A Pattern Search Algorithm for Feedback Control System Parameter Optimization.

摘要

The problem of parameter optimization for the purpose of synthesis and design of multi-loop feedback control systems is considered. A flexible performance index,based on performance specifications such as rise time,settling time,peak overshoot and bandwidth is developed and adapted to a modified version of Rosenbrock's pattern search algorithm. Modifications of Rosenbrock's method include scaling of parameters to make coordinate-orthogonalization more effective and employing variable step sizes at the beginning of each new stage,the steps being a function of previous failures and successes. The efficiency of each stage is thus improved and the sensitivity to initial estimates of step sizes is reduced. A stopping criterion for the search based on a given number of consecutive stages with no improvement of the performance index (within the precision of the computer) is shown to be effective. Calculation of sensitivity numbers and desired performance specifications around the optimal point is included to give a feel for system behavior near optimal conditions. The IBM/360Continuous System Modeling Program (CSMP) is employed,permitting system simulation directly from a block diagram. (Author)