@inproceedings {pub3867,
	title = {Surrogate Assisted Interactive Multiobjective Optimization in Building Energy System Design},
	author = {Pouya Aghaei-Pour AND Tobias Rodemann AND Markus Olhofer AND Jussi Hakanen AND Kaisa Miettinen},
	year = {2019},
	month = {June},
	abstract = {Managers of larger buildings are confronted with complex investment decisions concerning possible extensions of the energy system, like photo voltaics(PV), stationary batteries or heat storage. They have to consider a multitude of objectives, for example, investment and annual operation costs, CO2 emissions and module lifetime. A computer-assisted optimization and decision making process promises substantial benefits in this complex problem. In this research, we provide decision support in building energy system management by applying interactive multiobjective optimization methods. We consider five objectives (initial investment cost, running cost, CO 2 emissions, resilience to power outages, battery lifetime) for a system upgrade with both hardware additions and modifications of system controllers. We use a building simulation software based on the Modelica standard to simulate energy flows in a facility with different investment options.
For optimization, we apply evolutionary algorithms since, in the past, different evolutionary algorithms have been successfully tested to find viable investment plans for the problem considered. A major drawback of evolutionary algorithms is the long computing times (days to weeks) of a single optimization run, which stretches the patience of 
(high-level) decision makers (DMs) to the limit. In order to speed up the computation, we apply surrogate models. In particular, we focus on Kriging as surrogate models (also known as Gaussian processes) due to the ability of providing uncertainty information about the surrogates. With five conflicting objectives, it is not easy to get a good representation of
Pareto optimal solutions and, hence, we focus on applying interactive methods, which have not been applied in this problem before. The benefits of interactive
approaches are threefold. First, the computation time is reduced because the algorithm will focus on those Pareto optimal solutions that reflect the preference
information of the DM, not all Pareto optimal solutions. Second, the DM can direct the solution process to focus on those solutions that are interesting and
does not need to spend time with uninteresting solutions. Finally, and most importantly, thanks to the iterative nature of the methods, the DM can learn
about the nature of the problem and trade-offs involved as well as the feasibility of one{\textquoteright}s preferences. This will increase the confidence of the DM in the results
of the optimization process and increase the chances of actually realizing the final solution identified.We demonstrate how the problem considered can be solved with an interac-
tive surrogate-assisted multiobjective optimization method and discuss the findings. We apply variants of the reference vector guided evolutionary algorithm
(RVEA), incorporate surrogates in the consideration and perform an interactive optimization using surrogates. Finally, we compare the quality of the solutions
obtained to those of a noninteractive approach.},
	publisher = {MCDM},
	booktitle = {25th International Conference on Multiple Criteria Decision Making },
	city = {Istanbul}
}