Reference point-based evolutionary multi-objective optimization for industrial systems simulation

Abstract

In Multi-objective Optimization the goal is to present a set of Pareto-optimal solutions to the decision maker (DM). One out of these solutions is then chosen according to the DM preferences. Given that the DM has some general idea of what type of solution is preferred, a more efficient optimization could be run. This can be accomplished by letting the optimization algorithm make use of this preference information and guide the search towards better solutions that correspond to the preferences. One example for such kind of algorithms is the Reference point-based NSGA-II algorithm (R-NSGA-II), by which userspecified reference points can be used to guide the search in the objective space and the diversity of the focused Pareto-set can be controlled. In this paper, the applicability of the R-NSGA-II algorithm in solving industrial-scale simulation-based optimization problems is illustrated through a case study for the improvement of a production line. INTRODUCTIONSimulation-based multi-objective optimization problems are in many cases difficult to solve which is caused by the vast number of simulation runs that are needed in order to find a converged and diverse set of Pareto-optimal solutions. The standard goal usually pursued in multi-objective optimization is to find a converged set of solutions that is also featuring diversity. This goal requires to find a large number of solutions to cover the whole Pareto-front. Besides a complex problem structure the high number of simulation runs can also be caused by uncertainty in the input or output parameters of the model which has to be handled by simulating every solution multiple times and using the average objective values. Also a high number of problem objectives can make it necessary to execute many solution evaluations. This is because the Pareto-optimal front of a high-dimensional problem is a large subspace. In order to cover it sufficiently many Pareto-optimal solutions need to be found. For complex real-world simulation models the execution time usually is high so that the overall number of simulation runs that can be executed is limited. With this limited simulation time budget in many cases only an inferior Pareto-optimal front can be found. For many optimization problems however the DM has domain knowledge about which solutions are most interesting. After the optimization has been run only a single solution is chosen and implemented. If the DM can provide preference information that indicates where the DM expects to choose a solution the algorithm can exploit this information and guide the search towards this area. In this way only a small part of the Pareto-optimal front needs to be covered 978-1-4673-4781-5/12/$31.00 ©2012 IEEE

Reference point based evolutionary multi-objective optimization with dynamic resampling for production systems improvement

“…As R-NSGA-II uses non-domination sorting, it tends to prioritize population diversity before convergence to the RPs. Therefore, extensions have been proposed which limit the influence of the Pareto-dominance and allow the algorithm to focus faster on the RPs (Siegmund et al , 2012).…”

Investigating the Use of a Modified NSGA-II Solution for Land-Use Planning in Mediterranean Islands

“…Shaygan et al [5] proved that NSGA-II performs successfully in comparison to Goal Attainment-MultiObjective Land Allocation (GoA-MOLA). Researchers [25] presented a revised version of NSGA-II, named NSGA-III, to deal with issues, which include four or even more objectives while another version, named Reference point-based NSGA-II facilitates the specification of reference points by the users to "guide the search in the objective space and the diversity of the focused Pareto-set can be controlled" [26].…”

Reference point-based evolutionary multi-objective optimization for reversible logic circuit synthesis

“…REFERENCE POINT-BASED NSGA-II ALGORITHM NSGA-II [17] is a classic multiojective algorithm which was able to matintain a better spread of solutions and converge better in the obtained nondominated front. Crowding comparative operator ( n ) of NSGA-II defines a partial order of solutions to provide a basis for the two selection stage blocked in Fig.…”