Prediction of temperature field of moderator of heavy- water reactor based on cellular neural network

3/22/2017 2017 - #01 Modelling processes at nuclear facilities

Starkov S.O. Lavrenkov Y.N.

https://doi.org/10.26583/npe.2017.1.09

Calculation of 2D temperature fields in structural elements of installations with internal sources of heat is the necessary condition required for maintaining their safe operation. Methodology is suggested for forecasting moderator temperature in heavy-water reactors with application of cellular neural network. Application of intellectual computational structure allows simplifying the procedure of calculation of moderator temperature in different cross-sections of calandria heating of which is achieved both directly from the surfaces of fuel channels and, as well, as the result of volume heating. The difficulty to apply numerical methods is explained by the computational difficulty of determination of temperature gradients and heat fluxes along the fuel channel, as well as by the complex geometry of fuel assemblies. Neural network modeling facilitates removing potential critical situations by implementing the control of variation of temperature fields during transients, emergency situations and in the estimation of thermal stresses. Application of the technique of cellular neural networks is justified by the specific features of the architecture allowing targeted limitation of communications between neurons. Specific features of cellular networks correspond to the principles of quick tunable transformations which can be efficiently realized on hardware level. Reduction of the number of synoptical communications increases computational efficiency and makes it possible to use this class of networks for processing the high dimensionality data. Structural synthesis of cellular network with optical communications between neurons possessing high efficiency of data processing is examined. The suggested type of communications constitutes the basis for realization of modular intellectual structures consisting of homogeneous fragments. Training the neural network is accomplished using methods of local training with elements of co-evolutional interaction. Presence of powerful 32-bit microcontroller in each neural core justifies the above strategy despite associated high memory requirements. The suggested modular organization efficiently combines local training of neural cores with global training of the whole network. Comparison of results of forecasting with behavior of the reactor model demonstrated the efficiency of reconstructive analysis of complex systems with application of reconfigurable cellular neural networks, structure of which can be optimized for the specific calculational task. Modularity of the structure allows simultaneously constructing large number of networks on the same electronic chip which makes it possible to forecast moderator temperature fields in different sections of calandria and constitutes the basis for constructing 3D models.

References

1. Cacuci Dan Gabriel. Handbook of Nuclear Engineering. Springer US, 2010. 3574 p. DOI: 10.1007/978-0-387-98149-9.
2. Kenneth D Kok. Nuclear Engineering Handbook. CRC Press, 2009. 786 p. DOI: 10.1201/9781420053913-p1.
3. The Essential CANDU, A Textbook on the CANDU Nuclear Power Plant Technology, Editor-in-Chief Wm. J. Garland, University Network of Excellence in Nuclear Engineering (UNENE), ISBN 0-9730040.
4. Leon O. Chua, Tamas Roska. Cellular neural networks and visual computing: foundations and applications. Cambridge University Press New York, NY, USA, 2002. 396 p.
5. Vasiliev A.N., Tarkhov D.A. Principy i tekhnika nejrosetevogo modelirovaniya [Principles and techniques of neural network modeling]. Saint Petersburg. Nestor-Istoriya Publ., 2014. 218 p. (in Russian).
6. Shustov M.A. Skhemotekhnika. 500 ustrojstv na analogovykh mikroskhemakh [Circuitry. 500 analog circuits devices]. Saint Petersburg. Nauka i Tekhnika Publ., 2013. 352 p. (in Russian).
7. Juan R. Rabunal, Julian Dorado. Artificial Neural Networks in Real-Life Applications. IGI Global, 2005. 394 p. DOI: 10.4018/978-1-59140-902-1.
8. David Crecraft, Stephen Gergely. Analog Electronics: Circuits, Systems and Signal Processing. 1st Edition. Butterworth-Heinemann, 2002. 425 p.
9. Dorogov A.Yu. Teoriya i proektirovanie bystrykh perestraivaemykh preobrazovanij i slabosvyazannykh nejronnykh setej [Theory and Design of quick tunable transformations and loosely coupled neural networks]. Saint Petersburg. Politekhnika Publ., 2014. 328 p. (in Russian).
10. Simon Haykin. Neural networks. A Comprehensive Foundation. Second Edition. Prentice Hall, Inc., 1999. 1104 p.
11. Kashchenko S.A., Maiorov V.V. Modeli volnovoj pamyati [Wave memory models]. Moscow. Knizhnyi dom «LIBROKOM» Publ., 2014. 288 p. (in Russian).
12. Graupe Daniel. Principles of Artificial Neural Networks (Advanced Series in Circuits and Systems), 2-nd Edition, World Scientific Pub Co Inc, 2007. 238 p.
13. Nicholls John G., Martin A. Robert, Fuchs Paul A., Brown David A., Diamond Mathew E., Weisblat David. From Neuron to Brain, Fifth Edition. Sinauer Associates, Inc., 2011. 621 p.
14. Barsky A.B. Logicheskie nejronnye seti: uchebhoe posobie [Logical neural networks: schoolbook]. Moscow. Internet-Universitet Informatsionnykh Tekhnologiy; BINOM, Laboratoriya znaniy Publ., 2012. 352 p. (in Russian).
15. Osovsky S. Nejronnye seti dlya obrabotki informacii [Neural network for information processing]. Moscow. Finansy i statistika Publ., 2002. 344 p. (in Russian).
16. Madan M. Gupta, Liang Jin, and Noriyasu Homma. Static and Dynamic Neural Networks: From Fundamentals to Advanced Theory. ISBN: 978-0-471-21948-4, 752 pages, April 2003, Wiley-IEEE Press.
17. Terekhov V.A. Nejrosetevye sistemy upravleniya: uchebnoe posobie dlya vuzov [Neural network control systems: textbook for high schools]. Moscow. Vysshaya shkola Publ., 2002. 183 p. (in Russian).
18. Greshilov A.A. Matematicheskie metody prinyatiya reshenij: uchebnoe posobie s raschyotnymi programmami na opticheskom diske [Mathematical methods of decision-making: schoolbook. 2-nd iss., with modifications and amendments]. Moscow. MGTU im. N.E. Baumana Publ., 2014. 647 p. (in Russian).

cellular neural networks heavy water reactor oscillating fractal network optical neurons forecasting temperature fields the heating medium neutron moderator stochastic algorithms for neural networks training

Link for citing the article: Starkov S.O., Lavrenkov Y.N. Prediction of temperature field of moderator of heavy- water reactor based on cellular neural network. Izvestiya vuzov. Yadernaya Energetika. 2017, no. 1, pp. 94-106; DOI: https://doi.org/10.26583/npe.2017.1.09 (in Russian).