Huang Dexian

Application of non-negative matrix factorization to signal curve recognition in a real-time control software

This paper aims at upgrading the current pattern of electronic tubes real-time control software (ETRCS) application. It presents a new method based on non-negative matrix factorization (NMF) to recognize several types of curves in ETRCS. Theoretically, NMF method can extract the part features of objects that resemble intuitive notions in an unsupervised way, which is similar to human visual perception. The test results based on real data indicate that this method can effectively recognize different curves, and will contribute to improving the tubes quality

Wavelet networks based soft sensors and predictive control in fermentation process

Abstract In this paper, a kind of soft sensor for penicillin fermentation processes using wavelet neural network is developed. And then a predictive controller based on wavelet neural network model is presented. Comparing with other neural networks used in fermentation processes, the wavelet neural network is of good generalization and its structure is easy to specify. Weights of the wavelet neural network are of excellent statistic properties such as unbiased estimation The soft sensors show high capability to estimate states of penicillin fermentation process. Biomass soft sensor was used in plant and the estimation is accurate. By simulation the predictive controller based on wavelet network shows promising performance.