Study On Classification Association Rules Mining And Its Application In Complicated Industry Process

Nowadays, technique of automation and database has been widely applied into more and more industrial production systems. Huge numbers of historical process data have been accumulated in their real time databases, in which substantial information concerned with process monitoring and controlling, parameter optimization, product quality, and production management are hidden. All of these provide a great opportunity to the application of data mining technology in practical industrial process.Based on the background of complicated industrial process and the tool of classification association rules mining, several techniques in different phases of data mining process composed of multivariable supervised discretization algorithm, explorative analysis methods, and classification association rules based modeling methods are investigated in this dissertation. In addition, a practical application of data mining technique in a sulphuric acid production plant based on lead sintering machine off-gas is accomplished finally.The main contributions of this dissertation are described as follows,In data preprocessing phase, a novel multivariable supervised discretization (MSD) algorithm is proposed.Considering the features of industrial process data (large numbersof variables, strong coupling, and gigantic data), a multivariable supervised discretization (MSD) algorithm is proposed. MSD can be realized in two levels: first a coarse discretization is carried out (in this step, the interdependency information among all the conditional variables are considered); then a fine discretization is made (in this step, the class information is considered). Through considering the distribution information and class information of data, MSD automates the discretizaition step effectively. To prove its validity and efficiency, a comparative test is also conducted.In modeling phase, three data mining models are proposed.(1) A fuzzy classification association rules mining model (FCARM) is proposed.On the basis of existed classification association rules algorithm, a fuzzy classification association rules mining model (FCARM) is proposed. The new FCARM model has made three contributions: a new distance based definition of support is suggested and applied; MSD discretization method is introduced to automate its discretization tasks; and also the concept of fuzzy sets is inducted to soften the sharp boundaries brought by discretization. In the end, FCARM model is applied into the analysis of historical data from practical aromatic hydrocarbon extraction process.(2) A fuzzyCARs (fuzzy class association rules) based fuzzy system is proposed.There are two obstacles in the construction process of traditional fuzzy system. One is how to partition the input space (that is how to determine the number of discrete zone and the corresponding separating point of each attribute). The other is to avoid the problem of " the curse of dimensionality" caused by the sampling sparseness of high dimensional data. To solve the above two problems, a fuzzyCARs based fuzzy system is proposed, in which the two difficulty problems are solved successfully. And also a comparative experiment is conducted to demonstrate its effectiveness.(3) A novel fuzzy path-query system is proposed.All of the aforementioned systems run on the basis of the analysis of off-line historical data. To satisfy the need of automatic operation of factory, a novel online data mining system named fuzzy path-query system is proposed. It is composed of a rule base system and an intelligent query system. The system bears the following features:(a) Give online evaluation to current operations.(b) Make operational suggestions to current status through searching in its rule base.(c) Answer diversified queries from users (similar operations, the optimal operating, and fault detection)(d) Automatically update its rule baseFinally the proposed fuzzy path-query system is applied into the practical process of paraxylene adsorption and separation.In application phase, a practical industrial data mining project in a WSAsulphuric acid production plant has been fulfilled.Through analysis of historical data from WSA sulphuric acid production plant, the following aims have been reached: the correlations among operational variables and the key factors concerned with the acid concerntation are identified; corresponding visualized results are also provided. In the dissertation, the whole data mining procedures are discussed, which consists of data gathering, data preprocessing, explorative analysis, and application of FCARM model. At present, the software has put into service in the plant.