Planning and decision-making have become increasingly complex, especially in the energy sector. Modeling to Generate Alternatives (MGA) enables the creation of multiple alternative solutions, enhancing the decision-making process. However, when applied to large system optimization problems, the MGA method can become computationally cumbersome. Bio-inspired metaheuristics, a branch of artificial intelligence, have the potential to overcome these computational limitations by applying metaheuristic methods that can solve multiple model solutions simultaneously. Currently, a sophisticated integration of such heuristics with MGA, specifically focused on energy system optimization, has not yet been realized.

This thesis aims to answer the question: How does a combination of MGA and bio-inspired heuristics, aimed at optimizing energy system design, compare with existing deterministic MGA methods? To address this question, a metaheuristic-MGA algorithm was developed and compared with existing spatial energy system MGA results. First, a literature review was conducted to determine the most suitable metaheuristic for this study. The review confirmed the scarcity of literature on the combination of metaheuristics and MGA in energy systems. However, relevant studies applying a metaheuristic-MGA approach to general optimization problems were identified. Based on these findings, a genetic algorithm (GA) was selected as the most appropriate method for this thesis. The mathematical formulation presented in the literature was adapted to fit the spatial optimization problem of energy systems.

The complete mathematical process of the GA-MGA algorithm was developed in Python. Next, a small energy system test model was built in Calliope to evaluate the performance of the developed GA-MGA algorithm. The algorithm’s parameters were further fine-tuned using existing parameter-tuning methods, performance measurements, and assessments of the computational time required to complete the algorithmic process.

Before applying the developed GA-MGA algorithm to a large-scale model, it needed to be scaled to prevent errors or computational inefficiencies when generating results. It was determined that the desired resolution was not feasible due to the excessive computational time required for its completion. Instead, a time-masking method was applied to the resolution, preserving high-resolution characteristics while improving computational efficiency.

The GA-MGA algorithm was then applied to a large European energy system model, and the results were compared to existing MGA results. However, due to differences in resolution, the GA-MGA-generated results did not meet the standards of the existing MGA results, making direct comparisons less robust and reliable than desired. The comparison revealed a significant difference in battery capacity deployment, with the GA-MGA solutions deploying higher quantities of battery capacity. The lack of spatial distribution data for the large model was solved by comparing the GA-MGA results to a plot of the existing MGA results. While this was not an ideal comparison, it provided an opportunity to analyze spatial deployment differences between the GA-MGA and the existing MGA results. The comparison showed that the GA-MGA algorithm favored high-capacity deployment at specific locations, whereas the existing MGA results exhibited a more diverse capacity distribution.

The limitations of the results primarily stemmed from shortcomings in spatial comparison and differences in resolution between the two modeling techniques. Another key limitation was the algorithm’s structure, which presents several opportunities for improvement in optimizing the GA-MGA approach. Despite these challenges, the theoretical combination of GA-MGA demonstrated promising potential. Future research should focus on enhancing the algorithm’s performance and conducting more in-depth comparisons with MGA results to fully evaluate its effectiveness. Further research in this area could expand access to the MGA method for tackling large, complex problems, ultimately contributing to more effective planning and decision-making processes. This, in turn, would support efforts to address major societal challenges.