基于模型的多尺度间歇过程性能监控

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.

Model Based Multiscale Performance Monitoring for Batch Processes

Batch process is one of the most important processes in chemical industry, and how to monitor the performance of batch processes has always been one of the most active research areas in process control. In this paper, neural network (NN) is used to describe the nonlinear and dynamic behavior of batch processes, and the predicted residuals of NN is modeled through the extension of multiscale principal component analysis (MSPCA) to batch processes. Compared to the multiway principal component analysis (MPCA) with a linear model, the proposed method is a multi-model, nonlinear model-built method. Each of the residuals is decomposed into the approximations and details using wavelet analysis, and principal component analysis is employed to develop a statistical model at each scale. The advantage of proposed method over the traditional MPCA is demonstrated on the industrial streptomycin fermentation process, and the smaller detection delay is also obtained.