Nonparametric modeling of oxygen-converter processes

Abstract

Preliminary data analysis in the identification of multidimensional discrete–continuous processes is considered. A method is proposed for generating a working sample from an initial training sample consisting of normal operating data. The method somewhat resembles the bootstrap process. In the present case, the process begins with a training sample that reflects the properties of the object to be identified. By means of the proposed method, the unknown stochastic dependence at the limit of definition of the corresponding input–output variables for the object may be automatically derived. The identification of the oxygen-converter process in converter shop 2 at OAO EVRAZ ZSMK is considered in the case with insufficient available information and gaps in the observation sample. The model is based on a new working sample containing both the measurements and data generated by the proposed method. By using the working sample as a training sample, the precision of identification is doubled.