МЕТОД ОЦЕНКИ ДОСТОВЕРНОСТИ КОЛИЧЕСТВЕННОГО АНАЛИЗА РИСКА НА ОБЪЕКТАХ НЕФТЕГАЗОВОЙ ОТРАСЛИ

Abstract

Despite the existing arsenal of various methods of quantitative risk analysis both in Russian and international practice, problems arise related to the assessment of the reliability of the results obtained during their practical application, including at oil and gas facilities. The question remains to what extent the decision-maker can trust the results. The article considers the problem of assessing the reliability of quantitative risk analysis at oil and gas facilities. Existing methods for estimating reliability are investigated. It is proposed to use an approach based on ensuring the quality of the process of risk analysis itself. To increase the objectivity in assessing the reliability of the results of a risk analysis, a formal quantitative method was proposed. The article introduces 5 criteria that ensure the quality of the process of risk analysis at oil and gas facilities. A system of rules for coding the values of each of the basic criteria into three discrete qualitative levels was developed. The solution of the task was accomplished by constructing a classifier in which the reliability index of a quantitative risk analysis of oil and gas industry objects is a function of the values of the basic criteria. The reliability of the risk analysis was evaluated on the basis of a naive Bayesian classifier that takes into account the values of the five basic criteria in the evaluation framework. The results of the classifier work are based on a variety of training data that were previously evaluated by experts. The article suggests an approach to the assessment of the quality of the classifier itself, based on a cross-checking with successive exclusion of one copy of the training data. The merits of using the naive Bayesian classifier for assessing the reliability of quantitative risk analysis in oil and gas industry objects include the fact that the classification is carried out quite easily and quickly, surpasses many other algorithms, and requires a smaller amount of training data. A naive Bayesian classifier works very well with categorical features, which is exactly what is reflected in this article.