基于数据挖掘技术的全局优化算法

The Global Algorithms Based on Data Mining

在过程系统综合中,许多问题属于非线性规划(NLP)和混合整数非线性规划(MINLP)范畴.它们大都具有奇异、多峰、刚性等特性.人们很难有效地得到它们稳定的全局最优解.而知识性、经验性约束使基于梯度方向的Newton方法无法有效地获取该类问题的全局最优解.通常只能得到该类问题的局部最优解.遗传算法的随机性虽为求取NLP和MINLP问题的全局最优解提供了可能,但是随机过程中的盲目性及"伪穷举"性却又限制了该算法的搜索效率.针对过程系统综合问题的特殊性,在信息提取技术对搜索空间进行充分数据挖掘的基础上,用遗传算法的随机扰动来跳出局部极值陷井,获得全局最优解.对反应器网络综合问题的求解,显示了信息提取技术与遗传算法相结合求取全局最优解的能力.

Many problems in the process system synthesis belong to the non-linear problem （NLP） and mixed integrated non-linear problem （MINLP） category and they are almost singular, multipeaks and rigid. There is no effective means to get their stable global-optimal. Those solutions belong to local-optimal because Newton method based on grads, which subject to knowledge and experience didn＇t effectively get the solution domains of processing system synthesis. Although it is possible to get global-optimal by means of the randomicity of genetic algorithms, blindness and pseudo-enumeration of the randomicity may restrict the method search efficiency. The paper presents a way to combine information selection with genetic algorithms. Based on data mining in possible domains, the randomicity of genetic algorithms is used to dap out of the local-optimal trap. The test reactor networks synthesis showed that the method is effective and useful.