基于PCA-RBF神经网络的工业裂解炉收率在线预测软测量方法

为了解决工业裂解炉收率在线预测的问题,研究了基于PCA(principal component analysis)-RBF(radial basis function)神经网络模型的多输入多输出(MIMO)软测量方法及其在线校正技术。该方法由主元分析PCA、RBF神经网络和在线校正3部分组成,可以实现工业裂解炉收率的在线预测,具有实时性好、建模周期短、计算量小、校正方便等特点。本文给出的SRT-IV型工业裂解炉收率预测例子及其结果表明该软测量方法应用于工业裂解炉收率的在线预测是有效的。

A Soft-sensing Approach to On-line Predict the Yields of Industrial Pyrolysis Furnace Based on PCA-RBF Neural Networks

The industrial pyrolysis furnace is a key equipment in ethylene process. It is very essential and important to on-line obtain the accurate yields of products not only for advanced control, process optimization, production evaluation, but also for performance monitoring, safety supervision, and fault diagnosis. However, it is usually difficult to on-line measure the product yields because of many practical reasons in some ethylene plants. In order to solve the problem on on-line yields prediction of industrial pyrolysis furnace, a soft-sensing approach with multi-inputs and multi-outputs (MIMO), and on-line correcting methods are investigated based on PCA(principal component analysis)-RBF(radial basis function) neural networks. The topology structure of this soft-sensing approach is as follows: the first unit is PCA, the second is RBF neural network, and the third is correcting unit. So the soft-sensing approach combines the abilities of PCA to de-correlate the variables and reduce the dimensionality of the data matrix with that of neural network to approximate any complex nonlinear function. The approach, which is of good real-time property, short modeling time, little calculations, and easily correcting, can be applied to on-line predict the yields of industrial pyrolysis furnace. The good performance of PCA-RBF soft-sensing approach for on-line yields prediction is illustrated by the example from a SRT-IV furnace. The results show that the soft-sensing approach to on-line predict the yields of industrial pyrolysis furnace based on PCA-RBF neural network is effective.