Constructing Neural Network-Based Models for Simulating Dynamical Systems

Legaard, Christian Møldrup; Schranz, Thomas; Schweiger, Gerald; Drgoňa, Ján; Falay, Basak; Gomes, Cláudio; Iosifidis, Alexandros; Abkar, Mahdi; Larsen, Peter Gorm

doi:10.48550/arxiv.2111.01495

Cited by 8 publications

(19 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Specifically, using measurements corrupted by noise as the initial observation, we verify the accuracy of the predictive models by investigating how well they can "recover" the actual (ground truth) frequencies. Figure 4 illustrates the actual and predicted frequencies at 3 arbitrarily chosen generator buses under two different disturbance scenarios: 1) (top) disturbance -1 correspond to the load changes at buses 20, 23, 25, 36, 41, 42; and 2) (bottom) disturbance -2 correspond to the load changes at buses 4,8,12,15,18,27,44,46,47,48,51. Notice that during the transient period the model predictions demonstrate some undershoot -e.g.Robust DMD, deepDMD for disturbance-2 from 4s and STGCN for disturbance-1 throughout the prediction window -or overshoot behaviore.g.deepDMD predictions under disturbance-1 and 2 around 2s.…”

Section: Performance Evaluationmentioning

confidence: 99%

“…Figure 4: Actual and predicted frequencies corresponding to two randomly chosen disturbances using Robust DMD, deepDMD and STGCN at generator buses 57, 64, 66 for a period of 10 seconds (500 time-steps as per the given PMU sampling frequency.) The top plots correspond to disturbance -1 (load changes at buses 20,23,25,36,41,42) and the bottom plots correspond to disturbance -2 (load changes at buses 4,8,12,15,18,27,44,46,47,48,51). The solid lines show the actual noisy measurements whereas the dotted lines indicate the predictions.…”

Section: Performance Evaluationmentioning

confidence: 99%

“…State-of-the-art. Data-driven learning of predictive models for nonlinear dynamical systems has received some attention in recent times [4][5][6]. The work [4] presents a review of the different neural network-based learning models for nonlinear dynamical systems, including different variants of recurrent networks, e.g., long-short-short-term memory (LSTM) models [5,6].…”

Section: Introductionmentioning

confidence: 99%

“…Data-driven learning of predictive models for nonlinear dynamical systems has received some attention in recent times [4][5][6]. The work [4] presents a review of the different neural network-based learning models for nonlinear dynamical systems, including different variants of recurrent networks, e.g., long-short-short-term memory (LSTM) models [5,6]. In the context of power systems dynamics, transient stability assessment and prediction has been an active area of research over the years, as summarized in a recent review [1].…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Graph Neural Network and Koopman Models for Learning Networked Dynamics: A Comparative Study on Power Grid Transients Prediction

Nandanoori¹,

Guan²,

Kundu³

et al. 2022

Preprint

View full text Add to dashboard Cite

Continuous monitoring of the spatio-temporal dynamic behavior of critical infrastructure networks, such as the power systems, is a challenging but important task. In particular, accurate and timely prediction of the (electro-mechanical) transient dynamic trajectories of the power grid is necessary for early detection of any instability and prevention of catastrophic failures. Existing approaches for prediction of dynamic trajectories either rely on the availability of accurate physical models of the system, use computationally expensive time-domain simulations, or are applicable only at local prediction problems (e.g., a single generator). In this paper, we report the application of two broad classes of data-driven learning models -along with their algorithmic implementation and performance evaluation -in predicting transient trajectories in power networks using only streaming measurements and the network topology as input. One class of models is based on the Koopman operator theory which allows for capturing the nonlinear dynamic behavior via an infinite-dimensional linear operator. The other class of models is based on the graph convolutional neural networks which are adept at capturing the inherent spatio-temporal correlations within the power network. Transient dynamic datasets for training and testing the models are synthesized by simulating a wide variety of load change events in the IEEE 68-bus system, categorized by the load change magnitudes, as well as by the degree of connectivity and the distance to nearest generator nodes. The results confirm that the proposed predictive models can successfully predict the post-disturbance transient evolution of the system with a high level of accuracy.

show abstract

Section: Performance Evaluationmentioning

confidence: 99%

Section: Performance Evaluationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Graph Neural Network and Koopman Models for Learning Networked Dynamics: A Comparative Study on Power Grid Transients Prediction

Nandanoori¹,

Guan²,

Kundu³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…a) Electronic mail: rghiassi@ut.ac.ir b) Electronic mail: abkar@mpe.au.dk Nowadays, the data-driven models along with machine learning (ML) techniques are being extensively utilized in fluid-mechanics-related problems 15,16 . Although the interpretability and generalizability of data-driven models are among the most challenging issues, these models can reduce the need for computationally expensive physics-based models 17,18 . Several studies are using ML techniques, specifically deep learning tools, for predicting fluid-mechanicsrelated problems by utilizing mid-and high-fidelity data from RANS and LES (or direct numerical) simulations, e.g., flow over periodic hill [19][20][21] , wall mounted cube 22 , backward-facing step 23 , duct flow [24][25][26][27] , RANS correction 28,29 , flow with system rotation 30 , and wind-farm modeling [31][32][33] , among others.…”

Section: Introductionmentioning

confidence: 99%

Data-driven quantification of model-form uncertainty in Reynolds-averaged simulations of wind farms

Eidi,

Zehtabiyan-Rezaie,

Ghiassi

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Computational fluid dynamics using the Reynolds-averaged Navier-Stokes (RANS) remains the most costeffective approach to study wake flows and power losses in wind farms. The underlying assumptions associated with turbulence closures is one of the biggest sources of errors and uncertainties in the model predictions. This work aims to quantify model-form uncertainties in RANS simulations of wind farms by perturbing the Reynolds stress tensor through a data-driven machine-learning technique. To this end, the extreme gradient boosting algorithm is validated and employed to predict the perturbation amount and direction of the modeled Reynolds stress toward the limiting states of turbulence on the Barycentric map such that the perturbed Reynolds stress returns a more accurate representation of the Reynolds stress. The model is trained on high-fidelity data obtained from large-eddy simulation of a specific wind farm, and it is tested on two other (unseen) wind farms with distinct layouts. The results indicate that, unlike the data-free approach in which a uniform and constant perturbation amount is applied to the entire computational domain, the proposed framework yields an optimal estimation of the uncertainty bounds for the RANS-predicted quantities of interest, including the wake velocity, turbulence intensity, and power losses in wind farms.

show abstract

Friction modelling and the use of a physics-informed neural network for estimating frictional torque characteristics

Olejnik,

Ayankoso

2023

Meccanica

View full text Add to dashboard Cite

This paper presents an exploration of friction modeling encompassing theoretical and practical aspects, utilizing a planar or 2D contact system. Various white-box friction models, including static and dynamic variants, are introduced, highlighting the superior capability of dynamic models in comprehensively capturing friction effects, substantiated through numerical simulation. Practical aspects of friction measurement and data-driven friction modeling are elucidated. The discourse extends to the development of grey-box and black-box friction models. A significant contribution lies in the proposition of a physics-informed neural network-based friction modeling approach, presenting it as an advanced and preferable alternative for friction estimation. To exemplify its efficacy, a case study of a torsion-based frictional contact scenario, employing Physics-Informed Neural Network (PINN) and the Nelder–Mead (N–M) algorithm for concurrent dynamics and friction model identification, was examined. Experimental data from a double torsion pendulum system, characterized by discontinuous dynamics, is employed for training. Results demonstrate the PINN’s superiority, providing more accurate representation of stick–slip phases at the contact zone and exhibiting faster performance compared to the N–M algorithm. The paper concludes by deliberating on challenges, prospects, and future directions in friction modeling.

show abstract

Constructing Neural Network-Based Models for Simulating Dynamical Systems

Cited by 8 publications

References 0 publications

Graph Neural Network and Koopman Models for Learning Networked Dynamics: A Comparative Study on Power Grid Transients Prediction

Graph Neural Network and Koopman Models for Learning Networked Dynamics: A Comparative Study on Power Grid Transients Prediction

Data-driven quantification of model-form uncertainty in Reynolds-averaged simulations of wind farms

Friction modelling and the use of a physics-informed neural network for estimating frictional torque characteristics

Contact Info

Product

Resources

About