THE «DECISION TREE» METHOD FOR STATISTICAL CONTROL OF PARAMETERS INTERRELATIONS in MULTIDIMENSIONAL INFORMATION FLOWS

Abstract

The purpose of the study is to show the possibilities of machine learning methods for analyzing intra–system connections of multidimensional data. In modern automated process control systems and in particular, in the electric power industry, continuous monitoring of information exchange is necessary. Data flows are random and the parameters transmitted via communication channels have different ranges of variation and dimension. In these conditions, particularly relevant is the development of statistical control methods of such data intra-system connections. Methods. To solve the problem, the machine learning method “decision tree” is used. The possibilities of the approach are demonstrated by analyzing the data interconnections which model a stream containing 27 random parameters with different dimensions. The test was carried out on a sample of 100 to 500 values of each of the parameters. Results. It is shown that statistical control can be carried out without considering the structure of the decision tree itself, according to such indicators as the percentage of links recognition, ranges of splitting of parameter values during classification, the significance of individual parameters (attributes). Conclusions. Since the algorithm does not require a large sample of the analyzed parameters values, statistical control can be carried out in a sliding time window. It is shown that the approach can be used to analyze information exchange in the automated control system.