БАЙЕСОВСКИЙ ПОДХОД К ПОВЫШЕНИЮ ДОСТОВЕРНОСТИ КОНТРОЛЯ КАЧЕСТВА ВОД

Increased variability and, at the same time, a reduced frequency of selective measurements of controlled indicators of natural waters increase the probability of erroneous evaluation of their quality. The task is to increase the reliability of such an assessment by analyzing arrays of new data in conjunction with data accumulated in previous periods. To do this, a Bayesian approach was modified using the uniformity measure of the combined data. It is shown that in the latter case the combined estimate shifts from the Bayesian one to the maximum likelihood estimate from the newly obtained experimental data, thus "forgetting" the obsolete data. At the same time, the 90% confidence interval, in which the true values of the monitored indicators are concluded, is narrowed, which increases the reliability of the probabilistic assessment of water quality. The proposed approach is illustrated by the example of a universal nonparametric method for estimating the probability of the concentration of a certain pollutant in compliance with the requirements as the most common indicator of water quality. The example is brought to specific numerical values, allowing both to compare the classical and modified Bayesian approach, and to give recommendations on the rational use of the latter. The proposed approach can find wide application in the problems of analysis of statistical quality indicators in various subject areas with a shortage of experimental data. Keywords : water quality control, probabilistic estimation, Bayesian approach, mixture of distributions, maximum likelihood function  DOI: http://dx.doi.org/10.15826/analitika.2018.22.3.001 (Russian) O.M. Rozental’ 1 , L.N. Aleksandrovskaya 2 , A.V. Kirillin 2 1 Institute of water problems of RAS, ul. Gubkina, 3, Moscow, 125993,  Russian Federation 2 Moscow Aviation Institute (MAI), Volokolamskoe shosse, 4, Moscow, 125080, Russian Federation