Методика выбора алгоритмов оптимизации живучести энергетических инфраструктур

Abstract

The article considers one of the most difficult tasks of studying the resilience of energy infrastructures – finding effective combinations of measures to increase resilience. To solve this problem, the article describes an approach that considers it as a problem of structural-parametric optimization of energy infrastructures, which is built according to a two- or three-level scheme. The approach described in the article adds another layer to the middle of the above scheme, which checks the efficiency of the selected equipment under extreme conditions created by a given set of large disturbances. The main disadvantage that the approach inherits from the structural-parametric optimization of energy infrastructures is a high computational complexity of the multilevel optimization scheme. However, the unacceptable calculation time can be explained by the selection of inappropriate optimization algorithms. In the papers concerning the structural-parametric optimization of energy infrastructures publ in the literature, the question of comparing optimization algorithms with each other is clearly not raised. Therefore, this article proposes a three-stage methodology for selecting optimization algorithms, according to which, before solving a specific problem of optimizing the resilience of energy infrastructures, first test the algorithms, and then choose the best one based on a multi-criteria analysis of the test results. To apply the methodology, it is necessary to develop a special lightweight version of the task of optimizing resilience and prepare a testbed for organizing and conducting test computational experiments. The application of the methodology is demonstrated by the example of choosing heuristic methods for finding optimal solutions from the PaGMO library used at the external level of the resilience optimization scheme of the Unified Gas Supply System of Russia. In total, five popular evolutionary algorithms were tested, the most suitable of which turned out to be a genetic sorting algorithm without NSGA-II dominance.