A comparison of the performance of techniques for direct,on-line optimization

The performance of a number of well known optimization techniques has been compared with that of recently developed weighted-mean random techniques, in the context of on-line process control applications. It was found that one variation of the random technique compared favorably with the well known pattern search technique of Hooke and Jeeves in finding the maximum on simulated process response functions. The study was performed by representing the processes by mathematical expressions. The effects of multidimen-sionality, irregular response surfaces, scaling errors, and process noise were simulated in the tests that were performed for the comparisons.