Управление безопасностью критически важного объекта теплоэнергетической отрасли с использованием нейронного сетевого моделирования
Аннотация
В данной магистерской диссертации представлены алгоритм реализации методики для управления безопасностью критически важного объекта – критически важная структура, состоящая из нейросети-управления и нейросети-модуля, определяющих вероятность отказа отдельного участка мазутопровода и всей системы в целом. Для обучения нейросети-управления были собраны статистические данные по отказам мазутопровода Приморской ГРЭС, состоящие из 4470 наблюдений. В качестве данных для нейросети-модуля смоделирован набор данных из 1240 наблюдений, характеризующих участки мазутопровода с наибольшей вероятностью отказа. Для написания использованы программный пакет Statistica 10.0 и среда программирования на Python – PyCharm.
This master's dissertation presents an algorithm for implementing a methodology for managing the safety of a critically important facility - a critical structure consisting of a neural network control and a neural network module that determine the probability of failure of a separate section of the fuel oil pipeline and the entire system as a whole. To train the neural control network, statistical data on failures of the fuel oil pipeline at the Primorskaya State District Power Plant was collected, consisting of 4470 observations. As data for the neural network module, a data set of 1240 observations was simulated, characterizing sections of the fuel oil pipeline with the highest probability of failure. The Statistica 10.0 software package and the Python programming environment – PyCharm – were used for writing.
This master's dissertation presents an algorithm for implementing a methodology for managing the safety of a critically important facility - a critical structure consisting of a neural network control and a neural network module that determine the probability of failure of a separate section of the fuel oil pipeline and the entire system as a whole. To train the neural control network, statistical data on failures of the fuel oil pipeline at the Primorskaya State District Power Plant was collected, consisting of 4470 observations. As data for the neural network module, a data set of 1240 observations was simulated, characterizing sections of the fuel oil pipeline with the highest probability of failure. The Statistica 10.0 software package and the Python programming environment – PyCharm – were used for writing.