Research On Association Rules Mining And Its Application To Complex Industrial Process Control

Association rules discovery has emerged as an important problem in knowledge discovery and data mining. As the focus research of association rules mining, level wise search algorithm has been studied widely. In recent years, database technologies have permeated though many monitor systems of complex industrial process. A great deal of industrial data has been collected, which includes useful information about process control, parameter optimizing and administration. These have formed the base of application of .data mining in the integrate intelligent control of the complex industrial process and its optimizing.Three problems have been studied in this paper funded by scientific research project of the State Ministry of Education. Firstly, several classical frequent set discovery algorithms based on level wise framework are discussed. Secondly, the incremental algorithm and discretization of feature are studied for the dynamic and continue data set. Finally, an expert controller and its implement of complex industrial process with help of the association rules mining are introduced.In the study on level wise search algorithm, the structure and character of three basic frequent set discovery algorithms AIS, Apriori and DHP are studied by simulation. A model of level wise search algorithm is given and its time complexity is analyzed respectively on transaction scale, item length and support parameter. The result shows that the time complexity of algorithm is linear with the increment of transaction if the average length of transaction and frequent itemsets is invariable, but it is inefficient to the increament of item average length (including transaction length and frequent itemset length). On the other hand, the level wise search algorithm is analyzed theoretically with level space theory based on partial order. Several properties of level space are summarized and proved. The property of border theory and complexity conclusion of abstract level wise search model are also proved based on the above discuss.In order to handle dynamic data set in database of industrial field, several incremental algorithms including NBIA and FUP are discussed. An improved fast update algorithm with level wise search structure based on FUP algorithm is proposed for the incremental data. This algorithm avoids the needless scanning on the previous database, and can achieve better performance than FUP algorithm. For the data with quantitative attribute, several unsupervised discretization methods of continuous features are discussed. Three simple boolean discretization methods (equal width, equal frequency and cluster) and fuzzy discretization are simulated based on a real population statistical database. Simulation illustrates that the discovery result has dependency relation to the distribution of instance number of feature intervals, cluster method can help to generate more association rules with small quantity of frequent itemsets, and fuzzy discretization can obtain more frequent itemsets and association rules than simple boolean discretization method.Finally, an expert controller's design of sintering temperature of industrial rotary kiln is discussed. In order to solve the measurement problem of sintering temperature, a fuzzy fusion algorithm based on the multi-sensor data is proposed to judge the trend of innersintering temperature of rotary kiln, and some types of sensors in the fusion are also drawn by experience. A new fuzzy quantitative association mining method based on time series is proposed for the acquirement of MIMO expert control knowledge of complex industry process in field database. Subsequently, multi-control strategies in the expert controller are also emphasized for the kiln's complexity. Final application shows that the expert control system can achieve higher stability and better utility, which prove the efficiency of the above methods.