Abstract
In the field of energy system modelling, increasing complexity and optimization analysis are essential for understanding the most effective decarbonization options. However, the growing need for intricate models leads to increased computational time, which can hinder progress in research and policy-making. This study aims to address this issue by integrating machine learning algorithms with EnergyPLAN and EPLANopt, a coupling of EnergyPLAN software and a multi-objective evolutionary algorithm, to expedite the optimization process while maintaining accuracy. By saving computational time, we can increase the number of evaluations, thereby enabling deeper exploration of uncertainty in energy system modelling. Although machine learning models have been widely employed as surrogate models to accelerate optimization problems, their application in energy system modeling at the national scale, while preserving high temporal resolution and extensive sector-coupling, remains scarce. Several machine learning models were evaluated, and an artificial neural network was selected as the most effective surrogate model. The findings demonstrate that incorporating this surrogate model within the optimization process reduces computational time by 64% compared to the conventional EPLANopt approach, while maintaining an accuracy level close to that obtained by running EPLANopt without the surrogate model.