Rough Set Theory And Its Application In Fermentation Process Control

Rough Set Theory (RST) is a strong tool for intelligent data reasoning. It deals with representation, learning and generalization of uncertain information and has a bright future in the application to industrial process control. This thesis considers some related questions: What is the present state that RST is applied to process control? Can this technique be applied to complex fermentation process control, and how? Since not all of the decision rules generated by RST are completely credible, how to calculate the confidence of rules accurately?The main contents of this thesis are as follows:(1) The fundamental concepts and principles of RST are introduced. The state of the art of RST is summarized from both theory and application aspects. RST software is also introduced in this thesis.(2) Rules extracted by RST from data sets can not be trusted without doubt. To this problem, this thesis modifies a formula for computing confidence of rough rules, which is based on information incompletion, and consummates a computing method based on rough membership function. Considering the effects on confidence caused by both incomplete and incompatible information, this thesis presents a new approach to computing confidence in rough set theory. Then, the method for computing compound confidence of a decision from multiple rules is given, and uncertainty transfer is considered. Some examples validate the efficiency and precision of this approach. This approach avoids the blindness effectively when we use the decision rules.(3) Based on abundant papers, applications of RST to process control are summarized, including getting control algorithms, fault diagnose, process modeling, inspecting online, and so on. Then, the combination of RST and other intelligent control tools is analyzed to congregate advantages and improve control effects. As a newly developed intelligent control tool, rough control is worthy of research.(4) Fermentation process is so complicated that classical control methods based on mathematics models are often incompetent. On the other hand, there are usually a great deal of knowledge and rules of microorganism growth hiding in the history data. To this problem, this thesis gives primary schemes applying RST toidentifying occasions of substrate feeding, controlling feeding and fault diagnosis in fermentation process. A method of selecting decision rules is presented.In the end, some problems for further research are addressed.