Diagnostics of in-core neutron monitoring system based on artificial neural network
DOI:
https://doi.org/10.15276/opu.2.58.2019.04Keywords:
self-powered neutron detector, neutron flux measuring channel, in-core monitoring system, neural network, system diagnostics, signal reliabilityAbstract
One of the tasks of the diagnostics of in-core neutron control is the task of diagnostics of self-powered neutron detectors, namely the determination of reliability of self-powered neutron detectors signal, that is, the detection of the fault self-powered neutron detectors SPND. It is important to be able to estimate the power density in the fuel assembly with this self-powered neutron detector. The article discusses the features of the restoration of the self-powered neutron detectors signal based on the application of neural network technology. The restoration of the self-powered neutron detector signal means a model-based restoration of a signal that is not available due to physical damage of detector. A trained neural network, based on the monitoring of input information, can with a high degree of accuracy predict the appearance of defects in the equipment and assess the degree of its technical condition. The neural networks of three different architectures were considered: without hidden layers, with one hidden layer and two hidden layers. As the input of the neural network were taken self-powered neutron detectors signals from the various number of neutron flux measuring channels – from 3 to 63. As the output of the neural network were taken self-powered neutron detectors signals that were chosen for prediction. The simulation was carried out for different self-powered neutron detectors, both in terms of year of use and location in the core, as well as for different power units and fuel campaigns (26th and 27th fuel campaigns of ZNPP-5, 27th and 28th fuel campaigns of KhNPP-1, 11th and 12th fuel campaigns of KhNPP-2). The influence of the number of input signals, as well as the effect of the number of hidden layers on the error of the determination of the output signal, were investigated. A comparison of neuronal training algorithms (Levenberg-Marquardt and L-BFGS) was carried out. It is shown the importance of choosing such input signals for the neural network, which determine the nature of the output signal most. It is shown that the restoration of self-powered neutron detectors signals is possible with an error not more than 2%, provided neural network learning on a wide range of data, which allows to control the energy distribution in the fuel assembly with fault self-powered neutron detector.
Downloads
References
Саунин Ю.В., Добротворский А.Н., Семенихин А.В. Специализированное программное обеспе-чение для проведения комплексных испытаний системы внутриреакторного контроля реакторов ВВЭР. Обеспечение безопасности АЭС с ВВЭР. сб. тезисов докладов VI Междунар. научн.-технич. конф. Подольск. ОКБ Гидропресс, 26-29 мая 2009 г. 105 c.
Семенихин А.В., Саунин Ю.В., Жук М.М. Опробование системы диагностики входной инфор-мации СВРК на энергоблоке №1 Нововоронежской АЭС-2. Известия вузов. Ядерная энергетика. 2017. № 3. С. 88–95.
Messai A., Mellit A., Abdellani I., Massi P.A. On-line fault detection of a fuel rod temperature meas-urement sensor in a nuclear reactor core using ANNs. Progress in Nuclear Energy. March 2015. Vol. 79. P. 8–21.
Pirouzmand A., Dehdashti M.K. Estimation of relative power distribution and power peaking factor in a VVER-1000 reactor core using artificial neural networks. Progress in Nuclear Energy. November 2015. Vol. 85. P. 17–27.
Борисенко В.І., Горанчук В.В. Використання технології нейронних мереж для відновлення сиг-налів детекторів, які вийшли з ладу. Проблеми безпеки атомних станцій і Чорнобиля. Чорно-биль. 2018. Вип. 30. С. 17–23.
Haykin S. Neural networks and learning machines. Third Edition. Upper Saddle River. New Jersey : Pearson Education, 2009. 936 p.
ALGLIB-C++/C# numerical analysis library. URL: www.alglib.net (дата звернення 10.04.2019).
