OPTIMIZATION OF DISCRETE VARIABLE STOCHASTIC SYSTEMS BY COMPUTER SIMULATION (MANUFACTURING, COMPLEX SEARCH, CONFIDENCE INTERVAL, FIMS)

A heuristic algorithm (SIMICOM) has been designed and tested for optimization of discrete variable stochastic systems which are modeled through computer simulation. The systems under study are stochastic and complex in nature and their performances are functions of several discrete decision variables. However, due to the complexities involved, their performance measure can be evaluated only through computer simulation. The approach adopted utilizes an Integer Complex (Constrained Simplex) technique coupled with techniques of establishing confidence intervals for the system's response. It is capable of optimizing a general class of problems that could be constrained or unconstrained. In case of constrained problems, the constraints could either be explicit analytical functions of decision variables or be expressed as other responses of the simulation model. In addition to attempting to obtain a reasonably accurate solution, the economic aspect of obtaining the solution in as few simulation runs as possible has also been taken into consideration. This is done by a specially designed sequential procedure which disregards extremely suboptimal solutions after running the simulation model for a short time. The superiority of the algorithm over two other known techniques is demonstrated by applying it to several test problems.