建模技术在虚拟仪表中的应用

【作者】 马昌浩；

【导师】 韩志刚；

【作者基本信息】 黑龙江大学 ， 检测技术与自动化装置， 2012， 硕士

【摘要】 随着测控领域对测试要求的不断提高，传统仪表在精度上、性能上，功能上都取得了长足的进步，但是由于技术落后、成本高以及工业环境差等原因，传统仪表仍然无法满足工艺复杂、被测参数多的场合，尤其是在诸如冶金、化工、煤炭、石油等工业中，传统仪表仍然停留在手动操作阶段，却无法在线实时对参数进行测量。正是基于上述等原因，虚拟仪表以其独有的优势而出现，并且在测控领域逐渐地取代了传统仪表的地位，成为自动控制领域重点研究的方向。构建一个虚拟仪表的核心是对实际生产过程相应工艺参数进行数学建模，本文拟采用人工神经网络方法来解决这一问题，因而对基于统计建模思想的人工神经网络方法进行了深入研究。其中重点论述了BP神经网络和RBF神经网络，提出了基于BIC准则的BP神经网络隐层节点个数的优选方法，并利用遗传算法对BP网络初始连接权值和阈值进行优化；针对单个RBF网络分类精度不高的缺点，设计了基于Adaboost算法的RBF强分类器。仿真结果表明，本文提出的这些改进方法可以有效的进行网络结构优选并进一步提高网络的训练速度和分类精度。最后，根据本文所述的建模方法，利用LabVIEW软件编写了一个聚合物黏度软测量系统。该系统可以对从工业现场实时采集的数据进行处理分析，并根据这些数据最终给出聚合物的黏度值。更多还原

【Abstract】 With the test requirements increased of the monitoring and control areas, thetraditional instruments have made considerable progress with the accuracy, performanceand functions, but due to backward technology, high cost and poor industrialenvironment, they are still unable to meet the occasion of process complex, moremeasured parameters, especially in industries such as metallurgy, chemicals, coal, oiland so on, traditional instruments still remain at the stage of manual operation, can notmeasure parameters online in real time. In case of these reasons above, the virtualinstruments appeared with their unique advantages, and gradually replaced the status ofthe traditional instruments in the measurement and control field, which become thedirection of the key research in the field of automatic control.The core of builds a virtual instrument is the mathematical modeling of thecorresponding process parameters of actual generation process. This paper proposedartificial neural network approach to solve this problem, so conducted in-depth researchon artificial neural network method based on statistical modeling ideas. This paperfocuses on the BP neural network and RBF neural network, BP neural network hiddenlayer nodes select method based on BIC rule is proposed, and use of genetic algorithmsto optimize the BP network initial connection weights and thresholds; for a single RBFnetwork has the shortcoming of classification accuracy is not high, designs of a RBFstrong classifier based on Adaboost algorithm. The simulations result show that theseimprovements proposed in this paper can effectively optimizes network structure andfurther improve the training speed and classification accuracy of the network.Finally, based on the modeling approach described in this paper, a polymerviscosity soft measurement system is compiled using the LabVIEW software. Thesystem can deal with and analysis the data which real-time collected from an industrial site, and ultimately gives the value of the polymer viscosity according to these data.更多还原