Multi-Energy Systems (MES) offer a potential solution to the challenges faced by today’s energy sectors, as they balance economic, environmental, and technical needs. This thesis presents a comparative numerical study of five multi-objective optimization algorithms applied to real-world MES case studies simulated with EnergyPLAN and optimized through a Python black-box model using the Pymoo library. This approach allows flexible exploration of complex decision spaces without requiring explicit reformulations of the simulator. The evaluated algorithms include two Pareto-based methods (NSGA-II, SPEA2), one indicator-based approach (IBEA), and two variants of Multi-Objective Particle Swarm Optimization (MOPSO with crowding distance and with adaptive grid). The case studies analyzed Aalborg, CEIS, and Giudicarie, which vary in decision-variable dimensionality, constraints, and the number of objectives. Algorithm performance is assessed using six complementary indicators (Hypervolume, Additive Epsilon, Inverted Generational Distance, Overall Spread, Spacing, and Ratio of Non-Dominated Solutions), combined through a weighted scoring system and validated with non-parametric Friedman and Nemenyi statistical tests. Results show that MOPSO methods, particularly the adaptive‑grid variant, achieve the strongest performance in the Aalborg and Giudicarie cases. In contrast, CEIS yields similar and generally lower performance across all algorithms due to its high dimensionality and complexity. Statistical tests indicate that none of the algorithms outperform the others at α = 0.05, highlighting the strong dependence of algorithm suitability on problem structure.
I Sistemi Multi Energia (MES) rappresentano una potenziale soluzione alle sfide affrontate dagli attuali settori energetici, poiché cercano di bilanciare esigenze economiche, ambientali e tecniche. Questa tesi presenta uno studio numerico comparativo di cinque algoritmi di ottimizzazione multi obiettivo applicati a casi di studio reali di MES, simulati con EnergyPLAN e ottimizzati utilizzando un approccio black-box basato sulla libreria Python Pymoo. Questo approccio consente un’esplorazione flessibile di spazi decisionali complessi senza richiedere riformulazioni esplicite del simulatore. Gli algoritmi valutati includono due metodi basati su Pareto (NSGA II, SPEA2), un approccio basato su indicatori (IBEA) e due varianti della Multi Objective Particle Swarm Optimization (MOPSO con crowding distance e con adaptive grid). I casi di studio analizzati — Aalborg, CEIS e Giudicarie — differiscono per dimensionalità delle variabili decisionali, vincoli e numero di obiettivi. Le prestazioni degli algoritmi sono valutate mediante sei indicatori complementari (Hypervolume, Additive Epsilon, Inverted Generational Distance, Overall Spread, Spacing e Ratio of Non Dominated Solutions), combinati tramite un sistema di punteggio pesato e validati con i test statistici non parametrici di Friedman e Nemenyi. I risultati mostrano che i metodi MOPSO, in particolare la variante adaptive grid, ottengono le prestazioni migliori nei casi di Aalborg e Giudicarie. Il caso CEIS presenta prestazioni simili e generalmente inferiori per tutti gli algoritmi, a causa dell’elevata dimensionalità e complessità. I test statistici indicano che nessun algoritmo supera gli altri con un livello di confidenza inferiore a α = 0,05, evidenziando una forte dipendenza dell’idoneità degli algoritmi dalla struttura del problema.
Multi-objective optimization of multi‑energy systems: algorithmic comparison across real case studies
Urrea Sanchez, Sara
2025/2026
Abstract
Multi-Energy Systems (MES) offer a potential solution to the challenges faced by today’s energy sectors, as they balance economic, environmental, and technical needs. This thesis presents a comparative numerical study of five multi-objective optimization algorithms applied to real-world MES case studies simulated with EnergyPLAN and optimized through a Python black-box model using the Pymoo library. This approach allows flexible exploration of complex decision spaces without requiring explicit reformulations of the simulator. The evaluated algorithms include two Pareto-based methods (NSGA-II, SPEA2), one indicator-based approach (IBEA), and two variants of Multi-Objective Particle Swarm Optimization (MOPSO with crowding distance and with adaptive grid). The case studies analyzed Aalborg, CEIS, and Giudicarie, which vary in decision-variable dimensionality, constraints, and the number of objectives. Algorithm performance is assessed using six complementary indicators (Hypervolume, Additive Epsilon, Inverted Generational Distance, Overall Spread, Spacing, and Ratio of Non-Dominated Solutions), combined through a weighted scoring system and validated with non-parametric Friedman and Nemenyi statistical tests. Results show that MOPSO methods, particularly the adaptive‑grid variant, achieve the strongest performance in the Aalborg and Giudicarie cases. In contrast, CEIS yields similar and generally lower performance across all algorithms due to its high dimensionality and complexity. Statistical tests indicate that none of the algorithms outperform the others at α = 0.05, highlighting the strong dependence of algorithm suitability on problem structure.| File | Dimensione | Formato | |
|---|---|---|---|
|
2026_03_Urrea_Executive Summary.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
1.03 MB
Formato
Adobe PDF
|
1.03 MB | Adobe PDF | Visualizza/Apri |
|
2026_03_Urrea_Sara_Tesi.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
2.52 MB
Formato
Adobe PDF
|
2.52 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/253102