Application Study Of Data Mining In Acetone Refining Process

In this thesis, the data mining algorithms applied in process industrial production are researched, and the application of data mining methods represented by variable precision rough set in product quality prediction is studied under the background of acetone refining equipment in a petrochemical enterprise. Based on it, a data mining prototype system in acetone refining process is designed and implemented. The main contribution and results of the thesis can be summarized as follows:The status quo and development trends of researches and applications in data mining are surveyed. The characteristics and application difficulties of process industrial data are analyzed. The development and status of data mining software tools and the process model of data mining are presented. Based on the introduction of the rough set theories, the variable precision rough set theory and its mining algorithm are expounded.The production sets and procedures in producing acetone and phenol are described in this thesis, and some analyses on the process data in production are given. Contraposing the problems of acetone refining production data such as mass, high dimension, effects of multiple time marks, imperfectness, and multiple modes, the suited data preprocessing technologies are selected, and the fuzzy c-means method is employed to discretize the data sample.Then the variable precision rough set data mining algorithm is improved by introducing emergence frequency threshold and multi-confidence thresholds for selecting decision rules and adopting summation judgment to obtain classification results. The proposed new algorithm performs much better than BPNN and Apriori algorithms in quality prediction, and also better than standard rough set algorithm. The results show that the improved variable precision rough set data mining algorithm has practical value and research prospect.Base on the requirement analysis, a data mining system for acetone refining process is designed. It contains design for functional model, structure frame, implementation, software interface and modules, etc. Contraposing the requirement of continuous mining, a method of dynamic combination and update for rule library is proposed in this thesis with the concept of rule-support frequency, and its feasibility is validated by experiments.According to the system design, and supported by COM, XML, ADO.NET and SQL, etc., an algorithm library and database-based ADMS (acetone data mining system) software prototype system for acetone refining is implemented by the method of mixed programming of C# and Matlab. The system provides reliable instance of data mining industrial application, and is convenient for expansion and development. It lays the foundation of the application of data mining methods in practical production.A conclusion is given in the end of the thesis, as well as the remaining problems to be further researched.