Extraction operation know-how from historical operation data - Using characterization method of time series data and data mining method

In these days, it is very difficult to hand down experts' operation know-how to beginner, because of operation technique of a large and highly complex plant and reducing operators. On the other hand, data mining methods (See5, naive bayes, k-nearest neighbor, and so on) has been proposed as knowledge discovering methods from a huge amount of data. See5 outputs decision trees or IF-THEN rules as data mining results. However, See5 cannot recognize data as time series. In this study an extraction method of experts' operation know-how from historical operation data is proposed. Furthermore efficiencies of the proposed method are demonstrated by numerical experiments using a dynamic simulator.