基于径向基函数神经网络和NLJ优化算法的精馏塔控制

Optimization control of a distillation column based on the RBF neural network and NLJ algorithm

精馏塔是一个非常重要的操作单元,具有较强的非线性和时变性,很难进行基于机理建模分析的实时优化控制.通过对精馏塔的相关过程变量进行主元分析确定了5～6个关键变量作为神经网络的输入,建立了精馏塔多个质量指标的RBF神经网络的软仪表模型,实现了这些质量指标的在线估计.选取其中部分软仪表模型作为优化控制系统中的约束条件函数模型和目标函数模型,采用NLJ优化算法（变收缩系数的随机搜索算法）获取最优的决策变量设定值,从而得到了满足生产质量要求的精馏塔产品的最大采出,实现了精馏塔的卡边优化控制.

A distillation column is a very important operating unit in the process of chemical production. Its nonlinear and time varying characteristics make the modeling and analyzing based on system identification very difficult. In order to realize the optimization control of a distillation column, the paper presents some soft-sensing models of quality targets based on the RBF neural network. They are used as the constrained condition and target function modes required by the NIJ optimization control to get their online estimations. Then try to find the optimizing set value of the key process variable to get the maximum output of the distillation column by using the NLJ optimization algorithm.