and substitution. • Iterative criterion minimizes unmet demand and cost in disruptive future scenarios.