Accelerating the convergence of parameter search techniques during simulation-based optimization by dimension reduction methods

Simulation-based optimization gains more and more acceptance as a decision support technology for the analysis of complex systems. In practice, the identification of the best simulation model's parameter typically results in long computation times, which often makes practical applications impossible due to the large size of the parameter space. Though one can achieve a significant speed-up by using parallel or distributed simulations, computation times still remain the limiting factor for many applications. Hence, a faster convergence is required, especially for time-consuming simulations. In this paper we present an approach that uses a dimension reduction strategy in order to shrink the size of the search space of possible solutions during the search procedure. We tested our approach by applying it to a parameter search for a statistical physics problem, namely a Monte Carlo simulation of a magnetic system, where the magnetization is the physical quantity of interest. By least-squares minimization the simulated data is compared to experimental results in order to obtain an optimal parameter set. Furthermore, we have tested our method by applying it to two common benchmark function