A New Approach to Simplex Optimization -- the Super Modified Simplex.

摘要

A new, improved simplex optimization procedure has been developed which significantly increases the utility, reliability and efficiency of current state-of-the-art simplicial methods in the presence of random response fluctuations. Comparison of this improved procedure, the Super Modified Simplex (SMS), with several variations of the modified simplex approach using an interactively computer-controlled flame spectrophotometer reveals its inherent superiority in rapidly and efficiently conforming to multi-dimensional response surface topography. Data are presented which indicate fewer simplices and data points, as well as a reduction in experimental time results when the SMS procedure is employed. Moreover, failure to satisfactorily locate the optimum in the presence of noise due to premature diminution of the simplex, adherence to false ridges, or inability to accurately home in on the optimum is minimized. When the optimum region lies at or near a boundary constraint the SMS procedure allows the simplex to elongate along and more rapidly adhere to the constraint if such a parameter value is indicated.