Shafronenko A. Methods of dynamic intellectual analysis for data with missing values

The dissertation is devoted to the development of methods of dynamic intellectual analysis for distorting data in the tables "object - property" and the time series for data recovery in on-line mode. An adaptive neuro-fuzzy system that allows to solve the problem of restoring missing values in on-line mode with correction of recovered elements and centroids of clusters was proposed. A neuro-fuzzy method for recovery of distorted data and clustering based on neuro-fuzzy Kohonen maps, that allows to process the data in on-line mode and provide operation with overlapping class was proposed. Have got further development of methods for data clustering data with missing values, based on recurrent optimization of objective functions in special type whose observations are replaced by estimates obtained in the process of solving the problem; methods of adaptive fuzzy clustering of data with missing values that allow to process information using on strategy of nearest prototype-centroid in on-line mode. The problem of restoration of distorted data provided by the x-ray plant and service center using the proposed methods, making it possible speed up recovery hardware that is out of order, and early identification of potential problem.