J.B. Stiasny
info
Please Note
<p>This page displays the records of the person named above and is not linked to a unique person identifier. This record may need to be merged to a profile.</p>
15 records found
1
Configure-and-Bound
A fast heuristic for network topology reconfiguration
Congestion management is a key challenge in power systems, and topology reconfiguration offers a promising solution. This paper introduces the Configure-and-Bound (C&B) algorithm to efficiently solve network topology reconfiguration (NTR) problems, focusing on substation switching and busbar splitting. By exploiting the locality effects of switching maneuvers, the C&B algorithm significantly reduces the computational time required to solve the NP-hard NTR problems, while achieving most of the cost savings achieved by exact methods. We explore the conditions under which the proposed C&B algorithm is most effective by classifying congestion into two broad classes; near congestion and far congestion. The locality condition and the foundation of the proposed algorithm generalize to a broader class of (power system) optimization problems. Case studies done on IEEE 39, 118, 240, 300, 500, 588, and 793 bus systems demonstrate that the proposed algorithm can reduce the computational runtime by up to 99% and achieve up to 99.9% similar costs relative to the global optimal solution.
...
Congestion management is a key challenge in power systems, and topology reconfiguration offers a promising solution. This paper introduces the Configure-and-Bound (C&B) algorithm to efficiently solve network topology reconfiguration (NTR) problems, focusing on substation switching and busbar splitting. By exploiting the locality effects of switching maneuvers, the C&B algorithm significantly reduces the computational time required to solve the NP-hard NTR problems, while achieving most of the cost savings achieved by exact methods. We explore the conditions under which the proposed C&B algorithm is most effective by classifying congestion into two broad classes; near congestion and far congestion. The locality condition and the foundation of the proposed algorithm generalize to a broader class of (power system) optimization problems. Case studies done on IEEE 39, 118, 240, 300, 500, 588, and 793 bus systems demonstrate that the proposed algorithm can reduce the computational runtime by up to 99% and achieve up to 99.9% similar costs relative to the global optimal solution.
Time-domain simulations are crucial for ensuring power system stability and avoiding critical scenarios that could lead to blackouts. The next-generation power systems require a significant increase in the computational cost and complexity of these simulations due to additional degrees of uncertainty, non-linearity and states. Physics-Informed Neural Networks (PINN) have been shown to accelerate single-component simulations by several orders of magnitude. However, their application to current time-domain simulation solvers has been particularly challenging since the system’s dynamics depend on multiple components. Using a new training formulation, this paper introduces the first natural step to integrate PINNs into multi-component time-domain simulations. We propose PINNs as an alternative to other classical numerical methods for individual components. Once trained, these neural networks approximate component dynamics more accurately for longer time steps. Formulated as an implicit and consistent method with the transient simulation workflow, PINNs speed up simulation time by significantly increasing the time steps used. For explanation clarity, we demonstrate the training, integration, and simulation framework for several combinations of PINNs and numerical solution methods using the IEEE 9-bus system, although the method applies equally well to any power system size.
...
Time-domain simulations are crucial for ensuring power system stability and avoiding critical scenarios that could lead to blackouts. The next-generation power systems require a significant increase in the computational cost and complexity of these simulations due to additional degrees of uncertainty, non-linearity and states. Physics-Informed Neural Networks (PINN) have been shown to accelerate single-component simulations by several orders of magnitude. However, their application to current time-domain simulation solvers has been particularly challenging since the system’s dynamics depend on multiple components. Using a new training formulation, this paper introduces the first natural step to integrate PINNs into multi-component time-domain simulations. We propose PINNs as an alternative to other classical numerical methods for individual components. Once trained, these neural networks approximate component dynamics more accurately for longer time steps. Formulated as an implicit and consistent method with the transient simulation workflow, PINNs speed up simulation time by significantly increasing the time steps used. For explanation clarity, we demonstrate the training, integration, and simulation framework for several combinations of PINNs and numerical solution methods using the IEEE 9-bus system, although the method applies equally well to any power system size.
The cost of grid tariffs is expected to rise and account for an ever-increasing share of electricity consumers’ invoices. Hence, it is imperative to factor these costs in when modelling electricity demand behaviour in a market-driven environment. Accurate demand profiles are essential for optimising transmission expansion planning (TEP) as an accurate representation of electricity demand profiles aids in finding the most beneficial expansion plan. Common TEP formulations in the literature do not include the costs accumulated by grid tariffs. This paper proposes a revised problem representation incorporating a version of grid tariffs in the objective function of TEP optimisation. An analysis is carried out to estimate the sensitivity of the planning strategy to different ratios of grid tariffs to generation costs. It could be concluded that as the grid tariffs are of the same magnitude as the generation costs, the optimal planning strategy foresees to retrofit lines facilitating local generation over grid-sourced electricity. If the grid tariff is further increased, there will be no significant deviation in the expansion plan from the optimal expansion plan generated at grid tariff to generation cost parity.
...
The cost of grid tariffs is expected to rise and account for an ever-increasing share of electricity consumers’ invoices. Hence, it is imperative to factor these costs in when modelling electricity demand behaviour in a market-driven environment. Accurate demand profiles are essential for optimising transmission expansion planning (TEP) as an accurate representation of electricity demand profiles aids in finding the most beneficial expansion plan. Common TEP formulations in the literature do not include the costs accumulated by grid tariffs. This paper proposes a revised problem representation incorporating a version of grid tariffs in the objective function of TEP optimisation. An analysis is carried out to estimate the sensitivity of the planning strategy to different ratios of grid tariffs to generation costs. It could be concluded that as the grid tariffs are of the same magnitude as the generation costs, the optimal planning strategy foresees to retrofit lines facilitating local generation over grid-sourced electricity. If the grid tariff is further increased, there will be no significant deviation in the expansion plan from the optimal expansion plan generated at grid tariff to generation cost parity.
Time domain simulation (TDS) is an important tool for assessing power system security under various disturbances. However, its computational cost limits the number of disturbances that can be assessed. The need for fast assessment of numerous disturbances has increased with the rapid integration of renewable energy sources. Machine learning (ML) methods have been explored to accelerate power system TDS, but these methods are studied in interpolation scenarios, where they predict outputs for inputs within the training data distribution. This work uses a state-of-the-art ML model to explore the extrapolation behavior of ML models for TDS. First, we highlight the importance of ML models’ extrapolation capacity for fast assessment of numerous diverse disturbances. Next, we demonstrate that extrapolation for discrete disturbances is more challenging than for continuous disturbances. Subsequently, we investigate how transfer learning (TL) may be used to improve the performance of ML models in TDS extrapolation scenarios. Finally, we outline the limitations of TL for power system TDS and suggest alternative approaches for developing ML models with better extrapolation performance in TDS applications.
...
Time domain simulation (TDS) is an important tool for assessing power system security under various disturbances. However, its computational cost limits the number of disturbances that can be assessed. The need for fast assessment of numerous disturbances has increased with the rapid integration of renewable energy sources. Machine learning (ML) methods have been explored to accelerate power system TDS, but these methods are studied in interpolation scenarios, where they predict outputs for inputs within the training data distribution. This work uses a state-of-the-art ML model to explore the extrapolation behavior of ML models for TDS. First, we highlight the importance of ML models’ extrapolation capacity for fast assessment of numerous diverse disturbances. Next, we demonstrate that extrapolation for discrete disturbances is more challenging than for continuous disturbances. Subsequently, we investigate how transfer learning (TL) may be used to improve the performance of ML models in TDS extrapolation scenarios. Finally, we outline the limitations of TL for power system TDS and suggest alternative approaches for developing ML models with better extrapolation performance in TDS applications.
Conference paper
(2025)
-
Viktor Zobernig, Sarah Fanta, Stefan Stromer, Regina Hemm, Jochen Stiasny, Jochen L. Cremer, Laurens J. De Vries
The growing share of renewable energy in shortterm European electricity markets has significantly increased congestion management costs and demands. Therefore, current market design is not optional to keep congestion costs low. A proper market would incentivize the integration of flexibilities to boost competition and lower costs, while mitigating risks of manipulation. However, assessing behavioral impacts is challenging due to increasingly interconnected market structures. Studies modeling more than two markets often overlook the strategic opportunities that emerge from these interactions, focusing instead on large-scale dynamics. To capture the detailed impact of bidding strategies, we use reinforcement learning to explore multi-market strategies. By progressively training a Deep Reinforcement Learning (DRL) agent as a market participant - from replicating established behaviors to mastering intricate multimarket interactions - we employ Domain-Informed Curriculum Learning (DomCL), a structured approach that systematically guides learning through staged complexity. We validate our approach against established two-market studies, then evaluate it in two progressively complex four-market case studies spanning a 6-bus network, including historical data. Results show that our DRL-based method improves performance while uncovering challenges that arise as strategic opportunities expand, offering a structured approach for multi-market design analysis.
...
The growing share of renewable energy in shortterm European electricity markets has significantly increased congestion management costs and demands. Therefore, current market design is not optional to keep congestion costs low. A proper market would incentivize the integration of flexibilities to boost competition and lower costs, while mitigating risks of manipulation. However, assessing behavioral impacts is challenging due to increasingly interconnected market structures. Studies modeling more than two markets often overlook the strategic opportunities that emerge from these interactions, focusing instead on large-scale dynamics. To capture the detailed impact of bidding strategies, we use reinforcement learning to explore multi-market strategies. By progressively training a Deep Reinforcement Learning (DRL) agent as a market participant - from replicating established behaviors to mastering intricate multimarket interactions - we employ Domain-Informed Curriculum Learning (DomCL), a structured approach that systematically guides learning through staged complexity. We validate our approach against established two-market studies, then evaluate it in two progressively complex four-market case studies spanning a 6-bus network, including historical data. Results show that our DRL-based method improves performance while uncovering challenges that arise as strategic opportunities expand, offering a structured approach for multi-market design analysis.
Journal article
(2024)
-
Jochen Stiasny, Baosen Zhang, Spyros Chatzivasileiadis
The dynamic behaviour of a power system can be described by a system of differential–algebraic equations. Time-domain simulations are used to simulate the evolution of these dynamics. They often require the use of small time step sizes and therefore become computationally expensive. To accelerate these simulations, we propose a simulator – PINNSim – that allows to take significantly larger time steps. It is based on Physics-Informed Neural Networks (PINNs) for the solution of the dynamics of single components in the power system. To resolve their interaction we employ a scalable root-finding algorithm. We demonstrate PINNSim on a 9-bus system and show the increased time step size compared to a trapezoidal integration rule. We discuss key characteristics of PINNSim and important steps for developing PINNSim into a fully fledged simulator. As such, it could offer the opportunity for significantly increasing time step sizes and thereby accelerating time-domain simulations.
...
The dynamic behaviour of a power system can be described by a system of differential–algebraic equations. Time-domain simulations are used to simulate the evolution of these dynamics. They often require the use of small time step sizes and therefore become computationally expensive. To accelerate these simulations, we propose a simulator – PINNSim – that allows to take significantly larger time steps. It is based on Physics-Informed Neural Networks (PINNs) for the solution of the dynamics of single components in the power system. To resolve their interaction we employ a scalable root-finding algorithm. We demonstrate PINNSim on a 9-bus system and show the increased time step size compared to a trapezoidal integration rule. We discuss key characteristics of PINNSim and important steps for developing PINNSim into a fully fledged simulator. As such, it could offer the opportunity for significantly increasing time step sizes and thereby accelerating time-domain simulations.
Journal article
(2023)
-
Jochen Stiasny, Spyros Chatzivasileiadis
The simulation of power system dynamics poses a computationally expensive task. Considering the growing uncertainty of generation and demand patterns, thousands of scenarios need to be continuously assessed to ensure the safety of power systems. Physics-Informed Neural Networks (PINNs) have recently emerged as a promising solution for drastically accelerating computations of non-linear dynamical systems. This work investigates the applicability of these methods for power system dynamics, focusing on the dynamic response to load disturbances. Comparing the prediction of PINNs to the solution of conventional solvers, we find that PINNs can be 10 to 1’000 times faster than conventional solvers. At the same time, we find them to be sufficiently accurate and numerically stable even for large time steps. To facilitate a deeper understanding, this paper also present a new regularisation of Neural Network (NN) training by introducing a gradient-based term in the loss function. The resulting NNs, which we call dtNNs, help us deliver a comprehensive analysis about the strengths and weaknesses of the NN based approaches, how incorporating knowledge of the underlying physics affects NN performance, and how this compares with conventional solvers for power system dynamics.
...
The simulation of power system dynamics poses a computationally expensive task. Considering the growing uncertainty of generation and demand patterns, thousands of scenarios need to be continuously assessed to ensure the safety of power systems. Physics-Informed Neural Networks (PINNs) have recently emerged as a promising solution for drastically accelerating computations of non-linear dynamical systems. This work investigates the applicability of these methods for power system dynamics, focusing on the dynamic response to load disturbances. Comparing the prediction of PINNs to the solution of conventional solvers, we find that PINNs can be 10 to 1’000 times faster than conventional solvers. At the same time, we find them to be sufficiently accurate and numerically stable even for large time steps. To facilitate a deeper understanding, this paper also present a new regularisation of Neural Network (NN) training by introducing a gradient-based term in the loss function. The resulting NNs, which we call dtNNs, help us deliver a comprehensive analysis about the strengths and weaknesses of the NN based approaches, how incorporating knowledge of the underlying physics affects NN performance, and how this compares with conventional solvers for power system dynamics.
Conference paper
(2023)
-
Simon Stock, Jochen Stiasny, Davood Babazadeh, Christian Becker, Spyros Chatzivasileiadis
This paper introduces for the first time, to the best of our knowledge, the Bayesian Physics-Informed Neural Networks for applications in power systems. Bayesian Physics-Informed Neural Networks (BPINNs) combine the advantages of Physics-Informed Neural Networks (PINNs), being robust to noise and missing data, with Bayesian modeling, delivering a confidence measure for their output. Such a confidence measure can be very valuable for the operation of safety critical systems, such as power systems, as it offers a degree of 'trustworthiness' for the neural network output. This paper applies the BPINNs for robust identification of the system inertia and damping, using a single machine infinite bus system as the guiding example. The goal of this paper is to introduce the concept and explore the strengths and weaknesses of BPINNs compared to existing methods. We compare BPINNs with the PINNs and the recently popular method for system identification, SINDy. We find that BPINNs and PINNs are robust against all noise levels, delivering estimates of the system inertia and damping with significantly lower error compared to SINDy, especially as the noise levels increases.
...
This paper introduces for the first time, to the best of our knowledge, the Bayesian Physics-Informed Neural Networks for applications in power systems. Bayesian Physics-Informed Neural Networks (BPINNs) combine the advantages of Physics-Informed Neural Networks (PINNs), being robust to noise and missing data, with Bayesian modeling, delivering a confidence measure for their output. Such a confidence measure can be very valuable for the operation of safety critical systems, such as power systems, as it offers a degree of 'trustworthiness' for the neural network output. This paper applies the BPINNs for robust identification of the system inertia and damping, using a single machine infinite bus system as the guiding example. The goal of this paper is to introduce the concept and explore the strengths and weaknesses of BPINNs compared to existing methods. We compare BPINNs with the PINNs and the recently popular method for system identification, SINDy. We find that BPINNs and PINNs are robust against all noise levels, delivering estimates of the system inertia and damping with significantly lower error compared to SINDy, especially as the noise levels increases.
Conference paper
(2022)
-
J.B. Stiasny, Samuel Chevalier, Rahul Nellikkath, Brynjar Sævarsson, Spyros Chatzivasileiadis
Conference paper
(2022)
-
Samuel Chevalier, Jochen Stiasny, Spyros Chatzivasileiadis
Recent advances in deep learning have allowed neural networks (NNs) to successfully replace traditional numerical solvers in many applications, thus enabling impressive computing gains. One such application is time domain simulation, which is indispensable for the design, analysis and operation of many engineering systems. Simulating dynamical systems with implicit Newton-based solvers is a computationally heavy task, as it requires the solution of a parameterized system of differential and algebraic equations at each time step. A variety of NN-based methodologies have been shown to successfully approximate the trajectories computed by numerical solvers at a fraction of the time. However, few previous works have used NNs to model the numerical solver itself. For the express purpose of accelerating time domain simulation speeds, this paper proposes and explores two complementary alternatives for modeling numerical solvers. First, we use a NN to mimic the linear transformation provided by the inverse Jacobian in a single Newton step. Using this procedure, we evaluate and project the exact, physics-based residual error onto the NN mapping, thus leaving physics “in the loop”. The resulting tool, termed the Physics-pRojected Neural-Newton Solver (PRoNNS), is able to achieve an extremely high degree of numerical accuracy at speeds which were observed to be up to 31% faster than a Newton-based solver. In the second approach, we model the Newton solver at the heart of an implicit Runge-Kutta integrator as a contracting map iteratively seeking a fixed point on a time domain trajectory. The associated recurrent NN simulation tool, termed the Contracting Neural-Newton Solver (CoNNS), is embedded with training constraints (via CVXPY Layers) which guarantee the mapping provided by the NN satisfies the Banach fixed-point theorem; successive passes through the NN are therefore guaranteed to converge to a unique, fixed point. Explicitly capturing the contracting nature of Newton iterations leads to significantly increased NN accuracy relative to a vanilla NN. We test and evaluate the merits of both PRoNNS and CoNNS on three dynamical test systems.
...
Recent advances in deep learning have allowed neural networks (NNs) to successfully replace traditional numerical solvers in many applications, thus enabling impressive computing gains. One such application is time domain simulation, which is indispensable for the design, analysis and operation of many engineering systems. Simulating dynamical systems with implicit Newton-based solvers is a computationally heavy task, as it requires the solution of a parameterized system of differential and algebraic equations at each time step. A variety of NN-based methodologies have been shown to successfully approximate the trajectories computed by numerical solvers at a fraction of the time. However, few previous works have used NNs to model the numerical solver itself. For the express purpose of accelerating time domain simulation speeds, this paper proposes and explores two complementary alternatives for modeling numerical solvers. First, we use a NN to mimic the linear transformation provided by the inverse Jacobian in a single Newton step. Using this procedure, we evaluate and project the exact, physics-based residual error onto the NN mapping, thus leaving physics “in the loop”. The resulting tool, termed the Physics-pRojected Neural-Newton Solver (PRoNNS), is able to achieve an extremely high degree of numerical accuracy at speeds which were observed to be up to 31% faster than a Newton-based solver. In the second approach, we model the Newton solver at the heart of an implicit Runge-Kutta integrator as a contracting map iteratively seeking a fixed point on a time domain trajectory. The associated recurrent NN simulation tool, termed the Contracting Neural-Newton Solver (CoNNS), is embedded with training constraints (via CVXPY Layers) which guarantee the mapping provided by the NN satisfies the Banach fixed-point theorem; successive passes through the NN are therefore guaranteed to converge to a unique, fixed point. Explicitly capturing the contracting nature of Newton iterations leads to significantly increased NN accuracy relative to a vanilla NN. We test and evaluate the merits of both PRoNNS and CoNNS on three dynamical test systems.
Journal article
(2022)
-
Spyros Chatzivasileiadis, Andreas Venzke, Jochen Stiasny, George S. Misyris
We experience the power of machine learning (ML) in our everyday lives - be it picture and speech recognition, customized suggestions by virtual assistants, or just unlocking our phones. Its underlying mathematical principles have been applied since the middle of the last century in what is known as statistical learning. However, the enormous increase in computational power, even in devices as small as a smartphone, has enabled significant advances and wide adoption of ML in nearly every part of our lives and the scientific world.
...
We experience the power of machine learning (ML) in our everyday lives - be it picture and speech recognition, customized suggestions by virtual assistants, or just unlocking our phones. Its underlying mathematical principles have been applied since the middle of the last century in what is known as statistical learning. However, the enormous increase in computational power, even in devices as small as a smartphone, has enabled significant advances and wide adoption of ML in nearly every part of our lives and the scientific world.
Journal article
(2021)
-
Jochen Stiasny, Thierry Zufferey, Giacomo Pareschi, Damiano Toffanin, Gabriela Hug, Konstantinos Boulouchos
The presented work identifies the dominating influencing factors in electric vehicle (EV) modelling on low-voltage distribution grids to establish guidance for reliable impact assessments of increasing EV penetration. Seven aspects are distinguished with respect to the modelling of the load of EVs that influence the flows and voltages in the grid. For each of these aspects sensitivity analyses are carried out by running power flow simulations in a Monte-Carlo fashion to account for the stochasticity in the model parameters. The impacts are analysed using a variety of metrics including transformer and line loadings. The highest sensitivities are observed for the number of vehicles in the grid, the used charger power rating and the modelling of driving patterns. The grid configuration as well as locally higher EV shares gain significance for line loading assessments. Car modelling and people's charging behaviour play minor roles.
...
The presented work identifies the dominating influencing factors in electric vehicle (EV) modelling on low-voltage distribution grids to establish guidance for reliable impact assessments of increasing EV penetration. Seven aspects are distinguished with respect to the modelling of the load of EVs that influence the flows and voltages in the grid. For each of these aspects sensitivity analyses are carried out by running power flow simulations in a Monte-Carlo fashion to account for the stochasticity in the model parameters. The impacts are analysed using a variety of metrics including transformer and line loadings. The highest sensitivities are observed for the number of vehicles in the grid, the used charger power rating and the modelling of driving patterns. The grid configuration as well as locally higher EV shares gain significance for line loading assessments. Car modelling and people's charging behaviour play minor roles.
Conference paper
(2021)
-
Jochen Stiasny, George S. Misyris, Spyros Chatzivasileiadis
Varying power-infeed from converter-based generation units introduces great uncertainty on system parameters such as inertia and damping. As a consequence, system opera-tors face increasing challenges in performing dynamic security assessment and taking real-time control actions. Exploiting the widespread deployment of phasor measurement units (PMUs) and aiming at developing a fast dynamic state and parameter estimation tool, this paper investigates the performance of Physics-Informed Neural Networks (PINN) for discovering the frequency dynamics of future power systems. PINNs have the potential to address challenges such as the stronger non-linearities of low-inertia systems, increased measurement noise, and limited availability of data. The estimator is demonstrated in several test cases using a 4-bus system, and compared with state of the art algorithms, such as the Unscented Kalman Filter (UKF), to assess its performance.
...
Varying power-infeed from converter-based generation units introduces great uncertainty on system parameters such as inertia and damping. As a consequence, system opera-tors face increasing challenges in performing dynamic security assessment and taking real-time control actions. Exploiting the widespread deployment of phasor measurement units (PMUs) and aiming at developing a fast dynamic state and parameter estimation tool, this paper investigates the performance of Physics-Informed Neural Networks (PINN) for discovering the frequency dynamics of future power systems. PINNs have the potential to address challenges such as the stronger non-linearities of low-inertia systems, increased measurement noise, and limited availability of data. The estimator is demonstrated in several test cases using a 4-bus system, and compared with state of the art algorithms, such as the Unscented Kalman Filter (UKF), to assess its performance.
Conference paper
(2021)
-
Jochen Stiasny, Samuel Chevalier, Spyros Chatzivasileiadis
In order to drastically reduce the heavy computational burden associated with time-domain simulations, this paper introduces a Physics-Informed Neural Network (PINN) to directly learn the solutions of power system dynamics. In contrast to the limitations of classical model order reduction approaches, commonly used to accelerate time-domain simulations, PINNs can universally approximate any continuous function with an arbitrary degree of accuracy. One of the novelties of this paper is that we avoid the need for any training data. We achieve this by incorporating the governing differential equations and an implicit Runge-Kutta (RK) integration scheme directly into the training process of the PINN; through this approach, PINNs can predict the trajectory of a dynamical power system at any discrete time step. The resulting Runge-Kutta-based physics-informed neural networks (RK-PINNs) can yield up to 100 times faster evaluations of the dynamics compared to standard time-domain simulations. We demonstrate the methodology on a single-machine infinite bus system governed by the swing equation. We show that RK-PINNs can accurately and quickly predict the solution trajectories.
...
In order to drastically reduce the heavy computational burden associated with time-domain simulations, this paper introduces a Physics-Informed Neural Network (PINN) to directly learn the solutions of power system dynamics. In contrast to the limitations of classical model order reduction approaches, commonly used to accelerate time-domain simulations, PINNs can universally approximate any continuous function with an arbitrary degree of accuracy. One of the novelties of this paper is that we avoid the need for any training data. We achieve this by incorporating the governing differential equations and an implicit Runge-Kutta (RK) integration scheme directly into the training process of the PINN; through this approach, PINNs can predict the trajectory of a dynamical power system at any discrete time step. The resulting Runge-Kutta-based physics-informed neural networks (RK-PINNs) can yield up to 100 times faster evaluations of the dynamics compared to standard time-domain simulations. We demonstrate the methodology on a single-machine infinite bus system governed by the swing equation. We show that RK-PINNs can accurately and quickly predict the solution trajectories.
Conference paper
(2021)
-
George S. Misyris, Jochen Stiasny, Spyros Chatzivasileiadis
This paper proposes a tractable framework to determine key characteristics of
non-linear dynamic systems by converting physics-informed neural networks to a
mixed integer linear program. Our focus is on power system applications.
Traditional methods in power systems require the use of a large number of
simulations and other heuristics to determine parameters such as the critical
clearing time, i.e. the maximum allowable time within which a disturbance must
be cleared before the system moves to instability. The work proposed in this
paper uses physics-informed neural networks to capture the power system dynamic
behavior and, through an exact transformation, converts them to a tractable
optimization problem which can be used to determine critical system indices. By
converting neural networks to mixed integer linear programs, our framework also
allows to adjust the conservativeness of the neural network output with respect
to the existing stability boundaries. We demonstrate the performance of our
methods on the non-linear dynamics of converter-based generation in response to
voltage disturbances.
...
This paper proposes a tractable framework to determine key characteristics of
non-linear dynamic systems by converting physics-informed neural networks to a
mixed integer linear program. Our focus is on power system applications.
Traditional methods in power systems require the use of a large number of
simulations and other heuristics to determine parameters such as the critical
clearing time, i.e. the maximum allowable time within which a disturbance must
be cleared before the system moves to instability. The work proposed in this
paper uses physics-informed neural networks to capture the power system dynamic
behavior and, through an exact transformation, converts them to a tractable
optimization problem which can be used to determine critical system indices. By
converting neural networks to mixed integer linear programs, our framework also
allows to adjust the conservativeness of the neural network output with respect
to the existing stability boundaries. We demonstrate the performance of our
methods on the non-linear dynamics of converter-based generation in response to
voltage disturbances.