摘要
软测量技术可以有效解决复杂工业过程中一些重要参量难以由硬件在线检测的问题,由于化工过程具有连续性和累积性等特点,若采用传统的软测量建模方法往往会忽略信号的时间累积作用从而导致预测误差较大。针对上述问题,提出了基于改进的过程神经网络(PNN)的软测量建模方法。首先采用移动窗技术来确定包含过程正常运行大部分信息的时间序列,然后利用改进的PNN建立软测量模型并对主导变量进行连续预测,最后对软仪表进行校正以实现连续高精度预测。以某工厂高密度聚乙烯装置为例,验证了该方法具有较高的预测精度和跟踪性能,这对于工业过程的控制优化操作具有重要的应用价值。
Soft-sensing can effectively solve problems in complex industrial processes where hard sensors can not easily measure important parameters. Chemical processes are characterized by continuous, cumulative results and traditional soft sensing models tend to ignore the cumulative effect of time signals which may lead to large forecast errors. This paper presents a soft sensor modeling method based on a process neural network. A moving window technique is used to determine the time series containing most of the information, with the process neural network then used to drive the soft-sensor. The soft sensor is adjusted to achieve continuous, accurate estimates. A high-density polyethylene （HDPE） factory is used as an example to show that the method can provide accurate on-line estimates of the polymer＇s melt index with good tracking performance, with the adaptive capability ensuring the reliability of the online system.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2012年第9期1165-1170,共6页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金资助项目(61074153)
关键词
软测量
过程神经网络
Q统计量
高密度聚乙烯
soft-sensing
process neural network （PNN）
Q statistie
high density polyethylene （HDPE）
