基于神经网络辨识的灰色预测在精馏塔中的应用

石油化工生产中常用精馏塔的控制是一种滞后时间长、滞后常数不定的典型不确定滞后对象。这种对象控制困难,系统精度要求高,要对其进行有效控制就必须高精度预测它的输出,因为对象的不确定滞后特性,对其进行精确的输出预测始终是一个难题。针对精馏塔输出预测上的困难,提出利用神经网络首先辨识系统的滞后时间,之后在此基础上采用AR(p)(自回归)模型拟合残差的改进型灰色预测方法预测输出,基本灰色预测模型采用变步长单步灰色预测。将上述方法应用在精馏塔模型输出预测中,仿真结果表明改进后的预测方法对具有滞后、时变的系统有良好的预测效果,而且对系统参数突变、漂移等非失效型故障有一定的鲁棒性、容错性,比其他灰色预测方法更具优越性。

Application of Grey AR Prediction Based on Neural-Network-Identification in Distillation Column

The controling of distillation column used in petrochemical industry is a typical uncertain systems with long-time delay and unstable constant of delay.The requested precision for system of this kind is very high.To control the system effectively,high precision of prediction is necessary.It is a problem to forecast exactly the output of system for its uncertain time-delay characteristic.In order to overcome the difficulty in modeling,neural network was proposed to identify the delay time of such a system firstly.Moreover,the advanced grey prediction algorithm in which error was fitted with AR(p) modle was used to predict future behaviors of uncertain systems with time-delay.The based grey prediction model is one-mutative-step grey prediction.In computer simulation of distillation column's output prediction,the results show that the way introduced has higher prediction precision and better application than other models to time-delay and uncertain systems,it also has robustness and fault-tolerant towards uninvalidation troubles as parameters jump and excursion fault,and the method introduced is better than other grey prediction ways.