基于小波核函数极限学习机的模型预测控制模拟  

Simulation of Model Predictive Control Strategy Based on Wavelet Kernel Extreme Learning Machine

作　　者：徐凤[1] 刘爱伦[1] 

机构地区：[1]华东理工大学自动化系,上海200237

出　　处：《华东理工大学学报（自然科学版）》2015年第2期185-191,共7页Journal of East China University of Science and Technology

摘　　要：针对醋酸精馏控制中,产品质量采用常规的温度间接控制存在精度低的问题,提出了一种基于小波核函数极限学习机的模型预测控制(KMPC)策略,在醋酸浓度软测量的基础上直接控制产品质量。鉴于小波核函数极限学习机(KELM)算法训练速度快并且稳定的特点,该控制系统采用KELM建立醋酸浓度控制器预测模型,以预测控制器的输出作为再沸器蒸汽流量控制器的设定值,构成串级调节系统,同时,以灵敏板温度、塔底温度、再沸器入口温度、压力等变量作为扰动变量,实现了对复杂精馏过程的前馈控制和非线性预测控制。运用ASPEN DYNAMICS流程模拟软件建立的醋酸精馏塔动态模型对KMPC策略进行仿真研究,结果表明,与传统DMC预测控制方案比较,塔底醋酸浓度控制精度有较大提高,控制结构简单,易于实施,能够实现产品质量的卡边控制。In the acetic acid concentration, the temperature cannot be well controlled by indirect control. By using wavelet kernel extreme learning machine, this paper proposes a model predictive control strategy （KMPC）, which directly control the quality of products on the basis of acetic acid soft-sensor. It is known that wavelet kernel extreme learning machine （KELM） has the characteristics, e. g. , fast training speed, high precision, and strong generalization ability, which is utilized in the proposed KMPC to establish the prediction model of the acetic acid concentration. Moreover, the predictive controller is used as the master controller for the acetic acid concentration and a PID controller is used as the slave controller for the reboiler vapor flow, both of which constitute a cascade control system. Besides, the sensitive temperature, bottom temperature, inlet temperature, and pressure of reboiler are taken as the disturbance variables to achieve the feedforward control and nonlinear predictive control to the complex distillation column. By means of the simulation based on ASPEN DYNAMICS simulation software, it is shown that, compared with the traditional dynamic matrix control （DMC） method, the proposed strategy can improve the control efficiency of the acetic acid concentration and is easily applied due to its simple structure.

关 键 词：醋酸精馏 模型预测控制 软测量 小波核极限学习机 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]