SBOPT: A Simulation Based Optimization Algorithm

A common problem encountered in engineering practice, management science and systems analysis is to find the values of input variables which optimize some function of system performance. When systems are tab complicated to be described analytically, simulation is the appropriate tool for modeling purposes. Methods of optimization through simulation have recently become a steadily growing discipline. The present report describes a constrained optimization algorithm for a class of problems where part of the constraints consist of functional relationships between variables, determined through simulation models. This algorithm (SBOPT) utilizes sequential successive simulation approximations until the intermediary optimal solution reaches acceptable tolerance levels of discrepancy. In an example of application, the algorithm proves to converge rapidly (5-10 iterations) requiring a limited number of actual simulation performances. *School of Civil Engineering, Purdue University, W. Lafayette, IN 47907 **Department of Computer Science, Purdue University, w. Lafayette, IN 47907