An innovative nature-inspired heuristic combined with response surface methodology to find the optimal region in discrete event simulation models

The search for a stationary point in the study of Discrete Event Simulation models is a complex problem. This is because the equation of the objective function is never known a priori to the experimenter. In the case of restricted investigation domains the Response Surface Methodology typically provides, through the use of Central Composite Design, the experimental design most suitable for the construction of first and second order regression meta-models. The problem becomes more complex when the domain to be investigated is larger because, in that case, it becomes impossible to identify a meta-model regression able to describe the behaviour of the objective function on the entire domain. Such a limitation can be overcome by an appropriate use of research techniques such as gradiental or direct search methods. However, the presence in the domain of local stationary points may affect their effectiveness and forces the experimenter to track the investigation starting from several points of the domain, with a consequent increase in the number of function evaluations and computational time. In more recent times global research techniques have been developed, often inspired by natural processes. However they generally not perform well applied to Discrete Event Simulation models. For this reason the Authors have developed a new search algorithm called Attraction Force Optimization (AFO). The proposed approach applied to industrial problems up to 10-dimensional, offers significant advantages in terms of both exploration capacity and convergence speed. An application of the proposed technique to a real industrial case completes the discussion.