EVALUATING THE EFFECTIVENESS OF STATISTICAL PROCESSING PROCEDURES TO IMPROVE THE RELIABILITY OF THE RESULTS OF IN-SERVICE CONTROL OF THE CONDITION OF TRANSFORMER OILS

Using the results of periodic transformer oil condition tests on 91 transformers, 110 kV from four regions of Ukraine, we analyzed the effectiveness of statistical processing procedures to reduce the heterogeneity of data and highlight gross outliers from the results of operational control of transformer oil condition. In the process of researches we analyzed the efficiency of nonparametric two-sample Wilcoxon criterion for equality of mean values, parametric Fisher-Snedekor criterion for equality of variance and parametric criterion for significance of pair linear correlation of indicators and duration of operation, as well as their combinations. These criteria were used to test the results of physical and chemical analysis of transformer oils by 11 indicators and for the results of analysis of gases dissolved in oil. As a result of sequential testing, eight data sets for each indicator were formed and used to evaluate the effectiveness of the analyzed statistical criteria and their combinations. Evaluation of the effectiveness of the analyzed statistical criteria and their combinations was carried out on the basis of analysis of the closeness of the correlation relationship between the indicators with the subsequent assessment of the information significance for each of the indicators. As a result, we proposed a method for selecting the optimal number of the most informative indicators of quality, based on statistical analysis of the results of periodic tests of transformer oil. As an example of practical use of the suggested method a regression model to estimate oil aging degree has been developed and its adequacy has been proved.