A gradient---regression search procedure for simulation experimentation

A gradient search procedure for simulation experimentation with constrained systems that combines gradient search with curvilinear regression in moving toward a constrained optimal solution for a system involving n controllable variables is examined.

Abstract

This paper examines a gradient search procedure for simulation experimentation with constrained systems. This procedure combines gradient search with curvilinear regression in moving toward a constrained optimal solution for a system involving n controllable variables. In a direction-determining block, at least n+1 simulation trials are performed around a current base point to establish an improving direction. Then in a step determining block, t simulation trials are performed along the improving direction to establish the most favorable step in moving to the next base point. This sequential block process, in which each block is executed in one input to the computer, is repeated until an approximate solution is found which satisfies all system constraints.