The application of physics-informed radial basis function networks for solving boundary value problems describing piecewise homogeneous media is considered. A meshless algorithm for solving boundary value problems for piecewise homogeneous media is proposed, using the solution of individual problems for each region with different properties of the medium, and the conditions for the conjugation of media. To solve the coefficient inverse problem of determining the properties of a piecewise inhomogeneous medium, a parametric optimization algorithm is proposed that uses separate networks to approximate the properties of the medium and solve the direct problem. To train networks, a fast algorithm developed by the authors based on the Levenberg – Marquardt method was applied. The work of the proposed algorithms is demonstrated on model problems.
partial differential equations, piecewise homogeneous medium, inverse problems, physics informed neural-networks, radial basis function networks, neural network learning, Levenberg-Marquardt method.
Стр. 91-99.
References
1. Lagaris, I.E., A. Likas, D.I. Fotiadis. 1998. Artificial neural networks for solving ordinary and partial differential equations. IEEE Transactions on Neural Networks. 9(5): 987–1000. doi: 10.1109/72.712178.
2. Yadav, N., Yadav, A., Kumar, M. 2015. An Introduction to Neural Network Methods for Differential Equations. Dordrecht: Springer. 115 p.
3. Cybenko, G. 1989. Approximation by Superposition of a Sigmoidal Function. Mathematics of Control, Signals and Systems. 2: 303–314. doi: 0.1007/BF02551274.
4. Hornik, К., Stinchcombe, M., White, H. 1989. Multilayer feedforward networks are universal approximators. Neural networks. 2(5): 359–366. doi: 10.1016/0893-6080(89)90020-8.
5. Kurkova, V. 1992. Kolmogorov’s theorem and multilayer neural networks. Neural networks. 5(3): 501-506. doi: 0893-6080(92)90012-8
6. Hanin, B. 2019. Universal Function Approximation by Deep Neural Nets with Bounded Width and ReLU Activations. Mathematics. 7(10), 992: 1–9. doi: 10.3390/math7100992.
7. Raissi, M., Perdikaris, P., Karniadakis, G.E. 2019. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving non-linear partial differential equations. Journal of Computational Physics. 378: 686–707. doi: 10.1016/j.jcp.2018.10.045.
8. Bavdin, A.G., Pearlmutter, B.A., Radul, A.A., Siskind, J.M. 2018. Automatic Differentiation in Machine Learning: a Survey. Journal of Machine Learning Research. 18 (1): 1–43. Available at: https://jmlr.org/papers/volume18/17-468/17-468.pdf (accessed July 08, 2021).
9. Avrutskiy, V.I. 2021. Enhancing Function Approximation Abilities of Neural Networks by Training Derivatives. IEEE Transactions on Neural Networks and Learning Systems. 32(2): 916 – 924. doi: 10.1109/TNNLS.2020.2979706.
10. Aggarwal, C.C. 2018. Neural Networks and Deep Learning. Cham: Springer. 520 p.
11. Buhmann, M.D. 2004. Radial Basis Functions: Theory and Implementations. Cambridge Cambridge: University Press. 259 p.
12. Kansa, E.J. 1990. Multiquadrics — A scattered data approximation scheme with applications to computational fluid-dynamics — I surface approximations and partial derivative estimates. Computers&Mathematics with Applications. 19(8–9): 127–145. doi: 10.1016/0898-1221(90)90270-T.
13. Kansa, E.J. 1990. Multiquadrics — A scattered data approximation scheme with applications to computational fluid-dynamics — II solutions to parabolic, hyperbolic and elliptic partial differential equations. Computers&Mathematics with Applications. 19(8–9): 147–161. doi: 10.1016/0898-1221(90)90271-K.
14. Jianyu, L., Siwei, L., Yingjian, Q. 2003. Numerical solution of elliptic partial differential equation by growing radial basis function neural networks. Neural Networks. 16(5–6): 729–734. doi: 10.1016/S0893-6080(03)00083-2.
15. Chen, W., Fu, Z.J. 2014. Recent Advances in Radial Basis Function Collocation Methods. Cham: Springer. 90 p.
16. Gorbachenko, V.I., Zhukov, M.V. 2017. Solving boundary value problems of mathematical physics using radial basis function networks. Computational Mathematics and Mathematical Physics. 57(1): 145–155. doi: 10.1134/S0965542517010079.
17. Gorbachenko, V.I., Alqezweeni, M.M. 2019. Learning Radial Basis Functions Networks in Solving Boundary Value Problems.: 2019 International Russian Automation Conference. Sochi, Russia September 8-14, 2019: 1–6.
18. Marquardt, D.W. 1963. An Algorithm for Least-Squares Estimation of Nonlinear Parameters. Journal of the Society for Industrial and Applied Mathematics. 11(2): 431–441. doi: 10.1137/0111030.
19. Islam, M.R., Abou-Kassem, J.H., Farouq-Ali, S.M. 2020. Petroleum Reservoir Simulation: The Engineering Approach. Houston: Gulf Professional Publishing. 526 p.
20. Carslaw, H.S. 2018. Introduction to the Mathematical Theory of the Conduction of Heat in Solids. Blacweel: Franklin Classic. 286 p.
21. Anderson, M.P., Woessner, W.W., Hunt, R.J. 2015. Applied Groundwater Modeling: Simulation of Flow and Advective Transport. Cambridge: Academic Press. 564 p.
22. Samarskii, A.A., Vabishchevich, P.N. 1996. The Finite Difference Methodology, Volume 2, Computational Heat Transfer. New York: Wiley. 432 p.
23. Ngo-Cong, D., Tien, C.M.T., Nguyen-Ky, T., An-Vo, D.-A., Mai-Duy, N., Strunin, D.V., Tran-Cong, T. 2017. A generalised finite difference scheme based on compact integrated radial basis function for flow in heterogeneous soils. International Journal for Numerical Methods in Flu-ids. 85(7): 404-429. doi: 10.1002/fld.4386.
24. Chen, J.-S., Wang, L., Hu, H.-Y., Chi, S.-W. 2009. Sub-domain radial basis collocation method for heterogeneous media. International journal for numerical methods in en-gineering. 80(2): 163–190. doi: 10.1002/nme.2624.
25. Stenkin, D.A., Gorbachenko, V.I. 2021. Solving Equations Describing Processes in a Piecewise Homogeneous Medium on Radial Basis Functions Networks. Kryzhanovsky, B., Dunin-Barkowski, W., Redko, V., Tiumentsev, Y. (eds) Advances in Neural Computation, Machine Learning, and Cognitive Research IV. NEUROINFORMATICS 2020. Studies in Computational Intelligence. Cham: Springer. 925: 412–419. doi: 0.1007/978-3-030-60577-3_49.
26. Pakravan, S., Mistani, P.A., Aragon-Calvo, M.A., Gibou, F. Solving inverse-PDE problems with physics-aware neural networks. Available at: https://arxiv.org/abs/2001.03608 (accessed July 08, 2021).
27. Samarskii, A.A., Vabishchevich, P.N. 2007. Numerical Methods for Solving Inverse Problems of Mathematical Physics. Berlin: Walter de Gruyter. 454 p.
28. Morozov, V.A. 1984. Methods for Solving Incorrectly Posed Problems. New York: Springe. 280 p.
