Abstract
Deep learning-based surrogate modeling is becoming a promising approach for learning and simulating dynamical systems. However, deep-learning methods find it very challenging to learn stiff dynamics. In this paper, we develop DAE-PINN, the first effective physics-informed deep-learning framework for learning and simulating the solution trajectories of nonlinear differential-algebraic equations (DAE). DAEs are used to model complex engineering systems, e.g., power networks, and present a “form” of infinite stiffness, which makes learning their solution trajectories challenging. Our DAE-PINN bases its effectiveness on the synergy between implicit Runge–Kutta time-stepping schemes (designed specifically for solving DAEs) and physics-informed neural networks (PINN) (deep neural networks that we train to satisfy the dynamics of the underlying problem). Furthermore, our framework (i) enforces the neural network to satisfy the DAEs as (approximate) hard constraints using a penalty-based method and (ii) enables simulating DAEs for long-time horizons. We showcase the effectiveness and accuracy of DAE-PINN by learning the solution trajectories of a three-bus power network.
Similar content being viewed by others
Availability of data and materials
The data used for training the proposed methods of this article can be generated using the code available in Github (https://github.com/cmoyacal/DAE-PINNs).
Code availability
The code for this article is available in GitHub (https://github.com/cmoyacal/DAE-PINNs).
Notes
Observe that our proposed framework does not require supervision, i.e., it does not require to know target values of the solution trajectory.
References
Alvarado F, Oren S (2002) Transmission system operation and interconnection. National transmission grid study–Issue papers, pp A1–A35
Aristidou P, Fabozzi D, Van Cutsem T (2013) Dynamic simulation of large-scale power systems using a parallel schur-complement-based decomposition method. IEEE Trans Parallel Distrib Syst 25(10):2561–2570
Baker N, Alexander F, Bremer T et al (2019) Workshop report on basic research needs for scientific machine learning: Core technologies for artificial intelligence. Tech. rep, USDOE Office of Science (SC), Washington, DC (United States)
Brunton SL, Proctor JL, Kutz JN (2016) Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proc Natl Acad Sci 113(15):3932–3937
Brunton SL, Proctor JL, Kutz JN (2016) Sparse identification of nonlinear dynamics with control (sindyc). IFAC-PapersOnLine 49(18):710–715
Chao H (2002) Implementation of parallel-in-time Newton method for transient stability analysis on a message passing multicomputer. In: Proceedings of international conference on power system technology. IEEE, pp 1239–1243
Chen J, Li K, Bilal K et al (2018) A bi-layered parallel training architecture for large-scale convolutional neural networks. IEEE Trans Parallel Distrib Syst 30(5):965–976
Chen J, Li K, Philip SY (2021) Privacy-preserving deep learning model for decentralized vanets using fully homomorphic encryption and blockchain. IEEE Trans Intell Transp Syst
Chen RT, Rubanova Y, Bettencourt J et al (2018) Neural ordinary differential equations. arXiv preprint arXiv:1806.07366
Chiang HD (2011) Direct methods for stability analysis of electric power systems: theoretical foundation, BCU methodologies, and applications. Wiley, New York
Chiang HD, Wu FF, Varaiya PP (1994) A bcu method for direct analysis of power system transient stability. IEEE Trans Power Syst 9(3):1194–1208
Chung E, Leung WT, Pun SM et al (2021) A multi-stage deep learning based algorithm for multiscale model reduction. J Comput Appl Math 394(113):506
Fabozzi D, Chieh AS, Haut B et al (2013) Accelerated and localized newton schemes for faster dynamic simulation of large power systems. IEEE Trans Power Syst 28(4):4936–4947
Gupta A, Gurrala G, Sastry P (2018) An online power system stability monitoring system using convolutional neural networks. IEEE Trans Power Syst 34(2):864–872
Gurrala G, Dimitrovski A, Pannala S et al (2015) Parareal in time for fast power system dynamic simulations. IEEE Trans Power Syst 31(3):1820–1830
Hairer E, Lubich C, Roche M (2006) The numerical solution of differential-algebraic systems by Runge–Kutta methods, vol 1409. Springer, Berlin
He M, Zhang J, Vittal V (2013) Robust online dynamic security assessment using adaptive ensemble decision-tree learning. IEEE Trans Power Syst 28(4):4089–4098
Hiskens IA, Hill DJ (1989) Energy functions, transient stability and voltage behaviour in power systems with nonlinear loads. IEEE Trans Power Syst 4(4):1525–1533
Iserles A (2009) A first course in the numerical analysis of differential equations, vol 44. Cambridge University Press, New York
James J, Hill DJ, Lam AY et al (2017) Intelligent time-adaptive transient stability assessment system. IEEE Trans Power Syst 33(1):1049–1058
Ji W, Qiu W, Shi Z et al (2020) Stiff-pinn: physics-informed neural network for stiff chemical kinetics. arXiv preprint arXiv:2011.04520
Karniadakis GE, Kevrekidis IG, Lu L et al (2021) Physics-informed machine learning. Nat Rev Phys 3(6):422–440
Kim S, Ji W, Deng S et al (2021) Stiff neural ordinary differential equations. arXiv preprint arXiv:2103.15341
Kingma DP, Ba J (2014) Adam: a method for stochastic optimization. arXiv preprint arXiv:1412.6980
Kundur P (2007) Power system stability. Power system stability and control pp 7–1
LeCun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 521(7553):436–444
Li J, Stinis P (2019) Model reduction for a power grid model. arXiv preprint arXiv:1912.12163
Li J, Yue M, Zhao Y et al (2020) Machine-learning-based online transient analysis via iterative computation of generator dynamics. In: 2020 IEEE international conference on communications, control, and computing technologies for smart grids (SmartGridComm). IEEE, pp 1–6
Lu L, Jin P, Pang G et al (2021) Learning nonlinear operators via deeponet based on the universal approximation theorem of operators. Nat Mach Intell 3(3):218–229
Lu L, Meng X, Mao Z et al (2021) Deepxde: a deep learning library for solving differential equations. SIAM Rev 63(1):208–228
Lu L, Pestourie R, Yao W et al (2021c) Physics-informed neural networks with hard constraints for inverse design. arXiv preprint arXiv:2102.04626
Luenberger DG (1973) Introduction to linear and nonlinear programming, vol 28. Addison-Wesley, Reading
Milano F (2010) Power system modelling and scripting. Springer, Berlin
Misyris GS, Venzke A, Chatzivasileiadis S (2020) Physics-informed neural networks for power systems. In: 2020 IEEE power & energy society general meeting (PESGM). IEEE, pp 1–5
Moya C, Zhang S, Yue M et al (2022) Deeponet-grid-uq: A trustworthy deep operator framework for predicting the power grid’s post-fault trajectories. arXiv preprint arXiv:2202.07176
Pai M, Padiyar K, Radhakrishna C (1981) Transient stability analysis of multi-machine ac/dc power systems via energy-function method. IEEE Trans Power Appar Syst 12:5027–5035
Park B, Sun K, Dimitrovski A et al (2021) Examination of semi-analytical solution methods in the coarse operator of parareal algorithm for power system simulation. IEEE Trans Power Syst 36(6):5068–5080
Qin T, Wu K, Xiu D (2019) Data driven governing equations approximation using deep neural networks. J Comput Phys 395:620–635
Qin T, Chen Z, Jakeman JD et al (2021) Data-driven learning of nonautonomous systems. SIAM J Sci Comput 43(3):A1607–A1624
Raissi M, Perdikaris P, Karniadakis GE (2018) Multistep neural networks for data-driven discovery of nonlinear dynamical systems. arXiv preprint arXiv:1801.01236
Raissi M, Perdikaris P, Karniadakis GE (2019) Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. J Comput Phys 378:686–707
Roche M (1989) Implicit Runge–Kutta methods for differential algebraic equations. SIAM J Numer Anal 26(4):963–975
Roth J, Barajas-Solano DA, Stinis P et al (2021) A kinetic Monte Carlo approach for simulating cascading transmission line failure. SIAM J Multiscale Model Simul 19(1)
Rudin W et al (1976) Principles of mathematical analysis, vol 3. McGraw-Hill, New York
Schaeffer H (2017) Learning partial differential equations via data discovery and sparse optimization. Proc R Soc A Math Phys Eng Sci 473(2197):20160446
Schainker R, Miller P, Dubbelday W et al (2006) Real-time dynamic security assessment: fast simulation and modeling applied to emergency outage security of the electric grid. IEEE Power Energ Mag 4(2):51–58
Shu J, Xue W, Zheng W (2005) A parallel transient stability simulation for power systems. IEEE Trans Power Syst 20(4):1709–1717
Stott B (1979) Power system dynamic response calculations. Proc IEEE 67(2):219–241
Strogatz SH (2018) Nonlinear dynamics and chaos: with applications to physics, biology, chemistry, and engineering. CRC Press, Boca Raton
Tomim MA, Marti JR, Wang L (2009) Parallel solution of large power system networks using the multi-area thévenin equivalents (mate) algorithm. Int J Electr Power Energy Syst 31(9):497–503
Varaiya P, Wu FF, Chen RL (1985) Direct methods for transient stability analysis of power systems: Recent results. Proc IEEE 73(12):1703–1715
Wang S, Teng Y, Perdikaris P (2020) Understanding and mitigating gradient pathologies in physics-informed neural networks. arXiv preprint arXiv:2001.04536
Wanner G, Hairer E (1996) Solving ordinary differential equations II, vol 375. Springer Berlin Heidelberg, New York
Yazdani A, Lu L, Raissi M et al (2020) Systems biology informed deep learning for inferring parameters and hidden dynamics. PLoS Comput Biol 16(11):e1007575
Zhao T, Yue M, Wang J (2022) Structure-informed graph learning of networked dependencies for online prediction of power system transient dynamics. IEEE Trans Power Syst
Zheng H, DeMarco CL (2010) A bi-stable branch model for energy-based cascading failure analysis in power systems. In: North American power symposium 2010. IEEE, pp 1–7
Zheng Y, Hu C, Lin G et al (2022) Glassoformer: a query-sparse transformer for post-fault power grid voltage prediction. In: ICASSP 2022–2022 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE, pp 3968–3972
Zhou H, Zhang S, Peng J et al (2020) Informer: beyond efficient transformer for long sequence time-series forecasting. arXiv preprint arXiv:2012.07436
Zhu L, Hill DJ, Lu C (2019) Hierarchical deep learning machine for power system online transient stability prediction. IEEE Trans Power Syst 35(3):2399–2411
Funding
This work was supported by the National Science Foundation (DMS-1555072, DMS-1736364, DMS-2053746, and DMS-2134209), and Brookhaven National Laboratory Subcontract 382247, and U.S. Department of Energy (DOE) Office of Science Advanced Scientific Computing Research program DE-SC0021142.
Author information
Authors and Affiliations
Contributions
CM: Conceptualization, Methodology, Investigation, Formal Analysis, Writing—Original Draft. GL: Conceptualization, Supervision, Writing—Review, Editing, Funding Acquisition.
Corresponding author
Ethics declarations
Ethical approval
Not applicable.
Consent to participate
Not applicable.
Consent for publication
Not applicable.
Conflict of interest
The authors have no competing interests to declare that are relevant to the content of this article.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix A: The Van der Pol system
Appendix A: The Van der Pol system
In this appendix, we consider the classical Van der Pol (VDP) system [16, 49]:
where \((x, \dot{x})\) is the state vector and \(\mu\) is the system’s parameter. To facilitate our analysis, let us transform the VDP system as follows. Consider the following identity:
If we let \(\varphi (x) := \frac{1}{3}x^3 - x\) and \(z = \dot{x} + \mu \varphi (x)\), the above identity and the VDP system imply
If we also let \(y:= \frac{z}{\mu }\), then we can write the VDP system as follows:
Now suppose the initial condition \((x_0,y_0)\) is not too close to the line described by the function \(y = \varphi (x)\), i.e., assume \(y - \varphi (x) \sim O(1)\). Then Eq. (A1) implies that \(|\dot{x} |\sim O(\mu )\) and Eq. (A2) implies that \(|\dot{y} |\sim O(\mu ^{-1})\).
1.1 A.1 Non-stiff Van Der Pol system
It is well known [49] that whenever \(\mu = 1\), the Van Der Pol system is non-stiff, i.e., \(\vert \dot{x} |, \vert \dot{y} |\sim O(1)\), the time-scales are similar. Figure 8 presents the non-stiff solution trajectories for the VDP system with initial condition \((x_0, y_0) = (2,0)\). Figure 8 also presents the predicted trajectory using a modified version of DAE-PINN. Note that this version of DAE-PINN is trained to satisfy Eqs. (5a) and (5c) for the VDP system. The results illustrate that the proposed DAE-PINN can easily simulate the dynamic response of the non-stiff VDP system.
1.2 A.2 Increasing stiffness
To increase the stiffness of the VDP system, one must increase the value of the parameter \(\mu\). Clearly, \(|\dot{x} |\sim O(\mu ) \gg 1\) and \(|\dot{y} |\sim O(\mu ^{-1}) \ll 1\) when \(\mu\) increases and is much bigger than one. In particular, the velocity in the horizontal direction is considerable, while the velocity in the vertical direction is minimal [49]. These two widely separated time scales are a distinctive property of stiff systems. Figure 9 illustrates the solution trajectories for the VDP system with initial condition \((x_0, y_0) = (2.0, 0.0)\) and parameter \(\mu \in \{10, 100, 1000\}\). Note that we can observe both time scales in the graph of the x(t) trajectory. Figure 9 also illustrates the prediction of a modified DAE-PINN. This DAE-PINN can effectively predict/simulate the trajectory even for the very stiff VDP system, i.e., when \(\mu = 1000\).
1.3 A.3 Differential algebraic equations and infinite stiffness
If we let the parameter increase to infinity, i.e., if we let \(\mu \rightarrow \infty\), then the stiffness increases to infinity. Then, the differential equations of the VDP system become the following differential-algebraic equations:
Note that the claim that DAEs present a “form” of infinite stiffness [23] can be easily inferred from the previous analysis.
Rights and permissions
Springer Nature or its licensor holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Moya, C., Lin, G. DAE-PINN: a physics-informed neural network model for simulating differential algebraic equations with application to power networks. Neural Comput & Applic 35, 3789–3804 (2023). https://doi.org/10.1007/s00521-022-07886-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00521-022-07886-y