Mick Voogt, "Reducing Evaluation Time for a Multi-objective Energy System Sizing Problem using Dynamic Simulation Time", Leiden University (Netherlands), 2023.

A hybrid energy system integrates multiple energy sources, unlike systems that rely solely on a single source. However, incorporating these diverse energy sources can be intricate due to the variable nature of renewable energy supply. An approach to address these energy fluctuations is by incorporating energy storage, enhancing system resilience but also increasing costs. Another option would be to rely more on conventional energy sources, which is cheap but potentially causes more pollution to the environment. Achieving the optimal balance between cost, resilience, and environmental impact in these energy systems is challenging due to the absence of a universal solution. As such, this optimization can be approached as a Multi Objective Optimization Problem. In this thesis we experiment with evolutionary optimization and simulations to solve such a problem. When using evolutionary optimization to solve real world problems using a large simulation or computation as an evaluation function, the individual evaluation time can severely limit the number of evaluations and quality of the solutions within a set time limit. Speeding up these simulations is therefore essential but the new simulation still has to represent the original problem in order to generate relevant solutions. In this research we will be looking at a real world optimization problem that uses a simulator as an evaluation function for a hybrid energy system for an office space. Our goal is to try to keep quantitatively equivalent simulation results while reducing the simulated period. Continuing with the hybrid energy system optimization problem and framework from [10], we explore different options to reduce the simulation time for the evaluation function. To do this we change the simulated period from a full year to a single month to represent the full year. We also change this simulated month dynamically every generation, essentially creating a dynamic fitness function. The results from these experiments cannot directly match the quality of the baseline results in terms of hypervolume and IGD+ score which means we cannot directly reduce simulation time. However, we do explore an interesting possibility to start an optimization run using the dynamic fitness function and switching later to the original, in order to acquire a better solution set in the same amount of time.