论文部分内容阅读
利用神经网络对间歇过程的非线性和动态特征进行描述,神经网络的预测残差则利用多尺度主元分析进行建模,将多尺度主元分析扩展用于间歇过程的监控.这一方法突破了传统多向主元分析单模型、线性化的建模方式,是一种多模型非线性建模方法.它利用小波将每一残差信号分解为各个尺度上的近似部分和细节部分,而主元分析则用于分别建立各个尺度上的统计模型.通过对实际工业链霉素发酵过程数据的分析,表明文中所提出的方法与传统的多向主元分析方法相比,能够更早地发现故障,获得更好的监控性能.
The neural network is used to describe the nonlinear and dynamic characteristics of batch processes. The prediction residuals of neural networks are modeled using multi-scale principal component analysis, and the multi-scale principal component analysis is extended to the monitoring of batch processes. The traditional multi-direction principal component analysis of single model, linear modeling method is a multi-model nonlinear modeling method which uses wavelet to decompose each residual signal into the approximate part and the detail part of each scale, The principal component analysis (PCA) is used to establish the statistical models for each scale separately.Analysis of the actual industrial streptomycin fermentation process data shows that the proposed method can be used earlier than traditional multi-direction PCA methods Find fault, get better monitoring performance.