The Design And Implementation Of KDD System For Industrial Flow Object

With the development of computational intelligence, theory and technology of KDDused to solve engineering problems has become a primary task of its research. So, it has animportant practical meaning that KDD technologies--association rules extraction and objectmodeling are used in the industrial flow object to discovered knowledge in the massdepositional data. This paper is aimed at the application of flow object’s association rulesextraction and object modeling in the cement industry. We adopt the improved Apriorialgorithm and the flexible neural tree model of the structure optimization algorithm, to designand implement a KDD system for industrial flow object by J2EE. The whole system is mainlydivided into two functions: one function module is association rules extraction, the other oneis object modeling, and the original data were collected from the decomposing furnaceproduction link. The first function module is association rules extraction in which we adopt animproved Apriori algorithm--reducing the transaction data. The other function module isobject modeling in which we adopt the PIPE algorithm and SA algorithm to build andoptimize the FNT model. Encoding the entire system use java language and JSP, Servlet，JavaBean，Javascript，JfreeChart and some other techniques. Research process is as follows:(1) Depending on the characteristics of the flow object’s data, clean and transform theparameters. Adopt the methods of time-series match, data discretization and normalization toclean and transform the data sets for the preparation of KDD.(2) Research the improved Apriori algorithm, and implement the function of associationrules extraction by programming.(3) Research the theory of PIPE and SA algorithms, master the structure optimizationalgorithm based on Flexible Neural Trees, and implement the function of object modeling(generate the last generation of the best individual nodes, the parameter of activation function,and the weights of FNT’s branches) by programming.(4) Design structure of the system, and implement the functions of the KDD systembased on J2EE. Association rule algorithm encapsulation is used in the first function module;flexible neural tree model of the structure optimization algorithm encapsulation is used in thesecond function module. (5) Improve the second function module, achieve the other four subfunctions: generateFNT model, export mathematical formula of model, calculate the model output, generatetrend chart.As has been proven in practice, the cement industry accumulated a large amount ofdepositional data and we can excavate association rules and object modeling from these massdepositional data by our KDD system. According to the excavated association rules we canfind the influence relations between parameters of the decomposing furnace, thus, we can putforward the optimized proposals for the auxiliary decision-making of the production ofdecomposing furnace; According to object modeling we can quickly build the FNT model ofthe production process of decomposing furnace, which achieves the automatic filtering of theparameters. Thus we can calculate the decomposing furnace exit temperature predictive valuewith the model’s mathematical formula. When get the calculated predictive value, we can takeit to compare with the actual temperature of decomposing furnace exit. And after comparingwe can stabilize production, optimize control. Because the process of decomposing furnace isone of the most important processes of the cement industry, the two functions of our systemwill play an important role in enhancing production efficiency of the cement industry.