基于神经网络的实时专家控制系统及其PTA工业应用  

Neural network based real-time expert control system and its applications to PTA industry

作　　者：杜文莉[1] 钱锋[1] 

机构地区：[1]华东理工大学自动化研究所,上海200237

出　　处：《控制与决策》2005年第6期694-697,共4页Control and Decision

基　　金：国家973计划项目(2002CB3122000);国家自然科学基金项目(60074027);国家863计划项目(2003AA412010);上海市优秀学科带头人计划项目.

摘　　要：以精对苯二甲酸结晶过程为研究对象,提出一种基于神经网络模型的实时专家控制系统.该方法利用神经网络建模技术获取对象的机理知识,通过对影响模型特性的多个变量进行分析,自动得到常规专家控制系统难于获取的定性、定量知识,并按分级递阶的启发式搜索机制,实现了对工业过程对象的实时控制.实际应用表明:该方法不但克服了以往专家系统知识获取的瓶颈,而且有效实现了人机对话的功能,便于现场操作和更改专家知识库,为化工过程的多变量控制提供了新的思路.A kind of neural network model based real-time expert control system (MRECS) is proposed for PTA crystallization process. This system acquires mechanism knowledge of the object via artificial neural networks. The qualitative and quantitative knowledge can be easily achieved by analyzing the influences of the multiple input variables on PTA average particle size in the model. Hierarchical searching method is applied to the expert control system to implement real time control of PTA industrial crystallization process. Its successful application demonstrates the validity of the proposed MRECS. The proposed MRECS can not only overcome the bottleneck of knowledge acquisition, but also realize the online knowledge rerising through man-machine function. It provides a new idea of multi-variable control scheme suitable for chemical process.

关 键 词：实时专家控制系统 神经网络模型 知识自动获取 分级递阶搜索 多变量控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]