自适应调优专家仿真系统

Expert Simulation System for Adaptive Optimization

本文分析了调优问题的数学实质,指出关键是建立数值仿真模型,并将建模的主要方法分为理论分析、经验机理分析和统计分析三类.作者考察了若干种多元统计分析方法的特点后,采用了一种将判别和聚类分析相结合的模式分类方法.作者用实际数据说明了复杂过程的时变特性,提出用渐进式调整方法,可达到自适应调优目标.本文指出在线自适应调优的实现需要专家知识的帮助,提出了将专家知识和仿真计算结合而构成专家仿真系统的一种模式.最后还报道了调优软件在工业装置上实际试运行取得的效果示例.

In this paper the importance of mathematical modeling in solving optimization problems was emphasized.The principal methods for modeling were classified into three classes,namely theoretical,empirio-theoretical and statistical analyses.After examining the advantages and shortcomings of some multivariate statistical methods,a patternized classification method which combines the discriminating and clustering techniques was proposed.The time-dependent character of complex processes was demonstrated in this paper through statistical analysis,and an evolutionary scheme for adaptively optimizing process operation was suggested.The author indicated that the expert knowledge is crucial to on-line adaptive optimization procedure,and proposed a scheme of Expert Simulation System closely combining the domain knowledge with numerical calculations.The practical benifits of applying the adaptive optimization procedure to commercial chemical plants were reported.