An Application Of The Rough Set In The Data Analsys Of The Alumina Seed Precipitation

With the popularization of the application of automatic monitored control system and real-time database in alumina industry, a vast amount of historical data has been accumulated in alumina production. It has become an urgent need for enterprises to find from this enormous data the rules and laws that can play a guiding role in production and management.Based on rough set, this paper studies and probes some critical problems in the application of the alumina seed precipitation.Aiming at the importance of discretization algorithm in rough set mode-building and the users' perplexity at the great variety of the existing discretization algorithm, the paper sets the assessing standard based on the correction rate of classification prediction, comparing the influences of different discretization algorithm on the classification prediction accuracy in seed precipitation process. Taking the approximate precision as feedback information,it then proposes a better method, that is, a discretization method combining rough set and logical reasoning for data discretization in seed precipitation field in order to bring into being more effective and practical decision rules.Through analyzing the data features in alumina seed precipitation, a rough set model of data analysis in this field has been established and put into use in dealing with industrial data of sandy alumina production by two—stage method,realizing the predicted knowledge discovery process of agglomeration rate and obtaining good predicted results.In view of problems such as the failure of predicting some samples according to the obtained rules in rough set method and low prediction rate in seed precipitation data analysis through neural network, a data analysis model of rough-neural network has been propounded and applied to the analysis of industrial data of certain sandy alumina production by two—stage method, with low mean error of model prediction .This model has been improved to some extent in both time and space complexity due to the removal of some redundant attributes and samples, thus roughly realizing the regulation and prediction of alumina seed precipitation.